SOLID STATE PHYSICS PART II Optical Properties of Solids

M. S. Dresselhaus

Contents
1 Review of Fundamental Relations for Optical Phenomena 1.1 Introductory Remarks on Optical Probes . . . . . . . . . . . . . . . . 1.2 The Complex dielectric function and the complex optical conductivity 1.3 Relation of Complex Dielectric Function to Observables . . . . . . . . 1.4 Units for Frequency Measurements . . . . . . . . . . . . . . . . . . . . 2 Drude TheoryFree Carrier Contribution 2.1 The Free Carrier Contribution . . . . . . 2.2 Low Frequency Response: 1 . . . . 2.3 High Frequency Response; 1 . . . . 2.4 The Plasma Frequency . . . . . . . . . . . 1 1 2 4 7 8 8 10 11 11 15 15 19 19 19 20 21 23 27 27 30 36 36 37 38 41 46 46 47 47 50

. . . .

. . . .

. . . .

to the Optical Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . .

. . . .

3 Interband Transitions 3.1 The Interband Transition Process . . . . . . . . . . . . . . . . . . . . . 3.1.1 Insulators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1.2 Semiconductors . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1.3 Metals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 Form of the Hamiltonian in an Electromagnetic Field . . . . . . . . . . 3.3 Relation between Momentum Matrix Elements and the Eective Mass 3.4 Spin-Orbit Interaction in Solids . . . . . . . . . . . . . . . . . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

4 The Joint Density of States and Critical Points 4.1 The Joint Density of States . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2 Critical Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 Absorption of Light in Solids 5.1 The Absorption Coecient . . . . . . . . . . . . . . . . . . . . . 5.2 Free Carrier Absorption in Semiconductors . . . . . . . . . . . . 5.3 Free Carrier Absorption in Metals . . . . . . . . . . . . . . . . . 5.4 Direct Interband Transitions . . . . . . . . . . . . . . . . . . . . 5.4.1 Temperature Dependence of Eg . . . . . . . . . . . . . . . 5.4.2 Dependence of Absorption Edge on Fermi Energy . . . . . 5.4.3 Dependence of Absorption Edge on Applied Electric Field 5.5 Conservation of Crystal Momentum in Direct Optical Transitions 5.6 Indirect Interband Transitions . . . . . . . . . . . . . . . . . . . . 2

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

6 Optical Properties of Solids Over a Wide Frequency 6.1 KramersKronig Relations . . . . . . . . . . . . . . . . 6.2 Optical Properties and Band Structure . . . . . . . . . 6.3 Modulated Reectivity Experiments . . . . . . . . . . 6.4 Ellipsometry and Measurement of Optical Constants . 7 Impurities and Excitons 7.1 Impurity Level Spectroscopy . . . . . . . . . . . . 7.2 Shallow Impurity Levels . . . . . . . . . . . . . . 7.3 Departures from the Hydrogenic Model . . . . . 7.4 Vacancies, Color Centers and Interstitials . . . . 7.5 Spectroscopy of Excitons . . . . . . . . . . . . . . 7.6 Classication of Excitons . . . . . . . . . . . . . 7.7 Optical Transitions in Quantum Well Structures

Range . . . . . . . . . . . . . . . . . . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

57 57 62 64 71 73 73 73 77 79 83 87 91 97 97 98 104 108 108 108 110 112 121 126 128 132 132 134 135 136 137 137 137 141 144 144 146 146 147 152 153 154

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

8 Luminescence and Photoconductivity 8.1 Classication of Luminescence Processes . . . . . . . . . . . . . . . . . . . . 8.2 Emission and Absorption . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.3 Photoconductivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 Optical Study of Lattice Vibrations 10.1 Lattice Vibrations in Semiconductors . . . . . . . . . 10.1.1 General Considerations . . . . . . . . . . . . 10.2 Dielectric Constant and Polarizability . . . . . . . . 10.3 Polariton Dispersion Relations . . . . . . . . . . . . 10.4 Light Scattering . . . . . . . . . . . . . . . . . . . . 10.5 Feynman Diagrams for Light Scattering . . . . . . . 10.6 Raman Spectra in Quantum Wells and Superlattices 11 Non-Linear Optics 11.1 Introductory Comments . . . 11.2 Second Harmonic Generation 11.2.1 Parametric Oscillation 11.2.2 Frequency Conversion

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

12 Electron Spectroscopy and Surface Science 12.1 Photoemission Electron Spectroscopy . . . . . . . . 12.1.1 Introduction . . . . . . . . . . . . . . . . . 12.1.2 Energy Distribution Curves . . . . . . . . . 12.1.3 Angle Resolved Photoelectron Spectroscopy 12.1.4 Synchrotron Radiation Sources . . . . . . . 12.2 Surface Science . . . . . . . . . . . . . . . . . . . . 12.2.1 Introduction . . . . . . . . . . . . . . . . . 12.2.2 Electron Diraction . . . . . . . . . . . . . 12.2.3 Electron Energy Loss Spectroscopy, EELS . 12.2.4 Auger Electron Spectroscopy (AES) . . . . 12.2.5 EXAFS . . . . . . . . . . . . . . . . . . . . 3

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

12.2.6 Scanning Tunneling Microscopy . . . . . . . . . . . . . . . . . . . . . 13 Amorphous Semiconductors 13.1 Introduction . . . . . . . . . . . . . . . . . . . . . . 13.1.1 Structure of Amorphous Semiconductors . . 13.1.2 Electronic States . . . . . . . . . . . . . . . 13.1.3 Optical Properties . . . . . . . . . . . . . . 13.1.4 Transport Properties . . . . . . . . . . . . . 13.1.5 Applications of Amorphous Semiconductors 13.2 Amorphous Semiconductor Superlattices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

156 165 165 166 167 173 175 175 176

A Time Dependent Perturbation Theory 179 A.1 General Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179 A.2 Fermi Golden Rule . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183 A.3 Time Dependent 2nd Order Perturbation Theory . . . . . . . . . . . . . . . 184 B Harmonic Oscillators, Phonons, and the B.1 Harmonic Oscillators . . . . . . . . . . . B.2 Phonons . . . . . . . . . . . . . . . . . . B.3 Phonons in 3D Crystals . . . . . . . . . B.4 Electron-Phonon Interaction . . . . . . . Electron-Phonon Interaction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186 186 188 189 192

Chapter 1

Review of Fundamental Relations for Optical Phenomena

References:
G. Beke and A.H. Barrett, Electromagnetic Vibrations Waves and Radiation, MIT Press, Cambridge, MA J.D. Jackson, Classical Electrodynamics, Wiley, New York, 1975 Bassani and PastoriParravicini, Electronic States and Optical Transitions in Solids, Pergamon Press, NY (1975). Yu and Cardona, Fundamentals of Semiconductors, Springer Verlag (1996)

1.1

Introductory Remarks on Optical Probes

The optical properties of solids provide an important tool for studying energy band structure, impurity levels, excitons, localized defects, lattice vibrations, and certain magnetic excitations. In such experiments, we measure some observable, such as reectivity, transmission, absorption, ellipsometry or light scattering; from these measurements we deduce the dielectric function (), the optical conductivity (), or the fundamental excitation frequencies. It is the frequency-dependent complex dielectric function () or the complex conductivity (), which is directly related to the energy band structure of solids. The central question is the relationship between experimental observations and the electronic energy levels (energy bands) of the solid. In the infrared photon energy region, information on the phonon branches is obtained. These issues are the major concern of Part II of this course. 1

1.2

The Complex dielectric function and the complex optical conductivity

The complex dielectric function and complex optical conductivity are introduced through Maxwells equations (c.g.s. units) H 1 D 4 = j c t c 1 B =0 c t (1.1) (1.2) (1.3) (1.4)

E+

D =0 B =0 where we have assumed that the charge density is zero. The constitutive equations are written as: D = E B = H j = E

(1.5) (1.6) (1.7)

Equation 1.5 denes the quantity from which the concept of the complex dielectric function will be developed. When we discuss nonlinear optics (see Chapter 11), these linear constitutive equations (Eqs. 1.51.7) must be generalized to include higher order terms in E E and E E E. From Maxwells equations and the constitutive equations, we obtain a wave equation for the eld variables E and H:
2

E=

2 E 4 E + 2 c2 t2 c t

(1.8)

and

4 H 2 H + 2 . c2 t2 c t For optical elds, we must look for a sinusoidal solution to Eqs. 1.8 and 1.9
2

H=

(1.9)

E = E0 ei(Krt)

(1.10)

where K is a complex propagation constant and is the frequency of the light. A solution similar to Eq. 1.10 is obtained for the H eld. The real part of K can be identied as a wave vector, while the imaginary part of K accounts for attenuation of the wave inside the solid. Substitution of the plane wave solution Eq. 1.10 into the wave equation Eq. 1.8 yields the following relation for K: K 2 = 2 4i . c2 c2 c 2 (1.11)

If there were no losses (or attenuation), K would be equal to K0 = (1.12)

and would be real, but since there are losses we write K= complex c (1.13)

where we have dened the complex dielectric function as complex = + 4i = 1 + i2 . (1.14)

As shown in Eq. 1.14 it is customary to write 1 and 2 for the real and imaginary parts of complex . From the denition in Eq. 1.14 it also follows that complex = 4i 4i + = complex , 4i (1.15)

where we dene the complex conductivity complex as: complex = + 4i (1.16)

Now that we have dened the complex dielectric function complex and the complex conductivity complex , we will relate these quantities in two ways: 1. to observables such as the reectivity which we measure in the laboratory, 2. to properties of the solid such as the carrier density, relaxation time, eective masses, energy band gaps, etc. After substitution for K in Eq. 1.10, the solution Eq. 1.11 to the wave equation (Eq. 1.8) yields a plane wave z 4i 1 + E(z, t) = E0 eit exp i . c

(1.17)

For the wave propagating in vacuum ( = 1, = 1, = 0), Eq. 1.17 reduces to a simple plane wave solution, while if the wave is propagating in a medium of nite electrical conductivity, the amplitude of the wave exponentially decays over a characteristic distance given by = c c = N2 () k() (1.18)

where is called the optical skin depth, and k is the imaginary part of the complex index of refraction (also called the extinction coecient) N () = complex = 1 + 4i = n() + ik(). (1.19)

This means that the intensity of the electric eld, |E|2 , falls o to 1/e of its value at the surface in a distance 1 c = (1.20) abs 2 k() 3

where abs () is the absorption coecient for the solid at frequency . Since light is described by a transverse wave, there are two possible orthogonal directions for the E vector in a plane normal to the propagation direction and these directions determine the polarization of the light. For cubic materials, the index of refraction is the same along the two transverse directions. However, for anisotropic media, the indices of refraction may be dierent for the two polarization directions, as is further discussed in 2.1.

1.3

Relation of Complex Dielectric Function to Observables

In relating complex and complex to the observables, it is convenient to introduce a complex index of refraction Ncomplex Ncomplex = complex (1.21) where K= Ncomplex c (1.22)

and where Ncomplex is usually written in terms of its real and imaginary parts (see Eq. 1.19) Ncomplex = n + ik = N1 + iN2 . (1.23)

The quantities n and k are collectively called the optical constants of the solid, where n is the index of refraction and k is the extinction coecient. (We use the tilde over the optical constants n and k to distinguish them from the carrier density and wave vector which are denoted by n and k). The extinction coecient k vanishes for lossless materials. For non-magnetic materials, we can take = 1, and this will be done in writing the equations below. With this denition for Ncomplex , we can relate complex = 1 + i2 = ( + ik)2 n yielding the important relations 1 = n 2 k 2 2 = 2 k n (1.24) (1.25) (1.26)

where we note that 1 , 2 , n and k are all frequency dependent. Many measurements of the optical properties of solids involve the normal incidence reectivity which is illustrated in Fig. 1.1. Inside the solid, the wave will be attenuated. We assume for the present discussion that the solid is thick enough so that reections from the back surface can be neglected. We can then write the wave inside the solid for this one-dimensional propagation problem as Ex = E0 ei(Kzt) (1.27)

where the complex propagation constant for the light is given by K = (/c) Ncomplex . On the other hand, in free space we have both an incident and a reected wave: Ex = E1 ei(
z t) c

+ E2 ei(

z t) c

(1.28)

Figure 1.1: Schematic diagram for normal incidence reectivity. From Eqs. 1.27 and 1.28, the continuity of Ex across the surface of the solid requires that E0 = E 1 + E 2 . (1.29)

With E in the x direction, the second relation between E0 , E1 , and E2 follows from the continuity condition for tangential Hy across the boundary of the solid. From Maxwells equation (Eq. 1.2) we have H i E = H (1.30) = c t c which results in Ex i = Hy . z c (1.31)

The continuity condition on Hy thus yields a continuity relation for Ex /z so that from Eq. 1.31 (1.32) E0 K = E1 E2 = E0 Ncomplex c c c or E1 E2 = E0 Ncomplex . (1.33) The normal incidence reectivity R is then written as R= E2 E1
2

(1.34)

which is most conveniently related to the reection coecient r given by r= E2 . E1 (1.35)

From Eqs. 1.29 and 1.33, we have the results 1 E2 = E0 (1 Ncomplex ) 2 1 E1 = E0 (1 + Ncomplex ) 2 so that the normal incidence reectivity becomes 1 Ncomplex R= 1 + Ncomplex
2

(1.36) (1.37)

(1 n)2 + k 2 (1 + n)2 + k 2

(1.38)

where the reectivity R is a number less than unity. We have now related one of the physical observables to the optical constants. To relate these results to the power absorbed and transmitted at normal incidence, we utilize the following relation which expresses the idea that all the incident power is either reected, absorbed, or transmitted 1=R+A+T (1.39)

where R, A, and T are, respectively, the fraction of the power that is reected, absorbed, and transmitted as illustrated in Fig. 1.1. At high temperatures, the most common observable is the emissivity, which is equal to the absorbed power for a black body or is equal to 1 R assuming T =0. As a homework exercise, it is instructive to derive expressions for R and T when we have relaxed the restriction of no reection from the back surface. Multiple reections are encountered in thin lms. The discussion thus far has been directed toward relating the complex dielectric function or the complex conductivity to physical observables. If we know the optical constants, then we can nd the reectivity. We now want to ask the opposite question. Suppose we know the reectivity, can we nd the optical constants? Since there are two optical constants, n and k , we need to make two independent measurements, such as the reectivity at two dierent angles of incidence. Nevertheless, even if we limit ourselves to normal incidence reectivity measurements, we can still obtain both n and k provided that we make these reectivity measurements for all frequencies. This is possible because the real and imaginary parts of a complex physical function are not independent. Because of causality, n() and k() are related through the KramersKronig relation, which we will discuss in Chapter 6. Since normal incidence measurements are easier to carry out in practice, it is quite possible to study the optical properties of solids with just normal incidence measurements, and then do a KramersKronig analysis of the reectivity data to obtain the frequencydependent dielectric functions 1 () and 2 () or the frequencydependent optical constants n() and k(). In treating a solid, we will need to consider contributions to the optical properties from various electronic energy band processes. To begin with, there are intraband processes which correspond to the electronic conduction by free carriers, and hence are more important in conducting materials such as metals, semimetals and degenerate semiconductors. These intraband processes can be understood in their simplest terms by the classical Drude theory, or in more detail by the classical Boltzmann equation or the quantum mechanical density matrix technique. In addition to the intraband (free carrier) processes, there are interband 6

processes which correspond to the absorption of electromagnetic radiation by an electron in an occupied state below the Fermi level, thereby inducing a transition to an unoccupied state in a higher band. This interband process is intrinsically a quantum mechanical process and must be discussed in terms of quantum mechanical concepts. In practice, we consider in detail the contribution of only a few energy bands to optical properties; in many cases we also restrict ourselves to detailed consideration of only a portion of the Brillouin zone where strong interband transitions occur. The intraband and interband contributions that are neglected are treated in an approximate way by introducing a core dielectric constant which is often taken to be independent of frequency and external parameters.

1.4

Units for Frequency Measurements

The frequency of light is measured in several dierent units in the literature. The relation between the various units are: 1 eV = 8065.5 cm1 = 2.418 1014 Hz = 11,600 K. Also 1 eV corresponds to a wavelength of 1.2398 m, and 1 cm1 = 0.12398 meV = 3 1010 Hz.

Chapter 2

Drude TheoryFree Carrier Contribution to the Optical Properties

2.1 The Free Carrier Contribution

In this chapter we relate the optical constants to the electronic properties of the solid. One major contribution to the dielectric function is through the free carriers. Such free carrier contributions are very important in semiconductors and metals, and can be understood in terms of a simple classical conductivity model, called the Drude model. This model is based on the classical equations of motion of an electron in an optical electric eld, and gives the simplest theory of the optical constants. The classical equation for the drift velocity of the carrier v is given by dv mv m + = eE0 eit (2.1) dt where the relaxation time is introduced to provide a damping term, (mv/ ), and a sinusoidally time-dependent electric eld provides the driving force. To respond to a sinusoidal applied eld, the electrons undergo a sinusoidal motion which can be described as v = v0 eit so that Eq. 2.1 becomes (mi + m )v0 = eE0 (2.2)

(2.3)

and the amplitudes v0 and E0 are thereby related. The current density j is related to the drift velocity v0 and to the carrier density n by j = nev0 = E0 (2.4)

thereby introducing the electrical conductivity . Substitution for the drift velocity v 0 yields v0 = eE0 (m/ ) im 8 (2.5)

into Eq. 2.4 yields the complex conductivity = ne2 . m(1 i ) (2.6)

In writing in the Drude expression (Eq. 2.6) for the free carrier conduction, we have suppressed the subscript in complex , as is conventionally done in the literature. In what follows we will always write and to denote the complex conductivity and complex dielectric constant and suppress subscripts complex in order to simplify the notation. A more elegant derivation of the Drude expression can be made from the Boltzmann formulation, as is done in Part I of the notes. In a real solid, the same result as given above follows when the eective mass approximation can be used. Following the results for the dc conductivity obtained in Part I, an electric eld applied in one direction can produce a force in another direction because of the anisotropy of the constant energy surfaces in solids. Because of the anisotropy of the eective mass in solids, j and E are related by the tensorial relation, j = E (2.7)

thereby dening the conductivity tensor as a second rank tensor. For perfectly free electrons in an isotropic (or cubic) medium, the conductivity tensor is written as: 0 0 = 0 0 0 0

(2.8)

and we have our usual scalar expression j = E. However, in a solid, can have odiagonal terms, because the eective mass tensors are related to the curvature of the energy bands E(k) by 1 1 2 E(k) . (2.9) = 2 m k k h The tensorial properties of the conductivity follow directly from the dependence of the conductivity on the reciprocal eective mass tensor. As an example, semiconductors such as CdS and ZnO exhibit the wurtzite structure, which is a non-cubic structure. These semiconductors are uniaxial and contain an optic axis (which for the wurtzite structure is along the c-axis), along which the velocity of propagation of light is independent of the polarization direction. Along other directions, the velocity of light is dierent for the two polarization directions, giving rise to a phenomenon called birefringence. Crystals with tetragonal or hexagonal symmetry are uniaxial. Crystals with lower symmetry have two axes along which light propagates at the same velocity for the two polarizations of light, and are therefore called biaxial. Even though the constant energy surfaces for a large number of the common semiconductors are described by ellipsoids and the eective masses of the carriers are given by an eective mass tensor, it is a general result that for cubic materials (in the absence of externally applied stresses and magnetic elds), the conductivity for all electrons and all the holes is described by a single scalar quantity . To describe conduction processes in hexagonal materials we need to introduce two constants: for conduction along the high symmetry axis and for conduction in the basal plane. These results can be directly demonstrated by summing the contributions to the conductivity from all carrier pockets. 9

In narrow gap semiconductors, m is itself a function of energy. If this is the case, the Drude formula is valid when m is evaluated at the Fermi level and n is the total carrier density. Suppose now that the only conduction mechanism that we are treating in detail is the free carrier mechanism. Then we would consider all other contributions in terms of the core dielectric constant core to obtain for the total complex dielectric function () = core () + 4i/ so that () = ne2 /m (1 i )1 (2.11) (2.10)

in which 4/ denotes the imaginary part of the free carrier contribution. If there were no free carrier absorption, = 0 and = core , and in empty space = core = 1. From the Drude theory, = core + 4i ne2 = (1 + i2 ) = (n1 + ik2 )2 . m(1 i ) (2.12)

It is of interest to consider the expression in Eq. 2.12 in two limiting cases: low and high frequencies.

2.2

Low Frequency Response:

4ine2 . m

In the low frequency regime (

1) we obtain from Eq. 2.12 core + (2.13)

Since the free carrier term in Eq. 2.13 shows a 1/ dependence as 0, this term dominates in the low frequency limit. The core dielectric constant is typically 16 for geranium, 12 for silicon and perhaps 100 or more, for narrow gap semiconductors like PbTe. It is also of interest to note that the core contribution and free carrier contribution are out of phase. To nd the optical constants n and k we need to take the square root of . Since we are large, we can for the moment ignore the core contribution will see below that n and k to obtain: 4ne2 i = n + ik (2.14) m
i 1+i i=e4 = (2.15) 2 we see that in the low frequency limit n k, and that n and k are both large. Therefore the normal incidence reectivity can be written as

and using the identity

R=

( 12 ) + k 2 n ( + 12 ) + k 2 n

n2 + k 2 2 n 4 n =1 2 + k 2 + 2 2 + k2 n n n

2 . n

(2.16)

Thus, the Drude theory shows that at low frequencies a material with a large concentration of free carriers (e.g., a metal) is a perfect reector. 10

2.3

High Frequency Response;

4ne2 . m 2

In this limit, Eq. 2.12 can be approximated by: core (2.17)

As the frequency becomes large, the 1/ 2 dependence of the free carrier contribution guarantees that free carrier eects will become less important, and other processes will dominate. In practice, these other processes are the interband processes which in Eq. 2.17 are dealt with in a very simplied form through the core dielectric constant core . Using this approximation in the high frequency limit, we can neglect the free carrier contribution in Eq. 2.17 to obtain = core = real. (2.18) Equation 2.18 implies that n > 0 and k = 0 in the limit of 1, with R ( 1)2 n ( + 1)2 n (2.19)

where n = core . Thus, in the limit of very high frequencies, the Drude contribution is unimportant and the behavior of all materials is like that for a dielectric.

2.4

The Plasma Frequency

Thus, at very low frequencies the optical properties of semiconductors exhibit a metal-like behavior, while at very high frequencies their optical properties are like those of insulators. A characteristic frequency at which the material changes from a metallic to a dielectric response is called the plasma frequency p , which is dened as that frequency at which the real part of the dielectric function vanishes 1 ( p ) = 0. According to the Drude theory (Eq. 2.12), we have = 1 + i2 = core + 4i ne2 m(1 i ) 1 + i 1 + i (2.20)

where we have written in a form which exhibits its real and imaginary parts explicitly. We can then write the real and imaginary parts 1 () and 2 () as: 1 () = core 4ne2 2 m(1 + 2 2 ) 2 () = 4 ne2 . m(1 + 2 2 ) (2.21)

The free carrier term makes a negative contribution to 1 which tends to cancel the core contribution shown schematically in Fig. 2.1. We see in Fig. 2.1 that 1 () vanishes at some frequency ( p ) so that we can write 1 ( p ) = 0 = core which yields p = 2 4ne2 2 m(1 + p 2 ) 2 (2.22)

1 1 4ne2 2 = p 2 . mcore 2 11

(2.23)

Figure 2.1: The frequency dependence 1 (), showing the denition of the plasma frequency p by the relation 1 ( p ) = 0.

2 Since the term (1/ 2 ) in Eq. 2.23 is usually small compared with p , it is customary to neglect this term and to identify the plasma frequency with p dened by 2 p =

4ne2 mcore

(2.24)

in which screening of free carriers occurs through the core dielectric constant core of the medium. If core is too small, then 1 () never goes positive and there is no plasma frequency. The condition for the existence of a plasma frequency is core > 4ne2 2 . m (2.25)

The quantity p in Eq. 2.24 is called the screened plasma frequency in the literature. Another quantity called the unscreened plasma frequency obtained from Eq. 2.24 by setting core = 1 is also used in the literature. The general appearance of the reectivity as a function of photon energy for a degenerate semiconductor or a metal is shown in Fig. 2.2. At low frequencies, free carrier conduction dominates, and the reectivity is 100%. In the high frequency limit, we have R ( 1)2 n , ( + 1)2 n (2.26)

which also is large, if n 1. In the vicinity of the plasma frequency, 1 (1 ) is small by denition; furthermore, 2 (p ) is also small, since from Eq. 2.21 2 (p ) = ne2 4 mp 1 + (p )2 12 (2.27)

Figure 2.2: Reectivity vs for a metal or a degenerate semiconductor in a frequency range where interband transitions are not important and the plasma frequency p occurs near the minimum in reectivity R. and if p 1

core 2 (p ) = p

(2.28)

so that 2 (p ) is often small. With 1 (p ) = 0, we have from Eq. 1.25 n = k, and 2 (p ) = 2 . We thus see that n tends to be small near and consequently R is also 2 k n 2 n p small (see Fig. 2.2). The steepness of the dip at the plasma frequency is governed by the relaxation time ; the longer the relaxation time , the sharper the plasma structure. In metals, free carrier eects are almost always studied by reectivity techniques because of the high optical absorption of metals at low frequency. For metals, the free carrier conductivity appears to be quite well described by the simple Drude theory. In studying free carrier eects in semiconductors, it is usually more accurate to use absorption techniques, which are discussed in Chapter 11. Because of the connection between the optical and the electrical properties of a solid through the conductivity tensor, transparent materials are expected to be poor electrical conductors while highly reecting materials are expected to be reasonably good electrical conductors. It is, however, possible for a material to have its plasma frequency just below visible frequencies, so that the material will be a good electrical conductor, yet be transparent at visible frequencies. Because of the close connection between the optical and electrical properties, free carrier eects are sometimes exploited in the determination of the carrier density in instances where Hall eect measurements are dicult to make. The contribution of holes to the optical conduction is of the same sign as for the electrons, since the conductivity depends on an even power of the charge ( e2 ). In terms of the complex dielectric constant, we can write the contribution from electrons and holes as = core + n e e2 e n h e2 h 4i + me (1 ie ) mh (1 ih ) (2.29)

where the parameters ne , e , and me pertain to the electron carriers and nh , h , and mh are for the holes. The plasma frequency is again found by setting 1 () = 0. If there are 13

multiple electron or hole carrier pockets, as is common for semiconductors, the contributions from each carrier type is additive, using a formula similar to Eq. 2.29. We will now treat another conduction process in Chapter 3 which is due to interband transitions. In the above discussion, interband transitions were included in an extremely approximate way. That is, interband transitions were treated through a frequency independent core dielectric constant core (see Eq. 2.12). In Chapter 3 we consider the frequency dependence of this important contribution.

14

Chapter 3

Interband Transitions
3.1 The Interband Transition Process

In a semiconductor at low frequencies, the principal electronic conduction mechanism is associated with free carriers. As the photon energy increases and becomes comparable to the energy gap, a new conduction process can occur. A photon can excite an electron from an occupied state in the valence band to an unoccupied state in the conduction band. This is called an interband transition and is represented schematically by the picture in Fig. 3.1. In this process the photon is absorbed, an excited electronic state is formed and a hole is left behind. This process is quantum mechanical in nature. We now discuss the factors that are important in these transitions. 1. We expect interband transitions to have a threshold energy at the energy gap. That is, we expect the frequency dependence of the real part of the conductivity 1 () due to an interband transition to exhibit a threshold as shown in Fig. 3.2 for an allowed electronic transition. 2. The transitions are either direct (conserve crystal momentum k: Ev (k) Ec (k)) or indirect (a phonon is involved because the k vectors for the valence and conduction bands dier by the phonon wave vector q). Conservation of crystal momentum yields kvalence = kconduction qphonon . In discussing the direct transitions, one might wonder about conservation of crystal momentum with regard to the photon. The reason we need not be concerned with the momentum of the photon is that it is very small in comparison to Brillouin zone dimensions. For a typical optical wavelength of 6000 , the wave vector for the photon K = 2/ 105 cm1 , while a typical dimension A across the Brillouin zone is 108 cm1 . Thus, typical direct optical interband processes excite an electron from a valence to a conduction band without a signicant change in the wave vector. 3. The transitions depend on the coupling between the valence and conduction bands and this is measured by the magnitude of the momentum matrix elements coupling the valence band state v and the conduction band state c: | v|p|c |2 . This dependence results from Fermis Golden Rule (see Chapter A) and from the discussion on the perturbation interaction H for the electromagnetic eld with electrons in the solid (which is discussed in 3.2). 15

Figure 3.1: Schematic diagram of an allowed interband transition.

Figure 3.2: Real part of the conductivity for an allowed optical transition. We note that 1 () = (/4)2 ().

16

4. Because of the Pauli Exclusion Principle, an interband transition occurs from an occupied state below the Fermi level to an unoccupied state above the Fermi level. 5. Photons of a particular energy are more eective in producing an interband transition if the energy separation between the 2 bands is nearly constant over many k values. In that case, there are many initial and nal states which can be coupled by the same photon energy. This is perhaps easier to see if we allow a photon to have a small band width. That band width will be eective over many k values if Ec (k) Ev (k) doesnt vary rapidly with k. Thus, we expect the interband transitions to be most important for k values near band extrema. That is, in Fig. 3.1 we see that states around k = 0 make the largest contribution per unit bandwidth of the optical source. It is also for this reason that optical measurements are so important in studying energy band structure; the optical structure emphasizes band extrema and therefore provides information about the energy bands at specic points in the Brillouin zone. Although we will not derive the expression for the interband contribution to the conductivity, we will write it down here to show how all the physical ideas that were discussed above enter into the conductivity equation. We now write the conductivity tensor relating the interband current density j in the direction which ows upon application of an electric eld E in direction j = E (3.1) as = e2 m2
i,j

i|p |j j|p |i [f (Ei ) f (Ej )] Ei E j [i + 1/ + (i/ )(Ei Ej )] h

(3.2)

in which the sum in Eq. 3.2 is over all valence and conduction band states labelled by i and j. Structure in the optical conductivity arises through a singularity in the resonant denominator of Eq. 3.2 [i + 1/ + (i/ )(Ei Ej )] discussed above under properties (1) h and (5). The appearance of the Fermi functions f (Ei ) f (Ej ) follows from the Pauli principle in property (4). The dependence of the conductivity on the momentum matrix elements accounts for the tensorial properties of (interband) and relates to properties (2) and (3). In semiconductors, interband transitions usually occur at frequencies above which free carrier contributions are important. If we now want to consider the total complex dielectric constant, we would write = core + 4i [Drude + interband ] . (3.3)

The term core contains the contributions from all processes that are not considered explicitly in Eq. 3.3; this would include both intraband and interband transitions that are not treated explicitly. We have now dealt with the two most important processes (intraband and interband) involved in studies of electronic properties of solids. If we think of the optical properties for various classes of materials, it is clear from Fig. 3.3 that major dierences will be found from one class of materials to another. 17

Figure 3.3: Structure of the valence band states and the lowest conduction band state at the point in germanium.

18

Figure 3.4: Absorption coecient of germanium at the absorption edge cor3/2 responding to the transitions 25 1/2 2 (D1 ) and 25 2 (D2 ). The en1/2 ergy separation between the 25 and 3/2 25 bands is determined by the energy dierences between the D1 and D2 structures.

3.1.1

Insulators

Here the band gap is suciently large so that at room temperature, essentially no carriers are thermally excited across the band gap. This means that there is no free carrier absorption and that interband transitions only become important at relatively high photon energies (above the visible). Thus, insulators frequently are optically transparent.

3.1.2

Semiconductors

Here the band gap is small enough so that appreciable thermal excitation of carriers occurs at room temperature. Thus there is often appreciable free carrier absorption at room temperature either through thermal excitation or doping. In addition, interband transitions occur in the infrared and visible. As an example, consider the direct interband transition in germanium and its relation to the optical absorption. In the curve in Fig. 3.4, we see that the optical absorption due to optical excitation across the indirect bandgap at 0.7 eV is very small compared with the absorption due to the direct interband transition shown in Fig. 3.4. (For a brief discussion of the spinorbit interaction as it aects interband transitions see 3.4.)

3.1.3

Metals

Here free carrier absorption is extremely important. Typical plasma frequencies are p h = 10 eV which occur far out in the ultraviolet. In the case of metals, interband transitions typically occur at frequencies where free carrier eects are still important. Semimetals, like metals, exhibit only a weak temperature dependence with carrier densities almost inde19

pendent of temperature. Although the carrier densities are low, the high carrier mobilities nevertheless guarantee a large contribution of the free carriers to the optical conductivity.

3.2

Form of the Hamiltonian in an Electromagnetic Field

A proof that the optical eld is inserted into the Hamiltonian in the form p p e A/c follows. Consider the classical equation of motion: d 1 1 A 1 (mv) = e E + (v H) = e + v( dt c c t c A) (3.4)

where and A are, respectively, the scalar and vector potentials, and E and B are the electric and magnetic elds given by E= (1/c) A/t B= A. e (A v) c (3.5)

Using standard vector identities, the equation of motion Eq. 3.4 becomes d e (mv + A) = dt c (e) + (3.6)

where [ (A v)]j denotes vi Ai /xj in which we have used the Einstein summation convention that repeated indices are summed and where we have used the vector relations A dA = + (v dt t and [v ( If we write the Hamiltonian as H= and then use Hamiltons equations v= H 1 e = (p A) p m c (3.10) e 1 (p A)2 + e 2m c (3.9) A)]i = vj )A (3.7)

Ai Aj vj . xi xj

(3.8)

e (A v) (3.11) p = H = e + c we can show that Eqs. 3.4 and 3.6 are satised, thereby verifying that Eq. 3.9 is the proper form of the Hamiltonian in the presence of an electromagnetic eld, which has the same form as the Hamiltonian without an optical eld except that p p (e/c) A. The same transcription is used when light is applied to a solid and is then called the Luttinger transcription. The Luttinger transcription is used in the eective mass approximation where the periodic potential is replaced by the introduction of k (1/i) and m m . 20

The reason why interband transitions depend on the momentum matrix element can be understood from perturbation theory. At any instance of time, the Hamiltonian for an electron in a solid in the presence of an optical eld is H= (p e/cA)2 e2 A 2 p2 e Ap+ + V (r) = + V (r) 2m 2m mc 2mc2 (3.12)

in which A is the vector potential due to the optical elds, V(r) is the periodic potential. Thus, the one-electron Hamiltonian without optical elds is H0 = and the optical perturbation terms are H = e2 A 2 e Ap+ . mc 2mc2 (3.14) p2 + V (r) 2m (3.13)

Optical elds are generally very weak (unless generated by powerful lasers) and we usually consider only the term linear in A, the linear response regime. The form of the Hamiltonian in the presence of an electromagnetic eld is derived in this section, while the momentum matrix elements v|p|c which determine the strength of optical transitions also govern the magnitudes of the eective mass components (see 3.3). This is another reason why optical studies are very important. To return to the Hamiltonian for an electromagnetic eld (Eq. 3.9), the coupling of the valence and conduction bands through the optical elds depends on the matrix element for the coupling to the electromagnetic eld perturbation e H p A. (3.15) = mc With regard to the spatial dependence of the vector potential we can write A = A0 exp[i(K r t)] (3.16)

where for a loss-less medium K = n/c = 2 / is a slowly varying function of r since n 2 / is much smaller than typical wave vectors in solids. Here n, , and are, respectively, n the real part of the index of refraction, the optical frequency, and the wavelength of light.

3.3

Relation between Momentum Matrix Elements and the Eective Mass

Because of the relation between the momentum matrix element v|p|c , which governs the electromagnetic interaction with electrons and solids, and the band curvature ( 2 E/k k ), the energy band diagrams provide important information on the strength of optical transitions. Correspondingly, knowledge of the optical properties can be used to infer experimental information about E(k). We now derive the relation between the momentum matrix element coupling the valence and conduction bands v|p|c and the band curvature ( 2 E/k k ). We start with Schodingers equation in a periodic potential V (r) having the Bloch solutions r nk (r) = eikr unk (r), 21 (3.17)

Hnk (r) = En (k)nk (r) =

p2 + V (r) eikr unk (r) = En (k)eikr unk (r). 2m

(3.18)

Since p is an operator ( /i) , we can write h peikr unk (r) = eikr (p + k)unk (r). h Therefore the dierential equation for unk (r) becomes p2 k p 2 k2 h h + V (r) + + unk (r) = En (k)unk (r) 2m m 2m (3.20) (3.19)

giving the following dierential equation for the periodic function unk (r) = unk (r + Rm ) p2 2 k2 h k p h unk (r) = En (k) unk (r) + V (r) + 2m m 2m (3.21)

which we write as follows to put Eq. 3.21 in the canonical form for application of the perturbation theory formulae p2 + V (r) (3.22) H0 = 2m H = k p h m 2 k2 h 2m (3.23) (3.24) (3.25)

En (k) = En (k) to yield

[H0 + H ]unk (r) = En (k)unk (r).

Assume that we know the solution to Eq. 3.25 about a special point k0 in the Brillouin zone which could be a band extremum, such as k0 = 0. Then the perturbation formulae Eqs. 3.22 3.25 allow us to nd the energy and wave function for states near k0 . For simplicity, we carry out the expansion about the center of the Brillouin zone k = 0, which is the most important case in practice; the extension of this argument to an energy extremum at arbitrary k0 is immediate. Perturbation theory then gives: En (k) = En (0) + (un,0 |H |un,0 ) + (un,0 |H |un ,0 )(un ,0 |H |un,0 ) . En (0) En (0) n =n (3.26)

The rst order term (un,0 |H |un,0 ) in Eq. 3.26 normally vanishes about an extremum because of inversion symmetry, with H being odd under inversion and the two wavefunctions unk (r) both being even or both being odd. Since H = the matrix element is then written as (un,0 |H |un ,0 ) = h k (un,0 |p|un ,0 ). m (3.28) k p h m (3.27)

We now apply Eq. 3.26 to optical transitions, for the simplest case of a two band model. Here we assume that: 22

1. bands n and n (valence (v) and conduction (c) bands) are close to each other and far from other bands 2. interband transitions occur between these two bands separated by an energy gap E g . We note that the perturbation theory is written in terms of the energy E n (k) En (k) = En (k) 2 k2 h . 2m (3.29)

Assuming that the rst order term in perturbation theory (Eq. 3.26) can be neglected by parity (even and oddness) arguments, we obtain for En (k) about k= 0 En (k) = En (0) + |(v|p |c)(c|p |v)| 2 h k k 2 m Eg (3.30)

or in terms of the energy eigenvalues of Schrdingers equation (Eq. 3.18) o En (k) = En (0) + |(v|p |c)(c|p |v)| 2 k2 h 2 h + 2 k k . 2m m Eg 2 h 2 (3.31)

We dene the eective mass tensor by the relation En (k) = En (0) + so that k k
,

1 m

(3.32)

1 m

2 |(v|p |c)(c|p |v)| + 2 m m Eg

(3.33)

where is the unit matrix. This discussion shows that the non-vanishing momentum matrix element is responsible for the inequality between the free electron m and the eective mass m in the solid. With regard to the optical properties of solids we note that the same momentum matrix element that governs the eective mass formula (Eq. 3.33) also governs the electromagnetic interaction given by Eq. 3.15. Thus small eective masses tend to give rise to strong coupling between valence and conduction bands and large values for |(v|p|c)| 2 . On the other hand, small eective masses lead to a small density of states because of the m3/2 dependence of the density of states.

3.4

Spin-Orbit Interaction in Solids

Reference:
Jones and March, pp. 85-87, 89-94. Eisberg and Resnick, Quantum Physics pp. 278-281. 1020 A spin angular momentum Sz = /2 and a magnetic moment B = |e| /2mc = 0.927 h h erg/gauss is associated with each electron. The magnetic moment and spin angular momentum for the free electron are related by = |e| |e| h S= S mc mc 2 23 (3.34)

Figure 3.5: Schematic diagram showing the splitting of the = 1 level by the spinorbit interaction.

(S is a unit vector along S), and and S are oppositely directed because the electron is negatively charged. An electron in an atom sees a magnetic eld because of its own orbital motion and consequently there is an interaction called the spin-orbit interaction whereby the magnetic eld due to the orbital motion of the electron tends to line up its magnetic moment along the magnetic eld: HS.O. = H. (3.35) HS.O. = 1 ( V p) S 2m2 c2 (3.36)

since eE V . For an atom Eq. 3.36 results in HS.O.atom = (r)L S. (3.37)

A detailed discussion of this topic is found in any standard quantum mechanics text. This spin-orbit interaction gives rise to a spin-orbit splitting of the atomic levels corresponding to dierent values of the total angular momentum J J =L+S where L and S, respectively, denote the orbital and spin angular momentum. Thus J J = (L + S) (L + S) = L L + S S + (L S + S L) (3.39) (3.38)

in which the operators L and S commute. We take matrix elements in the |j, , s, mj representation, because m , ms are not good quantum numbers, to obtain, with j = | s|, (| s| + 1), . . . , + s, j(j + 1) = ( + 1) + s(s + 1) + 2 L S so that the expectation value of L S in the |j, , s, mj representation becomes: 1 L S = [j(j + 1) ( + 1) s(s + 1)] 2 (3.41) (3.40)

For p states, = 1, s = 1/2 and j = 3/2 or 1/2 as shown in Fig. 3.5. From Eq. 3.41 we can nd the expectation value of L S . In particular, we note that the degeneracy of an s-state is unaected by the spin-orbit interaction. On the other hand, a d-state is split 24

up into a doublet D5/2 (6-fold degenerate) and D3/2 (4-fold degenerate). Thus, the spinorbit interaction does not lift all the degeneracy for atomic states. To lift this additional degeneracy it is necessary to apply a magnetic eld. The magnitude of the spin-orbit interaction depends also on the expectation value of (r) dened by the following relation,

n, j, , s, mj |HS.O. |n, j, , s, mj = j, , s, mj |L S|j, , s, mj where the atomic wave function is written =Y

m (, )Rn

Rn (r) Rn dr

(3.42)

(r)

(3.43)

and Rn (r) denotes the radial part of the atomic wave function. We note that the integral over r in Eq. 3.42 increases rapidly with atomic number ( Z 3 or Z 4 ). The physical reason behind this sensitive dependence on Z is that heavier atoms have more electrons generating larger H elds, and therefore a greater spin-orbit splitting results. References for tabulated spin-orbit splittings are: C.E. Moore Atomic Energy Levels (National Bureau of Standards, Circular #467), vol. 1 (1949), vol. 2 (1952) and vol. 3 (1958). These references give the measured spectroscopic levels for any atom in a large number of excited congurations. The lowest Z values are in vol. 1, the highest in vol. 3. F. Herman and S. Skillman Atomic Structure Calculation (Prentice-Hall, Inc. 1963). Most complete listing of calculated atomic levels. Landolt and Bornstein Physical and Chemical Tables (many volumes in Reference section in the Science Library). For most atomic species that are important in semiconductor physics, the spin-orbit interaction is important. Some typical values are: semiconductor diamond silicon germanium tin InSb In Sb GaAs Ga As PbTe, HgTe Pb Hg Te atomic number Z=6 Z = 14 Z = 32 Z = 50 Z = 49 Z = 51 Z = 31 Z = 33 Z = 82 Z = 80 Z = 52 25 -point splitting = 0.006eV = 0.044eV = 0.290eV = 0.527eV = 0.274eV = 0.815eV = 0.103eV = 0.364eV = 1.746eV = 1.131eV = 1.143eV

Figure 3.6: Energy bands of Ge: (a) without and (b) with spinorbit interaction. The listing above gives the point splittings. The spin-orbit splittings are k-dependent and at the L-point are typically about 2/3 of the point value. The one-electron Hamiltonian for a solid including spin-orbit interaction is from Eq. 3.36 H= p2 1 + V (r) ( V p) S. 2m 2m2 c2 (3.44)

When the electron spin is considered, the wave functions consist of a spatial and a spin part. The eect of the spin-orbit interaction is to introduce a partial lifting of the degeneracy of band states at high symmetry points in the Brillouin zone. Also, it is a convention in the literature to use a dierent labeling scheme for the energy bands when the spin-orbit interaction is included. To show the eect of the spin-orbit interaction on the energy bands of a semiconductor, consider the energy bands for germanium. We show in Fig. 3.6 the E(k) vs. k along the (100) axis, (111) axis and (110) axes for no spin-orbit interaction and with spin-orbit interaction. As an example of the eect of the spin-orbit interaction, consider the valence band at the -point (k = 0) which is labeled by 25 when there is no spin-orbit interaction. The 25 band is triply degenerate at k = 0, each of the three orbital levels containing a spin up and a spin down electron. With spin-orbit interaction, this band splits into the + (doubly 8 degenerate) band and the + (non-degenerate) band. In the literature, the + band is 7 7 called the split-o band. In germanium the band gap is 0.8eV and the splitting between the + and + bands is 0.3eV. However, in InSb, the spin-orbit interaction is large and the 8 7 separation between the upper valence band and the split-o band is 0.9eV, which is much larger than the band gap of 0.2eV between the valence and conduction bands. 26

Chapter 4

The Joint Density of States and Critical Points

References:
Jones and March, Theoretical Solid State Physics: pp. 806-814 Bassani and PastoriParravicini, Electronic States and Optical Transitions in Solids: chapter 5 Yu and Cardona, Fundamentals of Semiconductors, pp. 251-258 Madelung, Introduction to Solid State Theory: pp. 262-271

4.1

The Joint Density of States

The detailed calculation of the contribution to () due to interband transitions is rather dicult. It is therefore instructive to obtain an approximate answer by use of the Fermi Golden Rule (Eq. A.32). The Golden Rule gives us the probability per unit time Wk that a photon of energy makes a transition at a given k point: h 2 | v|H |c |2 [Ec (k) Ev (k) ] h Wk = h (4.1)

where the matrix element for the electromagnetic perturbation H is taken between the valence and conduction band Bloch states at wave vector k and the -function [Ec Ev ] h which expresses energy conservation is also evaluated at k. In writing Eq. 4.1, we exploit the fact that the wave vector for the light is small compared to the Brillouin zone dimensions. Because the electronic states in the Brillouin zone are quasicontinuous functions of k, to obtain the lineshape for an interband transition, we must integrate over k. Recognizing that both the perturbation matrix elements and the joint density of states are k-dependent, we obtain upon integration of Eq. 4.1 over k space W = 2 h | v|H |c |2 2 (Ec (k) Ev (k) ) d3 k h 8 3 27 (4.2)

for a 3D system. For 2D and 1D systems, we replace [d3 k/(2)3 ] by [d2 k/(2)2 ] and [dk/(2)], respectively. The perturbation Hamiltonian for the electromagnetic interaction is simply eA p H = (4.3) mc where the time dependence of the vector potential A has already been taken into account, so that A is a vector with only spatial dependence. In taking matrix elements of the perturbation Hamiltonian, we need then only consider matrix elements of the momentum operator connecting the valence and conduction bands. In practical cases it is often not necessary to evaluate these matrix elements explicitly because it is precisely these momentum matrix elements that determine the experimentally measured eective masses (see 3.3). If we assume for simplicity that | v|H |c |2 is independent of k, then the remaining integral is the joint density of states between the valence and conduction bands cv ( ). For a 3D system, h we thus dene cv ( ) as h cv ( ) h 2 8 3 [Ec (k) Ev (k) ] d3 k h (4.4)

and cv ( ) is the number of states per unit volume per unit energy range which occur with h an energy dierence between the conduction and valence bands equal to the photon energy. As explained above, cv ( ) can be evaluated in a similar manner for 2D and 1D systems. h We would now like to look at this joint density of states (Eq. 4.4) in more detail to see why the optical properties of solids give unique information about the energy band structure. The main point is that optical measurements provide information about the bands at particular k points in the Brillouin zone, usually points of high symmetry and near energy band extrema. This can be understood by casting cv ( ) in a more transparent h form. We start with the denition of the joint density of states given in Eq. 4.4. It is convenient to convert this integral over k-space to an integral over energy. This is done by introducing a constant energy surface S in k-space such that the energy dierence E c Ev = is the photon energy. Then we can introduce the constant energy surfaces S and S + dS h in reciprocal space as corresponding to a constant energy dierence between the conduction and valence bands at each k point and: d3 k = dS dkn (4.5)

where dkn is an element of a wave vector normal to S, as shown in Fig. 4.1. By denition of the gradient, we have | k E|dkn = dE so that for surfaces with energy dierence Ec Ev we write: | Therefore d3 k = dkn dS = dS so that cv ( ) = h 2 8 3
k (Ec

Ev )|dkn = d(Ec Ev ). d(Ec Ev ) | k (Ec Ev )|

(4.6)

(4.7)

dS d(Ec Ev )(Ec Ev ) h . | k (Ec Ev )| 28

(4.8)

Figure 4.1: Adjacent constant energy dierence surfaces in reciprocal space, S and S + dS, where the energy dierence is between valence and conduction bands, and dkn is the normal to these constant energy dierence surfaces.

We now carry out the integral over d(Ec Ev ) to obtain cv ( ) = h 2 8 3 dS . | k (Ec Ev )|Ec Ev = h (4.9)

Of special interest are those points in the Brillouin zone where (Ec Ev ) is stationary and k (Ec Ev ) vanishes. At such points, called joint critical points, the denominator of the integrand in Eq. 4.9 vanishes and especially large contributions can be made to cv ( ). h This can be understood on the basis of physical considerations. Around critical points, the photon energy = (Ec Ev ) is eective in inducing electronic transitions over a relatively h larger region of the Brillouin zone than would be the case for transitions about non-critical points. The relatively large contributions to the transition probability for critical points gives rise to structure observed in the frequency dependence of the optical properties of solids. Critical points generally occur at high symmetry points in the Brillouin zone, though this is not necessarily the case. As an illustration, let us consider the energy bands of the semiconductor germanium (see Fig. 4.2). Here we see that both the valence and conduction bands have extrema at the point, k = 0, although the lowest conduction band minimum is located at the L point. For the band extrema at k = 0, the condition [Ec (k = 0) Ev (k = 0)] = gives h rise to critical points in the joint density of states. Notice also that around the L points, extrema occur in both valence and conduction bands, and a critical point therefore results. Since the energy dierence [Ec Ev ] has a relatively small gradient as we move away from the L point, this critical point participates more fully in the interband transitions. In fact, for germanium, Fig. 4.2 shows that there are large regions along the (100) and (111) axes where the energy separation between valence and conduction bands (E c Ev ) is roughly constant. These large regions in k-space make very large contributions to the dielectric function. We can see these features directly by looking at the frequency dependence of the real and imaginary parts of the dielectric function for germanium (see Fig. 4.3). Here we see that at low photon energies (below 2 eV), where the interband transitions from the 29

Figure 4.2: E(k) for a few high symmetry directions in germanium, neglecting the spin-orbit interaction.

25 valence band to the 2 conduction band dominate, the contributions to the real and imaginary parts of the dielectric function are small. On the other hand, the contributions from the large regions of the Brillouin zone along the (100) and (111) axes between 2 and 5 eV are very much more important, as is seen in Fig. 4.3 for both 1 () and 2 (). In describing this contribution to the dielectric function of germanium we say that the valence and conduction bands track each other and in this way produce a large joint density of states over large regions of the Brillouin zone. A similar situation occurs in silicon and in common III-V semiconductors. The diagram in Fig. 4.2 shows that beyond 5 eV there is no longer any signicant tracking of the valence and conduction bands. Consequently, the magnitudes of 1 () and 2 () fall sharply beyond 5 eV. The absolute magnitudes of 1 and 2 for germanium and other semiconductors crystallizing in the diamond or zincblende structure are relatively large. We will see shortly when we discuss the Kramers-Kronig relations in 6.1 that these large magnitudes of 1 and 2 are responsible for the large value of 1 ( 0) in these materials. For germanium 1 (0) is 16 from Fig. 4.3.

4.2

Critical Points

For a 3D system, critical points (often called Van Hove singularities) are classied into four categories depending on whether the band separations are increasing or decreasing as we move away from the critical point. This information is found by expanding [Ec (k) Ev (k)] in a Taylor series around the critical point k0 which is at an energy dierence extremum. 30

Figure 4.3: Frequency dependence of the real (1 ) and imaginary (2 ) parts of the dielectric function for germanium. The solid curves are obtained from an analysis of experimental normal-incidence reectivity data while the dots are calculated from an energy band model.

31

Figure 4.4: Summary of the joint density of states for a 3D system near each of the distinct type of critical point. Thus,
3

Ec (k) Ev (k) = Eg (k0 ) +

i=1

ai (ki k0i )2

(4.10)

where the energy gap at the expansion point is written as Eg (k0 ) and the sum is over the three directions x, y, and z. The coecients ai represent the second derivative of the 2 energy dierence k2 [Ec (k) Ev (k)]. The classication of the critical points in a 3D system i shown in Fig. 4.4 is made according to how many ai coecients in Eq. 4.10 are negative. The shapes given for the joint density of states curves of Fig. 4.4 are obtained as is here illustrated for the case of an M0 singularity for a 3D system. In the case of 2D and 1D systems, there are 3 and 2 types of critical points, respectively, using the same denition of the coecients ai to dene the type of critical point. As an example we will calculate cv ( ) for an M0 singularity in a 3D system assuming h simple parabolic bands (see Fig. 4.5). Here, Ec (k) = 2 k2 h Eg + 2 2mc 32 (4.11)

Figure 4.5: Bands associated with a M0 critical point for a 3D system. and

Eg 2 k2 h (4.12) 2 2mv where Eg is the energy gap, and mc and mv are eective masses for the conduction and valence bands, respectively, and mv is taken as a positive number. We thus obtain Ev (k) = Ec (k) Ev (k) = Eg + 1 2 k2 1 h + 2 mc mv = Eg + 2 k2 h 2mr (4.13)

where we dene the reduced mass mr through the relation 1 1 1 = + . mr mc mv Taking the gradient of Ec Ev yields
k (Ec Ev ) =

(4.14)

2k h mr

(4.15)

so that the joint density of states becomes cv ( ) = h or cv ( ) = h 2 8 3 dS | k (Ec Ev )| Ec Ev = h =

Ec Ev = h

(4.16)

2 4 k 2 mr 8 3 2 k h

mr k 2 2 h

.
Ec Ev = h

(4.17)

We evaluate k in Eq. 4.17 from the condition Ec E v = = E g + h or k= 2mr ( Eg ) h 2 h 33 2 k2 h 2mr

1/2

(4.18)

(4.19)

Figure 4.6: Two cases of band extrema which are associated with M 0 critical points. (a) Conduction band minimum and a valence band maximum and (b) Both bands showing minima. so that cv ( ) = h 1 2mr 2 2 2 h

3/2

Eg h

(4.20)

as shown in Fig. 4.4 for an M0 critical point. The expression for cv ( ) in Eq. 4.20 is not h singular but represents a discontinuity in slope at = Eg . This discontinuity in slope h corresponds to a threshold for the absorption process, as discussed in Chapter 5. On the other hand, the situation is quite dierent for the joint density of states corresponding to an M0 critical point for a 3D system in a magnetic eld, as we will see in Part III of the class notes. At a critical point, the joint density of states in a magnetic eld does show singularities where the density of states in a magnetic eld becomes innite. These singularities in a magnetic eld make it possible to carry out resonance experiments in solids, despite the quasicontinuum of the energy levels in the energy bands E(k). We note that we can have M0 -type critical points for bands that look like Fig. 4.6a or like Fig. 4.6b. It is clear that the dierence Ec Ev in Fig. 4.6b varies more slowly around the critical point than it does in Fig. 4.6a. Thus, bands that tend to track each other have an exceptionally high joint density of states and contribute strongly to the optical properties. Examples of bands that track each other are found in common semiconductors like germanium along the (111) direction (see Figs. 4.2 and 4.3). In addition to the M0 critical points, we have M1 , M2 , and M3 critical points in 3D systems. The functional forms for the joint density of states for < Eg and > Eg h h are given in Table 4.1. From the table we see that in 2D, the M0 and M2 critical points correspond to discontinuities in the joint density of states at Eg , while the M1 singularity corresponds to a saddle point logarithmic divergence. In the case of the 1D system, both the M0 and M1 critical points are singular.

34

Table 4.1: Functional form for the joint density of states for various types of singularities below and above the energy gap Eg for 3D, 2D, and 1D systems vc ( ). h Type M0 M1 M2 M3 M0 M1 M2 M0 M0 < Eg h 0 C (Eg )1/2 h C (Eg )1/2 h 0 ln(Eg ) h C 0 (Eg )1/2 h > Eg h ( Eg )1/2 h C C ( Eg )1/2 h 0 C ln( Eg ) h 0 ( Eg )1/2 h 0

3D

2D

1D

35

Chapter 5

Absorption of Light in Solids

References:
Ziman, Principles of the Theory of Solids: Chapter 8 Bassani and PastoriParravicini, Electronic States and Optical Transitions in Solids: chapter 5 Yu and Cardona, Fundamentals of Semiconductors, Chapter 6 Wolfe, Holonyak and Stillman, Physical Properties of Semiconductors, Chapter 7

5.1

The Absorption Coecient

Measurement of the absorption of light is one of the most important techniques for optical measurements in solids. In the absorption measurements, we are concerned with the light intensity I(z) after traversal of a thickness z of material as compared with the incident intensity I0 , thereby dening the absorption coecient abs (): I(z) = I0 eabs ()z (5.1)

where the absorption constant is shown schematically in Fig. 5.1. Since the intensity I(z) depends on the square of the eld variables, it immediately follows that abs () = 2 k() c (5.2)

where the factor of 2 results from the denition of abs () in terms of the light intensity, which is proportional to the square of the elds. This expression tells us that the absorption coecient is proportional to k(), the imaginary part of the complex index of refraction (extinction coecient), so that k is usually associated with power loss. We note that Eq. 5.2 applies to free carrier absorption in semiconductors in the limit 1, and p . We will now show that the frequency dependence of the absorption coecient is quite dierent for the various physical processes which occur in the optical properties of solids. We will consider here the frequency dependence of the absorption coecient for: 1. Free carrier absorption 36

Figure 5.1: Frequency dependence of the absorption coecient near a threshold for interband transitions.

(a) typical semiconductor abs () 2 2. Direct interband transitions

(b) metals at low frequencies abs () 2

( Eg ) 2 h h
1

(b) conservation of crystal momentum

(a) form of absorption coecient abs ()

(c) relation between m and momentum matrix element (d) form of abs () for direct forbidden transition 3. Indirect interband transitions (a) form of absorption coecient abs () ( Eg q )2 h h
( Eg ) 2 h h
3

(b) phonon absorption and emission processes

The summary given above is for 3D systems. In the case of 2D and 1D systems, the functional dependence is sensitive to the dimensionality of the system for each process.

5.2

Free Carrier Absorption in Semiconductors

For free carrier absorption we use the relation for the complex dielectric function () = 1 () + i2 () given by 4i () = 0 + (5.3) where 0 is the core dielectric constant in the optical frequency range above the lattice mode frequencies and 0 is here assumed to be independent of . The electronic polarizability is related to the frequency dependent electrical conductivity by the frequency dependent Drude term ne2 = . (5.4) m (1 i ) 37

The plasma frequency p is then given by the vanishing of 1 (), that is 1 (p ) = 0 or

2 p =

4ne2 . m 0

(5.5)

For semiconductors, the core dielectric constant 0 is typically a large number and the contribution due to the free carriers is small at infrared and visible frequencies. For metals, the free carrier absorption is dominant over the entire optical frequency range. For semiconductors, the typical frequency range of interest is that above the optical phonon frequencies, and for these frequencies it is generally true that 1. We can then write: 2 i0 p (1 + i ) 4ine2 (1 + i ) () = 0 + = 0 + (5.6) m [1 + ( )2 ] [1 + ( )2 ] or for 1 2 2 i0 p 2 0 p 2 . (5.7) () 0 + 3 3 In the range of interest for optical measurements in a semiconductor, the relation p is satised. It is then convenient to express the complex dielectric function () in terms of the optical constants n() and k() according to the denition () = [ () + ik()]2 n where n() is the index of refraction and k() is the extinction coecient. We can then write for the real part of the dielectric function: 1 () n2 () k 2 () 0 (5.8)

where the index of refraction n() is large and the extinction coecient k() is small. For the imaginary part of the dielectric function, we have 2 () 2 ()k() 2 0 k() = n which is small, since p
2 0 p 2 3 3

(5.9)

. Thus the absorption coecient can be written as: 2 2 0 p 2 k() 2 0 p abs () = = c c 2 0 3 c 2

(5.10)

and thus abs () is proportional to 1/ 2 or to 2 for free carrier absorption in semiconductors for the case where 1 and p . Figure 5.2 shows a plot of the optical absorption coecient for InAs vs wavelength on a log-log plot for various carrier densities, showing that abs () p where p is between 2 and 3 for a wide range of donor concentrations. The dependence of the reectivity spectra (vs wavelength) for various donor concentrations for heavily doped n-type InSb is shown in Fig. 5.3. The dependence of the plasma frequency on the carrier concentration is readily visible from these data.

5.3

Free Carrier Absorption in Metals

The typical limits for metals are somewhat dierent than for semiconductors. In particular we consider here the case where 1, p , |0 | 4/, so that n k. Thus we obtain 4i 4ine2 () i2 () 2i k 2ik 2 n (5.11) m 38

Figure 5.2: Free carrier absorption in n-type InAs at room temperature for six dierent carrier concentrations (in units of 1017 cm3 ) A: 0.28; B: 0.85; C: 1.4; D: 2.5; E: 7.8; and F: 39.0.

39

Figure 5.3: Plasma edges observed in the room temperature reectivity spectra of n-type InSb with carrier concentration n varying between 3.5 1017 cm3 and 4.0 1018 cm3 . The solid curves are theoretical ts to the experimental points, including consideration of the energy dependence of m due to the strong interband coupling (called non-parabolic eects).

40

 This gives us for the extinction coecient k() k() = and the absorption coecient becomes: 8ne2 2 k() = (5.13) c m c2 For this limit abs () is proportional to . Usually, the convenient observable for metals is the reectivity. In the limit appropriate for metals, n = k, and both n and k are large. We thus have abs () = R= ( 1)2 + k 2 n n2 2 + 1 + k 2 n 4 n = =1 2 + k2 2 + 2 + 1 + k 2 2 + k 2 + 2 + 1 ( + 1) n n n n n R1 (5.14) (5.15) 2ne2 m (5.12)

2 4 n 1 . n n2 + k 2 But from Eq. 5.12 and the condition n k 1, we obtain n() so that the reectivity goes as R() 12 m . 2ne2 2me2 m

(5.16)

(5.17)

Equation 5.17 is known as the Hagen-Rubens relation which holds well for most metals in the infrared region of the spectrum. This formula also applies to degenerate semiconductors below the plasma frequency.

5.4

Direct Interband Transitions

To calculate the absorption due to direct interband transitions we go back to the denition for the absorption coecient abs () which is dened as the power removed from the incident beam per unit volume per unit incident ux of electromagnetic energy: abs () = ( ) number of transitions/unit volume/unit time h . incident electromagnetic ux c Re(E H). 8 (5.18)

The incident electromagnetic ux is calculated from the Poynting vector S= (5.19)

It is convenient to relate the eld variables to the vector potential: E= 1 A i A = c t c 41 (5.20)

H = B =

A.

(5.21)

In nonmagnetic materials we can take the permeability to be unity. In taking the curl of A, we assume a plane wave form A = A0 ei(Krt) (5.22)

where the propagation constant for the light is denoted by the wave vector K. We thus obtain for the Poynting vector S= or S= i c Re A (iK A) 8 c (5.23)

Re (A A)K (A K)A . 8

(5.24)

Utilizing the fact that for a transverse plane wave A K = 0, we obtain S= n 2 |A| K 8 c (5.25)

where n denotes the real part of the complex index of refraction and K is a unit vector along the Poynting vector. This quantity |S| in Eq. 5.25 becomes the denominator in Eq. 5.18 which is the expression dening the absorption coecient. The transition probability/unit time/unit volume is calculated from the Fermi Golden Rule W = 2 | Hvc |2 cv ( ). h h (5.26)

If we wish to consider the absorption process at nite temperature, we also need to include the Fermi functions to represent the occupation of the states at nite temperature f (Ev )[1 f (Ec )] f (Ec )[1 f (Ev )] (5.27)

in which the rst group of terms represents the absorption process which depends on the valence band (v) being nearly full and the conduction band (c) being nearly empty. The second group of terms represents the emission process which proceeds if there are occupied conduction states and unoccupied valence states. Clearly, the Fermi functions in Eq. 5.27 simply reduce to [f (Ev ) f (Ec )]. The matrix elements |Hvc |2 in Eq. 5.26 can be written in terms of the electromagnetic interaction Hamiltonian Hvc = v|Hem |c = e mc v|A(r, t) p|c . (5.28)

We show in 5.5 that the matrix element v|A(r, t)p|c coupling the valence and conduction bands for the electromagnetic interaction is diagonal in wave vector k since the wave vector for light K is small relative to Brillouin zone dimensions. As a result also the spatial dependence of the vector potential can be ignored. Thus the square of the matrix elements coupling the valence and conduction bands becomes |Hvc |2 = e mc
2

|A|2 | v|p|c |2 ,

(5.29)

42

or

where | v|p|c |2 couples states with the same electron wave vector in the valence and conduction bands. Since | v|p|c |2 is slowly varying with k in comparison to cv ( ) it is convenient h 2 and evaluate this quantity at the M critical point. to neglect the k dependence of | v|p|c | i Thus for direct interband transitions, we obtain the following expression for the absorption coecient e h ( )[ 2 ( mc )2 |A|2 | v|p|c |2 cv ( )][f (Ev ) f (Ec )] h h abs () = (5.30) n 2 8 c |A| 16 2 e2 | v|p|c |2 cv ( )[f (Ev ) f (Ec )] h (5.31) m2 c n where n in Eqs. 5.30 and 5.31 denotes the index of refraction. To get an idea of the functional forms of the quantities in Eq. 5.31, we will consider a rather simplied picture of two simple parabolic bands with an allowed optical transition, i.e., a non-vanishing momentum matrix element coupling them. Writing the joint density of states from Eq. 4.4 for the case of an M0 critical point (as occurs near k = 0 for many semiconductors) 1 2mr 3/2 Eg h (5.32) cv ( ) = 2 h 2 2 h abs () = where mr is the reduced mass for the valence and conduction bands, we can estimate the absorption coecient abs (). At very low temperature, a semiconductor has an essentially lled valence band and an empty conduction band; that is f (Ev ) = 1 and f (Ec ) = 0. We can estimate | v|p|c |2 from the eective mass sumrule (Eq. 3.33) | v|p|c |2 m 0 Eg m 0 2 m (5.33)

where m0 is the free electron mass. After substitution of Eqs. 5.32 and 5.33 into Eq. 5.31 we obtain the following frequency dependence for the absorption coecient for direct allowed transitions: 1 abs () Eg h (5.34) so that the direct opticallyallowed interband transitions are characterized by a threshold at the energy gap Eg as shown in Fig. 5.1. We thus see a very dierent frequency dependence of abs () for the various physical processes. It is sometimes convenient to express the absorption in terms of the imaginary part of the dielectric function nc abs () (5.35) 2 () = which from Eq. 5.31 becomes 2 () = 4e m
2

| v|p|c |2 cv ( )[f (Ev ) f (Ec )]. h

(5.36)

If we introduce the dimensionless quantity fvc , which is usually called the oscillator strength and is dened by 2| v|p|c |2 2| v|p|c |2 fvc = = , (5.37) m[Ec (k) Ev (k)] m h 43

Figure 5.4: Plot of the square of the absorption coecient of PbS as a function of photon energy showing the linear dependence of [abs ()]2 on . The intercept h with the x-axis denes the direct energy gap.

we obtain the following result for 2 () at T = 0 2 () = 8 2 e2 h fvc cv ( ). h m (5.38)

We further discuss how 1 () for interband transitions is obtained from 2 () in 6.2 using the KramersKronig relation. To illustrate the t between these simple models and the behavior of the absorption coecient near the fundamental absorption edge, we show in Fig. 5.4 a plot of [ abs ]2 vs for PbS, with the intercept of [abs ]2 on the photon energy axis giving the direct energy h band gap. By plotting abs () on a log scale vs , a value for the energy gap can also be h obtained as shown in Fig. 5.5 for InSb. The derivation of the functional form for the absorption coecient for direct forbidden transitions proceeds as in the derivation of Eq. 5.31, except that | v|p|c |2 is now dependent 44

Figure 5.5: Semilogarithmic plot of the absorption coecient of InSb at 5 K as a function of photon energy. The lled circles represent experimental results. The curves have been calculated using various models. Best results are obtained when the dependence of the matrix elements on k are included. The intercept with the x-axis gives the direct bandgap of InSb.

45

on k 2 so that abs () follows a ( Eg )3/2 dependence for direct forbidden interband h transitions.

5.4.1

Temperature Dependence of Eg

Because of the expansion and contraction of the lattice with temperature, the various band parameters, particularly the energy gap is temperature dependent. Although calculations are available to predict and account for the T dependence of the band gap at the fundamental absorption edge (threshold), Eg (T ) is best found by empirical ts. We give below expressions for such ts which are useful for research purposes Eg (T ) = 1.165 2.84 104 T Eg (T ) = 0.742 3.90 104 T Eg (T ) = 1.522 Eg (T ) = 2.338
5.8104 T 2 T +300 6.2104 T 2 T +460

(eV) (eV) (eV) (eV)

Si Ge GaAs GaP

For Group IV and IIIV compound semiconductors, Eg (T ) decreases with increasing T , as shown above, but for IVVI compounds, Eg (T ) increases with increasing T .

5.4.2

Dependence of Absorption Edge on Fermi Energy

For lightly doped semiconductors, EF lies in the bandgap and the absorption edge occurs at Eg , neglecting excitonic eects which are discussed in Chapter 7. However, for heavily doped semiconductors, EF lies in the valence or conduction bands and the threshold for optical absorption is shifted. This shift in the absorption edge is often referred to as the Burstein shift, and is illustrated in Fig. 5.6 where it is shown that the threshold for absorption occurs when 1 2 kBS h 2 2 kBS 1 h 2 + = Eg + = EF 4kB T (5.39) = Eg + h 2 m mh 2m e r in which m is the reduced mass, (1/m ) = (1/m ) + (1/m ), kBS is the wave vector r r e h corresponding to the Burstein shift dened in Eq. 5.39. Referring to Eq. 5.27 where we introduce the probability that the initial state is occupied and the nal state is unoccupied, we nd that since doping aects the position of the Fermi level, the Fermi functions will depend on carrier concentration for heavily doped semiconductors. In particular the quantity (1 f0 ) denoting the availability of nal states will be aected by the Burstein shift. If we write 2 k2 h = E Ec 2m e (5.40)

where Ec is the energy at the bottom of the conduction band, then the probability that the nal state is empty is 1 1 = (5.41) 1 f0 = ( Eg )m h 1 + exp[(EF E)/kB T ] EF E h 1 + exp kB T (m +m )kB T
e h

and Eq. 5.41 should be used for the probability of nal states in evaluating f (Ec ) in Eq. 5.31. 46

Figure 5.6: Diagram showing how the fundamental absorption edge of an n-type semiconductor is shifted to higher energy by doping. The wave vector for the Burstein shift k BS is dened in Eq. 5.39.

5.4.3

Dependence of Absorption Edge on Applied Electric Field

The electron wave functions in the valence and conduction bands have an exponentially decaying amplitude in the energy gap. In the presence of an electric eld E, a valence band electron must tunnel through a triangular barrier to reach the conduction band. In the absence of photon absorption, the height of the barrier is Eg and its thickness is Eg /e|E| where |E| is the magnitude of the electric eld, as shown in Fig. 5.7(a). The eect of the photon, as shown in Fig. 5.7(b), is to lower the barrier thickness to t( ) = h Eg h e|E| (5.42)

so that the tunneling probability is enhanced by photon absorption. Figure 5.8 shows that the absorption edge is eectively lowered by the presence of the electric eld, and the eect of the electric eld on abs is particularly pronounced below the zero eld band gap. The eect of an electric eld on the fundamental absorption edge is called the FranzKeldysh eect.

5.5

Conservation of Crystal Momentum in Direct Optical Transitions

For clarity we now show why the momentum matrix elements coupling two Bloch states for a perfect crystal are diagonal in k and conserve crystal momentum. It is this property of the momentum matrix elements that is responsible for direct interband transitions. We 47

Figure 5.7: Energy band diagram in an electric eld showing the wavefunction overlap (a) without and (b) with absorption of a photon of energy . h write the momentum matrix elements coupling two bands (for example, the valence and conduction bands) as n k |p|n, k = Operating with d3 reik r u k (r) n h i eikr unk (r). (5.43)

on the product function of the Bloch state yields n k |p|n, k = d3 reik r u k (r)eikr ( k + h n h )unk (r). i (5.44)

Now the term in k can be integrated immediately to give knn (k k ) and is thus h h diagonal in both band index and crystal momentum. This term therefore does not give rise to interband transitions. The remaining term in Eq. 5.44 is d3 rei(kk )r h u (r) unk (r). i nk (5.45)

The function u k (r) unk (r) in Eq. 5.45 is periodic under the translation r r + Rn n where Rn is any lattice vector. But any spatially periodic function can be Fourier expanded Fm eiGm r =
m

h u (r) unk (r) i nk

(5.46)

in terms of the reciprocal lattice vectors Gm . We thus obtain for the integral in Eq. 5.45 d3 rei(kk )r Fm eiGm r 48 (5.47)

Figure 5.8: Electric eld and photon energy dependence of the band-to-band absorption for GaAs.

49

Figure 5.9: Indirect optically induced transitions of electrons (a) from the initial state 0 in the valence band to nal states 1 and 2 in the conduction band and (b) from initial states 1 and 2 in the valence band to nal state 0 in the conduction band. In both (a) and (b) a phonon labeled by ( s , qs ) is absorbed in the indirect transition process. h which vanishes unless k k + Gm = 0. (5.48)

Since k k must be within the rst Brillouin zone, k and k can only dier by the reciprocal lattice vector Gm 0. Thus Eq. 5.47 vanishes unless k = k and we have demonstrated that because of the periodicity of the crystal lattice, the momentum matrix elements coupling two bands can only do so at the same value of crystal momentum k. Since the probability for optical transitions involves the same momentum matrix elements as occur in the determination of the eective mass in the transport properties, study of the optical properties of a solid also bears an important relation to the transport properties of that material.

5.6

Indirect Interband Transitions

In making indirect transitions, the semiconductor can either emit or absorb a phonon of energy q h = E f Ei q h h (5.49)

in which Ef and Ei are, respectively, the energies of the nal and initial electron states and the signs refer to phonon emission (+ sign) or absorption ( sign). To review indirect interband transitions in a semiconductor, we derive below an expression for the absorption coecient for the situation where a phonon is absorbed in the indirect process, as shown schematically in Fig. 5.9. Similar arguments can then be applied to the case where a phonon is emitted. 50

Figure 5.10: Schematic diagram of an indirect transition showing the notation used in the text. Ec is the energy of the L-point conduction band at wave vector kc , while Ec is the thermal energy gap. The valence band maximum Ev is taken at the zero of energy. En and (kn kc ), respectively, denote the energy and momentum of an excited electron, while Ep and kp , respectively, denote the corresponding parameters for the holes near k = 0. It is customary to place the zero of energy at the valence band maximum.

The conservation of energy principle is applied to the total process, consisting of the direct optical transition and the absorption of a phonon q , yielding h = Eg q + h h h 2 2 (kn kc )2 2 kp h + 2mn 2mp (5.50)

in which the notation in Eq. 5.50 is dened in Fig. 5.10, and Eg is the thermal gap or energy dierence between the conduction band minimum (e.g., at L) and the valence band maximum at . The negative sign in front of the phonon energy q in Eq. 5.50 corresponds h to the phonon absorption process. In Eq. 5.50, the term (kn kc ) denotes the dierence h between the crystal momentum kn of an excited electron in the Lpoint conduction band h and the crystal momentum kc at the Lpoint conduction band minimum. Thus the kinetic h energy of the excited electron with crystal momentum kn is h En E c = 2 (kn kc )2 h 2mn (5.51)

where En is the energy above the conduction band minimum Ec , and mn in Eq. 5.51 is the eective mass of an electron near the conduction band minimum. Since the valence band extremum is at k = 0, then kp is the crystal momentum for the h hole that is created when the electron is excited, corresponding to the kinetic energy of the hole 2 kp h 2 Ep = . (5.52) 2mp 51

The sign convention that is used in this discussion is to take Ep as a positive number and the zero of energy is taken at the valence band maximum (see Fig. 5.10). In terms of these sign conventions, conservation of energy yields = Eg q + (En Ec ) + Ep h h and conservation of momentum requires q = kn kp (5.54) (5.53)

where q is the wave vector for the absorbed phonon. We now nd the frequency dependence of the absorption edge for indirect transitions in order to make a distinction between direct and indirect transitions just from looking at the frequency dependence of the optical absorption data. Let us then consider the transition from some specic initial state Ep to a specic nal state En . The density of states c (En ) (number of states/unit volume/unit energy range) for the nal state conduction band has an energy dependence given by c (En ) (En Ec )1/2 . (5.55)

Using the conservation of energy relation in Eq. 5.53, c (En ) can be expressed in terms of Ep as c (En ) ( Eg Ep + q )1/2 . h h (5.56) Thus we see that transitions to a state En take place from a range of initial states, since Ep can vary between Ep = 0 where all of the kinetic energy is given to the electron and the opposite limit where En Ec = 0 and all of the kinetic energy is given to the hole. Let the energy denote the range of possible valence band energies between these limits = E g + q . h h The density of initial states for the valence band has an energy dependence given by
1/2 v (Ep ) Ep

(5.57)

(5.58)

where we are using the convention Ev 0 for dening the zero of energy, so that Ep vanishes at the top of the valence band. Thus the eective density of states for the phonon absorption process is found by summing over all Ep values which conserve energy,

( ) h

c (En )v (Ep )dEp

Ep

Ep dEp .

(5.59)

The integral in Eq. 5.59 can be carried out through integration by parts, utilizing the notation u = Ep , and v = Ep , and writing the limits of the integration in terms of the variable Ep 2 2v u 2 1 uv du = uv + tan . (5.60) = 4 4 u 0 8 0 0 Substitution in Eq. 5.57 for in Eqs. 5.59 and 5.60 results in ( ) h ( Eg + q )2 h h 8 52 (5.61)

where Ca is a constant for the phonon absorption process. To nd the absorption coecient for the indirect absorption process that involves the emission of a phonon, we must nd the eective density of states for the emission process. The derivation in this case is very similar to that given above for phonon absorption, except that the energy conservation condition now involves the phonon energy with the opposite sign. Furthermore, the probability of emission of a phonon is proportional to [n( q ) + 1] h which is given by
h [n( q ) + 1] = 1 + [e q /kB T 1]1 = h

so that the absorption coecient for indirect transitions in which a phonon is absorbed becomes ( Eg + q )2 h h abs () = Ca (5.63) exp( q /kB T ) 1 h

which gives the frequency dependence for the indirect interband transitions involving phonon absorption. Also, the probability for the absorption of a phonon is proportional to the BoseEinstein factor 1 (5.62) n( q ) = h exp( q /kB T ) 1 h

1 1
h e q /kB T

(5.64)

so that the absorption constant for phonon emission becomes ems () = Ce ( Eg q )2 h h 1 exp( q /kB T ) h (5.65)

where Ce is a constant for the phonon emission process. At low temperatures, the phonon emission process dominates because there are so few phonons available for the absorption process. Furthermore, as a function of photon energy, dierent thresholds are obtained for the absorption and emission processes. In the absorption process, absorption starts when = Eg q (see Fig. 5.11), while for the h h emission process, the optical absorption starts when = Eg + q . So if we plot abs () h h vs , as is shown in Fig. 5.11, then abs () in the low photon energy range will go as h abs () ( Eg + q ) while ems () will be proportional to ( Eg q ). h h h h Experimentally, a superposition of the absorption and emission processes will be observed. Some experimental data illustrating indirect interband transitions are given in Fig. 5.12. The shift of the curves in Fig. 5.12 as a function of photon energy is due to the temperature dependence of the indirect gap in silicon. In Fig. 5.12 it is easy to separate out the lower energy absorption contribution which is associated with the phonon absorption process (compare Figs. 5.11 and 5.12). At higher energies it is also easy to separate out the phonon emission contribution. By carrying out measurements at several dierent temperatures it is possible to obtain a more accurate value for q . Figure 5.12 shows that the phonon h absorption process becomes more favorable as the temperature is raised, while the emission process is less sensitive to temperature. The physical reason behind this is that for the absorption process to occur in the rst place, phonons of the appropriate wave vector must be available. In Ge the phonon assisted process requires phonons of wave vector q extending from to L, while for Si we need a phonon qvector from to min (where min corresponds to the point conduction band minimum). Since lattice vibrations are thermally excited, there are few available phonons at low temperatures, but more are 53

Figure 5.11: Schematic diagram showing the frequency dependence of the square root of the absorption coecient for indirect interband transitions near the thresholds for the phonon emission and absorption processes. The curves are for four dierent temperatures. At the lowest temperature (T4 ) the phonon emission process dominates, while at the highest temperature (T1 ) the phonon absorption process is most important at low photon energies. The magnitude of twice the phonon energy is indicated.

available at high temperatures. On the other hand, phonon emission does not depend upon the availability of phonons since the emission process itself generates phonons; for this reason the phonon emission process is relatively insensitive to temperature. Since silicon is a relatively hard material (with a Debye temperature of D = 658 K) there will only be a few large wavevector phonons excited at room temperature. Therefore the phonon emission process will dominate in the optical absorption for photon energies where such emission is energetically possible. These arguments account for the dierent slopes observed for the phonon absorption and emission contributions to the absorption coecient of Fig. 5.12. Another complication that arises in real materials is that there are several types of phonons present for a given q-vector, i.e., there are acoustic and optical branches and for each branch there are longitudinal and transverse modes. An example of the analysis of optical absorption data to obtain the frequencies of the various phonons at q = 0 is given in 1/2 h Fig. 5.13 where abs vs is plotted for the indirect gap semiconductor GaP, from which it is possible to measure q for various LO, LA, TO and TA phonons. Today such optical h data are seldom taken, because it is now customary to use inelastic neutron diraction data to plot out the entire dispersion curve for each of the phonon branches. When the phonon frequencies are high, electron energy loss spectroscopy can be helpful in obtaining q (q) for the various phonon branches as is discussed in Chapter 12.

54

Figure 5.12: Plots of the square root of the absorption coecients of Si versus photon energy at several temperatures. The two segments of a straight line drawn through the experimental points represent the two contributions associated with phonon absorption and emission. (From Macfarlane, et al., Phys. Rev. 111, 1249 (1958)).

55

Figure 5.13: Plots of the square root of the absorption coecients of GaP vs photon energy at two dierent temperatures. The labels denote the various absorption thresholds associated with the emission of various phonon modes. The observation of these phonon modes is made possible by the enhanced absorption associated with excitons at the absorption threshold (see Chapter 7). The apparent shift in the phonon frequencies is in most part due to the variation of the bandgap energy with temperature (see Figs. 5.11 and 5.12).

56

Chapter 6

Optical Properties of Solids Over a Wide Frequency Range

6.1 KramersKronig Relations

References
Jones and March, Theoretical Solid State Physics: pp. 787-793 Jackson, Classical Electrodynamics: pp. 306-312 Measurement of the absorption coecient gives the imaginary part of the complex index of refraction while the reectivity is sensitive to a complicated combination of 1 () and 2 (). Thus from measurements such as abs () we often have insucient information to determine 1 () and 2 () independently. However, if we know either 1 () or 2 () over a wide frequency range, then 2 () or 1 () can be determined from the KramersKronig relation given by ( ) 2 2 d (6.1) 1 () 1 = P 2 2 0 and 2 2 () = P
0

in which P denotes the principal value. These relations are based on causality, linear response theory and the boundedness of physical observables. The KramersKronig relations relate 1 () and 2 () so that if either of these functions is known as a function of the other is completely determined. Because of the form of these relations, it is clear that the main contribution to 1 () comes from the behavior of 2 ( ) near due to the resonant denominator in Eqs. 6.1 and 6.2. What this means is that to obtain 1 () we really should know 2 ( ) for all , but it is more important to know 2 ( ) in the frequency range about than elsewhere. This property is greatly exploited in the analysis of reectivity data, where measurements are available over a nite range of values. Some kind of extrapolation procedure must be used for those frequencies that are experimentally unavailable. We now give a derivation of the KramersKronig relations after some introductory material. 57

1 ( ) d 2 2

(6.2)

This theorem is generally familiar to electrical engineers in another context. If a system is linear and obeys causality (i.e., there is no output before the input is applied), then the real and imaginary parts of the system function are related by a Hilbert transform. Let us now apply this causality concept to the polarization in a solid. We have the constitutive equation which denes the polarization of the solid: E = D = E + 4 P so that P = (6.3)

1 E ()E (6.4) 4 where () denes the polarizability, and P is the polarization/unit volume or the response of the solid to an applied eld E. The polarizability () in electrical engineering language is the system function () = r () + ii () (6.5) in which we have explicitly written the real and imaginary parts r () and i (), respectively. Let E(t) = E0 (t) be an impulse eld at t = 0. Then from the denition of a -function, we have: E0 E(t) = E0 (t) = cos td. (6.6) 0 The response to this impulse eld yields an in-phase term proportional to r () and an out-of-phase term proportional to i (), where the polarization vector is given by P (t) = E0
0

r () cos t + i () sin t d,

(6.7)

in which () is written for the complex polarizability (see Eq. 6.5). Since P (t) obeys causality and is bounded, we nd that the integral of ()eit is well behaved along the contour C as R and no contribution to the integral is made along the contour C in the upper half plane (see Fig. 6.1). Furthermore, the causality condition that P (t) vanishes for t < 0 requires that () have no poles in the upper half plane shown in Fig. 6.1. To nd an explicit expression for () we must generate a pole on the real axis. Then we can isolate the behavior of () at some point 0 by taking the principal value of the integral. We do this with the help of Cauchys theorem. Since () has no poles in the upper half-plane, the function [()/( 0 )] will have a single pole at = 0 (see Fig. 6.2). If we run our contour just above the real axis, there are no poles in the upper-half plane and the integral around the closed contour vanishes: ()d = 0. 0 (6.8)

Let us now consider the integral taken over the various portions of this closed contour:
C

1 The contribution over the contour C vanishes since () remains bounded, while 0 0 as R (see Fig. 6.2). Along the contour C, we use Cauchys theorem to obtain

() d + 0

0 R

() d + 0

() d + 0

R 0 +

()d = 0. 0

(6.9)

lim

0 C

() d = i(0 ) 0 58

(6.10)

Figure 6.1: Contours used in evaluating the complex polarizability integral of Eq. 6.7.

Figure 6.2: Contour used to evaluate Eq. 6.9.

59

in which (0 ) is the residue of () at = 0 and the minus sign is written because the contour C is taken clockwise. We further dene the principal part P of the integral in the limit R and 0 as
0
R

lim

() d + 0

0 +

() d P 0

() d. 0

(6.11)

The vanishing of the integral in Eq. 6.8 thus results in the relation r (0 ) + ii (0 ) = 1 P i r () + ii () d. 0 (6.12)

Equating real and imaginary parts of Eq. 6.12, we get the following relations which hold for < < ; () 1 i r (0 ) = P d (6.13) 0 where r () is even and i (0 ) = 1 P

where i () is odd. We would like to write these relations in terms of integrals over positive frequencies. We can do this by utilizing the even- and oddness of r () and i (). If we now multiply the integrand by ( + 0 )/( + 0 ) and make use of the even- and oddness of the integrands, we get: ()( + ) ()d 1 2 i 0 i r (0 ) = P d = P (6.15) 2 2 2 2 0 0 0 i (0 ) = 1 P

r () d 0

(6.14)

We have now obtained the KramersKronig relations. To avoid explicit use of the principal value of a function, we can subtract out the singularity at 0 , by writing r (0 ) + ii (0 ) = 1 i

r ()( + 0 ) 2 d = P 2 2 0

0 r ()d 2 . 2 0

(6.16)

() (0 ) 0

+ 0 d. + 0

(6.17)

Using the evenness and oddness of r () and i () we then obtain r (0 ) = and i (0 ) = 2 2

0 0

i () 0 i (0 ) d 2 2 0 0 r () 0 r (0 ) d. 2 2 0

(6.18)

(6.19)

To obtain the KramersKronig relations for the dielectric constant itself just substitute () = 1 + 4() = 1 () + i2 () to obtain 1 (0 ) 1 = 2
0

(6.20)

2 ( ) 0 2 (0 ) d 2 2 0 60

(6.21)

2 (0 ) =

The KramersKronig relations are very general and depend, as we have seen, on the assumptions of causality, linearity and boundedness. From this point of view, the real and imaginary parts of a physical quantity Q can be related by making the identication Qreal r Qimaginary i . (6.23) (6.24)

0 1 ( ) 0 1 (0 ) d . 2 2 0

(6.22)

Thus, we can identify 1 () 1 with r (), and 2 () with i (). The reason, of course, why the identication r () is made with [1 () 1] rather than with 1 () is that if 2 () 0 for all , we want 1 () 1 for all (the dielectric constant for free space). Thus, if we are interested in constructing a KramersKronig relation for the optical constants, then we again want to make the identication for ( + i k) n [ () 1] r () n k() i (). (6.25) (6.26)

From Eqs. 6.21 and 6.22, we obtain the KramersKronig relations for the optical con stants n() and k() 2 k( ) k() d (6.27) n() 1 = 0 2 2 and 2 k() =
0

where we utilize the denition relating the complex dielectric function () to the optical constants n() and k() where () = [ () + ik()]2 . n It is useful to relate the optical constants to the reection coecient r() exp[i()] dened by n() 1 + ik() r() exp[i()] = (6.29) n() + 1 + ik() and the reectivity is given as R() = r 2 (). From Eq. 6.29, we can then write n() = k() = 1 r 2 () 1 + r 2 () 2r() cos () 1+ r 2 () 2r() sin () 2r() cos () (6.30) (6.31)

( ) () n n d 2 2

(6.28)

so that once r() and () are found, the optical constants n() and k() are determined. In practice r() and () are found from the reectivity R which is measured over a wide frequency range and is modeled outside the measured range. A KramersKronig relation can be written for the conjugate variables ln r() and (), from which () is found: ln r() = 2
0

( ) () d 2 2 61

(6.32)

() =

where ln R() = 2 ln r(). From a knowledge of the frequency dependent reectivity R(), the reection coecient r() and the phase of the reectivity coecient () can be found. We can then nd the frequency dependence of the optical constants n() and k() which in turn yields the frequency dependent dielectric functions 1 () and 2 (). Starting with the experimental data for the reectivity R() for germanium in Fig. 6.3(a), the KramersKronig relations are used to obtain results for 1 () and 2 () for germanium as shown in Fig. 6.3(b). The KramersKronig relations for the conjugate variables 1 () and 2 (); n() and k(); and ln r() and () are widely used in quantitative studies of the optical properties of specic materials, as for example germanium in Fig. 6.3.

ln r( ) ln r() d . 2 2

(6.33)

6.2

Optical Properties and Band Structure

If we are interested in studying the optical properties near the band edge such as the onset of indirect transitions or of the lowest direct interband transitions, then we should carry out absorption measurements (Chapter 5) to determine the absorption coecient abs () and thus identify the type of process that is dominant (indirect, direct, allowed, forbidden, etc.) at the band edge. However, if we are interested in the optical properties of a semiconductor over a wide energy range, then we want to treat all bands and transitions within a few eV from the Fermi level on an equal footing. Away from the band edge, the absorption coecients become too high for the absorption technique to be useful, and reectivity measurements are made instead. Experimentally, it is most convenient to carry out reectivity measurements at normal incidence. From these measurements, the Kramers Kronig analysis (see 6.1) is used to get the phase angle () for some frequency 0 , if the reection coecient r() is known throughout the entire range of photon energies (0 ) = 20
0

ln r() ln r(0 ) d. 2 2 0

(6.34)

From a knowledge of r() and (), we can then nd the frequency dependence of the optical constants n() and k() using Eqs. 6.30 and 6.31 and the frequency dependent dielectric function 1 () = n2 k 2 (6.35) 2 () = 2 k. n (6.36) As an example of such an analysis, let us consider the case of the semiconductor germanium. The normal incidence reectivity is given in Fig. 6.3(a) and the results of the Kramers Kronig analysis described above are given for 1 () and 2 () in Fig. 6.3(b). Corresponding to the structure in the reectivity, there will be structure observed in the real and imaginary parts of the dielectric function. These structures in the reectivity data are then identied with special features in the energy band structure. It is interesting to note that the indirect transition (0.66 eV) from the 25 valence band to the L1 conduction band (see Part I of the notes) has almost no impact on the reectivity data. Nor does the direct band gap, which is responsible for the fundamental absorption edge in germanium, 62

Figure 6.3: (a) Frequency dependence of the reectivity of Ge over a wide frequency range. (b) Plot of the real [1 ()] and imaginary [2 ()] parts of the dielectric functions for Ge obtained by a KramersKronig analysis of (a).

63

Table 6.1: The measured energies (eV) of the prominent structures in the optical spectra of some semiconductors with the diamond and zinc-blende structures. All energies are low temperature values except that of the E0 transition in Si, which was measured at room temperature. Transition E0 E0 + 0 E1 E1 + 1 E0 E0 + 0 E2 E1 Si 4.185 4.229 3.45 3.378 4.33 5.5 Ge 0.898 1.184 2.222 2.41 3.206 3.39 4.49 5.65 GaAs 1.5192 1.859 3.017 3.245 4.488 4.659 5.11 6.63 InP 1.4236 1.532 3.15 3.835 4.54 5.05 GaP 2.869 2.949 3.785 4.77 5.21 6.8

have a signicant eect on the reectivity data. These eects are small on the scale of the reectivity structures shown in Fig. 6.3(a) and must be looked for with great care in a narrow frequency range where structure in the absorption data is found. The big contribution to the dielectric constant comes from interband transitions L3 L1 for which the joint density of states is large over large volumes of the Brillouin zone. The sharp rise in 2 () at 2.1 eV is associated with the L3 L1 transition. For higher photon energies, large volumes of the Brillouin zone contribute until a photon energy of about 5 eV is reached. Above this photon energy, we cannot nd bands that track each other closely enough to give interband transitions with intensities of large magnitude.

6.3

Modulated Reectivity Experiments

If we wish to study the critical point contribution to the optical reectivity in more detail, it is useful to carry out modulated reectivity measurements. If, for example, a small periodic perturbation is applied to a sample then there will be a change in reectivity at the frequency of that perturbation. The frequency dependence of this change in reectivity is small (parts in 103 or 104 ) but it is measurable. As an example, we show in Fig. 6.4, results for the reectivity R() and for the wavelength modulated reectivity (1/R)(dR/dE) of GaAs. Structure at E0 would be identied with the direct band gap, while the structure at E0 + 0 corresponds to a transition from the split-o valence band at k = 0 which arises through the spin-orbit interaction. The transitions at E1 and E1 + correspond to point and L point transitions, also showing spin-orbit splitting. Also identied in Fig. 6.5 are the E0 transition from the 7 valence band to the 6 conduction band, and the E2 transition from X5 X5 at the X point. Although the band structure and notation given in Fig. 6.5 applies to Ge in detail, the results for other group IV and IIIV semiconductors is qualitatively similar, with values for the pertinent interband transitions given in Table 6.1 for Si, Ge, GaAs, InP and GaP. In the vicinity of a critical point, the denominator in the joint density of states is small, so that a small change in photon energy can produce a signicant change in the joint density 64

Figure 6.4: Reectance and frequency modulated reectance spectra for GaAs. (a) Room temperature reectance spectrum and (b) the wavelength modulated spectrum (1/R)(dR/dE) (the solid curve is experimental and the broken curve is calculated using a pseudopotential band structure model. Adapted from Yu and Cardona).

65

Figure 6.5: The band structure of Ge including spin-orbit interaction and showing the various direct transitions responsible for the structures that are observed in the imaginary part of the dielectric function 2 () and in the modulated reectivity. Although the band structure in this gure is for Ge, a similar notation is used to identify the various interband transitions in other group IV or IIIV compound semiconductors (see Table 6.1).

of states. Hence, modulation spectroscopy techniques emphasize critical points. There are a number of parameters that can be varied in these modulation spectroscopy experiments: electric eld wavelength stress light intensity temperature electro-reectance wavelength modulation piezoreectance photo-reectance thermo-reectance.

The various modulated reectivity experiments are complementary rather than yielding identical information. For example, certain structures in the reectance respond more sensitively to one type of modulation than to another. If we wish to look at structure associated with the L point (111 direction) transitions, then a stress along the (100) direction will not produce as important a symmetry change as application of stress along a (111) direction; with a stress along a (111) direction, the ellipsoid having its longitudinal axis along (111) will be aected one way while the other three ellipsoids will be aected in another way. However, stress along the (100) direction treats all ellipsoids in the same way. The reason why modulation spectroscopy emphasizes critical points can be seen by the following argument. For a direct interband transition, the optical absorption coecient has a frequency dependence abs () = C ( Eg ) h h . (6.37)

Therefore, a plot of abs () vs. exhibits a threshold [Fig. 6.6(a)], but no singularity in h 66

Figure 6.6: (a) The frequency dependence of the optical absorption coecient showing a threshold for interband transitions at the band gap. (b) The derivative of (a) which is measured in the modulated reectivity shows a sharp singularity associated with the threshold energy. the frequency plot. However, when we take the derivative of Eq. 6.37 abs () C C = ( Eg )1/2 h ( Eg )1/2 h 2 2 h (6.38)

a sharp structure is obtained in the modulated reectivity due to the singularity in the rst term of Eq. 6.37 at = Eg [see Fig. 6.6(b)]. If we modulate the light with any arbitrary h parameter x, then abs abs = , (6.39) x x and structure in the reectivity is expected as x is varied. Thus all modulation parameters can be expected to produce singularities in the optical absorption. For some variables such as stress, the modulated signal is sensitive to both the magnitude and the direction of the stress relative to the crystal axes. For thermomodulation, the spectrum is sensitive to the magnitude of the thermal pulses, but the response is independent of crystalline direction. Thermomodulation is, however, especially sensitive to transitions from and to the Fermi level. Thus, the various modulation techniques can be used in optical studies to obtain additional information about symmetry, which can then be used for more reliable identication of structure in the optical properties. The modulation technique specically emphasizes interband transitions associated with particular points in the Brillouin zone. The identication of where in the Brillouin zone a particular transition is occurring is one of the most important and dicult problems in optical studies of solids. It is often not the case that we have reliable band models available to us when we start to do optical studies. For this reason, symmetry is a very powerful tool for the study of optical properties. The high sensitivity of modulation spectroscopy provides valuable information about the band structure that would be dicult to obtain otherwise, and some examples are cited below. One example of the use of modulation spectroscopy is to determine the temperature dependence of the bandgap of a semiconductor, as shown in Fig. 6.7 for the direct point gap in Ge. This measurement takes advantage of the high resolution of modulation 67

Figure 6.7: Temperature dependence of the direct gap (E0 ) of Ge.

68

Figure 6.8: Dependence of the energies of the E0 , E0 + 0 , E1 , E1 + 1 , E0 , and E2 electroreectance peaks on x in the Ge1x Six alloy system at room temperature.

spectroscopy and is especially useful for measurements at elevated temperatures. Another example is the dependence of the various band separations identied in Fig. 6.5 as a function of alloy concentration x in Ge1x Six alloys (Fig. 6.8). Here again the high resolution of the modulation spectroscopy is utilized. A third example is the isotope dependence of the direct absorption edge of Ge as shown in Fig. 6.9. Modulation spectroscopy has also been applied to studying interband transitions in metals. For example, Fig. 6.10 shows modulated spectroscopy results from a gold surface taken with both the thermal modulation and piezoreectance techniques. The results show that transitions involving states at the Fermi level (either initial or nal states) are more sensitively seen using thermal modulation because small temperature variations aect the Fermi tail of the distribution function strongly. Thus, thermo-reectance measurements on the noble metals give a great deal of wellresolved structure, compared with electro-reectance and piezoreectance measurements as illustrated in Fig. 6.10. In this gure, we see that in gold the piezoreectance 69

Figure 6.9: Photo-modulated reectivity of Ge showing the E0 direct gap at k = 0 of single crystals of nearly isotopically pure 70 Ge, 74 Ge, and 76 Ge, at T = 6 K. Note the remarkable dependence of E0 on isotopic composition.

Figure 6.10: Thermo-reectance and normal incidence reectivity spectra of gold near liquid nitrogen temperature (from W.J. Scouler, Phys. Rev. Letters 18, 445 (1967)) together with the room temperature piezoreectance spectrum (M. Garfunkel, J.J. Tiemann, and W.E. Engeler, Phys. Rev. 148, 698 (1966)).

70

is much more sensitive than ordinary reectivity measurements near 4 eV, but the thermoreectance technique is most powerful for transitions made to states near the Fermi level.

6.4

Ellipsometry and Measurement of Optical Constants

Ellipsometry is a standard method for measuring the complex dielectric function or the complex optical constants N = n + ik of a material. Since two quantities are measured in an ellipsometry measurement, n and k can both be determined at a single frequency. The ellipsometry measurements are usually made over a range of frequencies, especially for frequencies well above the fundamental absorption edge where semiconductors become highly absorbing. At these higher frequencies very thin samples would be needed if the method of interference fringes were used to determine n, which is a very simple method for measuring the wavelength in a non-absorbing medium. One drawback of the ellipsometry technique is the high sensitivity of the technique to the quality and cleanliness of the surface. Ellipsometry is limited by precision considerations to measurements on samples with absorption coecients abs > 1 10 cm1 . Ellipsometers can be made to operate in the near infrared, visible and near ultraviolet frequency regimes, and data acquisition can be made fast enough to do real time monitoring of (). In the ellipsometry method the reected light with polarizations p (parallel) and s (perpendicular) to the plane of incidence [see Fig. 6.11(a)] is measured as a function of the angle of incidence and the frequency of the light. The corresponding reectances R s = |rs |2 and Rp = |rp |2 are related to the complex dielectric function () = 1 ()+i2 () = ( +ik)2 n by the Fresnel equations which can be derived from the boundary conditions on the elds at the interface between two surfaces with complex dielectric functions a and s as shown in Fig. 6.11(a). From the gure the complex reection coecients for polarizations s and p are Esr Na cos Ns cos t rs = = (6.40) Esi Na cos + Ns cos t and rp = in which Ns cos t = (s a sin2 )1/2 (6.42) and rs and rp are the respective reection coecients, s and Ns denote the complex dielectric function and complex index of refraction within the medium, while a and Na are the corresponding quantities outside the medium (which is usually vacuum or air). When linearly polarized light, that is neither s- nor p-polarized, is incident on a medium at an oblique angle of incidence, the reected light will be elliptically polarized. The ratio ( r ) of the complex reectivity coecients rp /rs r is then a complex variable which is measured experimentally in terms of its phase (or the phase shift relative to the linearly polarized incident light) and its magnitude, which is the ratio of the axes of the polarization ellipse of the reected light [see Fig. 6.11(a)]. These are the two measurements that are made in ellipsometry. The complex dielectric function of the medium s () = 1 () + i2 () can 71 s Na cos a Ns cos t Epr = Epi s Na cos + a Ns cos t (6.41)

Figure 6.11: (a) Electric eld vectors resolved into p and s components, for light incident (i), reected (r), and transmitted (t) at an interface between media of complex indices of refraction Na and Ns . The propagation vectors are labeled by ki , kr , and kt . (b) Schematic diagram of an ellipsometer, where P and S denote polarizations parallel and perpendicular to the plane of incidence, respectively. then be determined from the angle , the complex r , and the dielectric function a of the ambient environment using the relation s = a sin2 + a sin2 tan2 1 r 1 + r
2

(6.43)

and in a vacuum environment a = 1. The experimental set-up for ellipsometry measurements is shown in Fig. 6.11(b). Light from a tunable light source is passed through a monochromator to select a frequency and the light is then polarized linearly along direction E to yield the Is and Ip incident light intensities. After reection, the light is elliptically polarized along E(t) as a result of the phase shift Epr and Esr have each experienced. The compensator introduces a phase shift which cancels the + phase shift induced by the reection at the sample surface, so that the light becomes linearly polarized again as it enters the analyzer. If the light is polarized at an angle of /2 with respect to the analyzer setting, then no light reaches the detector. Thus at every angle of incidence and every frequency, (, ) is determined by Eq. 6.43 from measurement of the magnitude and phase of r . Another common method to determine the optical constants is by measurement of the normal incidence reectivity over a wide frequency range and using the KramersKronig analysis as discussed in 6.2 to determine the optical constants n() and k().

72

Chapter 7

Impurities and Excitons

7.1 Impurity Level Spectroscopy

Selected impurities are frequently introduced into semiconductors to make them ntype or ptype. The introduction of impurities into a crystal lattice not only shifts the Fermi level, but also results in a perturbation to the periodic potential, giving rise to bound impurity levels which often occur in the band gap of the semiconductor. Impurities and defects in semiconductors can be classied according to whether they result in a minor or major perturbation to the periodic potential. Any disturbance to the periodic potential results in energy levels diering from the energy levels of the perfect crystal. However, when these levels occur within the energy band gap of a semiconductor or of an insulator, they are most readily identied, and these are the levels which give rise to welldened optical spectra. Impurity levels are classied into two categories: 1. shallow levels, 2. deep levels, corresponding, respectively, to a minor or a major perturbation of the periodic potential. Impurities are also classied according to whether they give rise to electron carriers (donors) or hole carriers (acceptors). We will now discuss the optical spectra for impurities.

7.2

Shallow Impurity Levels

An example of a shallow impurity level in a semiconductor is a hydrogenic donor level in a semiconductor like Si, Ge or the III-V compounds. Let us briey review the origin of shallow donor levels in ntype semiconductors, where conduction is predominantly by electron carriers. Suppose we add donor impurities such as arsenic, which has 5 valence electrons, to germanium which has 4 valence electrons (see Part I, Fig. 4.1). Each germanium atom in the perfect crystal makes 4 bonds to its tetrahedrally placed neighbors. For the arsenic impurity in the germanium lattice, four of the valence electrons will participate in the tetrahedral bonding to the germanium neighbors, but the fth electron will be attracted back to the arsenic impurity site because the arsenic ion on the site has a positive charge. 73

Figure 7.1: Hydrogenic impurity levels in a semiconductor. This interaction is described by the Coulomb perturbation Hamiltonian, H (r) = e2 0 r (7.1)

where 0 is the static dielectric constant which is 16 for germanium and 12 for silicon. This Coulomb interaction is a screened Coulomb potential, screened by the static dielectric constant. The approximation of taking 0 to be independent of distance is however not valid for values of r comparable to lattice dimensions, as discussed below. In simple terms, H given by Eq. 7.1 is the same as in the hydrogen atom except that the charge is now e/ 0 and the mass which enters the kinetic energy is the eective mass hydrogen m . Since the levels in the hydrogen atom are given by the Bohr energy levels En
hydrogen En =

m 0 e4 2 2 n2 h

(7.2)

then the energy levels in the hydrogenic impurity problem are to a rst approximation given impurity by hydrogenic levels En
impurity En =

m e4 . 2 2 2 n2 h 0

(7.3)

The impurity levels are shown in Fig. 7.1 where the donor levels are seen to lie in the gap below the conduction band minimum. hydrogen impurity For the hydrogen atom E1 = 13.6eV, but for germanium E1 6 103 = 0.12m representing an average of the eective eV where we have used a value of m 0 mass over the entire conduction band pocket. From measurements such as the optical absorption we nd that the thermal energy gap (which is the energy dierence between the 74

L point lowest conduction band and the point highest valence band) is 0.66 eV at room temperature. But the donor level manifold is only 6 103 eV wide (ranging from the E1 level to the ionization limit) so that these impurity levels are very close to the bottom of the conduction band. Another quantity of interest in this connection is the orbital radius of the impurity. Unless the orbital radius is greater than a few lattice dimensions, it is not meaningful to use a dielectric constant independent of r in constructing the perturbation Hamiltonian, since the dielectric constant used there is conceptually meaningful only for a continuum. Therefore, it is of interest to calculate the hydrogen Bohr radius using the usual recipe for the hydrogen atom n2 2 h hydrogen rn = (7.4) m 0 e2 where = 1.054 1027 erg/sec, the mass of the free electron is m0 = 9.1 1028 g, and h the charge on the electron is e = 4.8 1010 esu. The value for the Bohr radius in the hydrogen hydrogen atom is r1 = 0.5 and for the screened hydrogenic states in the impurity A problem, we have n 2 2 0 h impurity rn = (7.5) m e2 which is larger than the hydrogen Bohr radius by a factor 0 m0 /m . Using typical numbers impurity for germanium we get the ground state radius r1 70. Thus, the electron travels A over many lattice sites in germanium and the dielectric constant approximation used in Eq. 7.1 is valid. From this discussion we see that only a very small energy is needed to ionize a bound donor electron into the conduction band, and because this binding energy is small, these impurity hydrogenic donor levels are called shallow impurity levels. Since rn a where a is the lattice constant, these electrons are well localized in momentum space according to the uncertainty principle. Shallow donor levels are associated with the kpoint where the conduction band minima occur. Thus the simple hydrogenic view of impurity levels in a semiconductor predicts that the impurity spectrum should only depend on the host material and on the charge dierence between the host and impurity. To see how well this model works, let us look at the experimental results summarized in Fig. 7.2. This picture is for silicon where the Bohr radius is 20. Here the agreement with the hydrogenic type model is good except for the A ground state, where the dielectric constant approximation is not as valid as for germanium. The actual calculation referred to in this picture is solved for the case where the eective mass tensor components are included in the calculation. Such a calculation cannot be done exactly for an ellipsoidal constant energy surface 2 2 /x2 ( 2 /y 2 + 2 /z 2 ) h + 2 ml mt e2 (r) = E(r) r (7.6)

for which ml = mt but can be done exactly for the hydrogen atom for which ml = mt . In practice Eq. 7.6 is solved using a variational principle. To take into account that for small r, we have (r) 1 and for large r, we have (r) 0 where 0 is the static dielectric constant, a spatial dependence for (r) thus needs to be assumed and this spatial dependence can be incorporated into the variational calculation. The inclusion of screening eects by the 75

Figure 7.2: Energy levels of donor states in silicon, experiment and theory. In recent work very good agreement is achieved between theory and experiment in transitions between shallow impurity states at low temperature (4 K) and low carrier concentrations (1014 /cm3 ).

introduction of a spatial dependence to the dielectric function (r), is called the central cell correction (see Part I 4.3). The impurity spectra are studied most directly by infrared absorption and transmission measurements. As an example of such spectra we see in Fig. 7.3 the absorption spectrum from phosphorus impurities in Si. Note that the photon energies used in these measurements are small so that far infrared frequencies must be employed. The ground state donor level is a 1s state and allowed transitions are made to a variety of pstates. Since the constant energy surface is ellipsoidal, the 2p levels break up into a 2p(ml = 0) level and a 2p(ml = 1) level which is doubly degenerate (see Fig. 7.3). Transitions from the 1s to both kinds of p levels occur, and account for the sharp features in the spectrum shown. The sensitivity of the spectra is somewhat improved using modulated spectroscopy techniques as shown in Fig. 7.4, where transitions to higher quantum states (n = 6) can be resolved and to higher angular momentum states (f levels where = 3), noting that electric dipole transitions always occur between states of opposite parity. For both Figs. 7.3 and 7.4, the initial state is the 1s impurity ground state. Analysis of such spectra gives the location of the donor impurity levels including the location of the ground state donor level, which is more dicult to calculate because of the central cell correction. In absorption measurements the impurity level transitions are observed as peaks. On the other hand, impurity spectra can also be taken using transmission techniques, where the impurity level transitions appear as minima in the transmission spectra. 76

Figure 7.3: Absorption spectrum of phosphorus donors in Si for a sample at liquid helium temperature containing 1.2 1014 cm3 phosphorus. The inset shows the 2p0 line on an expanded horizontal scale.

7.3

Departures from the Hydrogenic Model

While the simple hydrogenic model works well for the donor states in silicon and germanium, it does not work so well for the degenerate valence bands. In this case the spectra are sensitive to the impurity species from column III in the periodic table, because the valence band masses are heavier and the eective Bohr radius is therefore more comparable to the lattice constant. Calculations for the acceptor impurity levels are now suciently accurate so that good agreement between theory and experiment is obtained in recent work. It would be naive to assume that the simple hydrogenic model works for all kinds of impurity centers. If the eective Bohr radius is comparable with atomic separations, then clearly the Coulomb potential of the impurity center is not a small perturbation to the periodic potential seen by an electron. Specic cases where the impurity radius becomes small are materials with either (1) a large m or (2) a small 0 which imply a small interband coupling. When these conditions are put into Eqs. 7.3 and 7.5, we see that a small Bohr radius corresponds to a large En value. Thus deep impurity levels are not well described by simple eective mass theory. In order to make any progress at all with deep impurity level problems, we must consider the energy band structure throughout the Brillouin zone. When an electron is localized in real space, a suitable description in momentum space must include a large range of k values. When the impurity concentration becomes so large that the Bohr orbits for neighboring impurity sites start to overlap, the impurity levels start to broaden, and eventually impurity bands are formed. These impurity bands tend to be only half lled because of the Coulomb repulsion which inhibits placement of both a spin up and a spin down carrier in the same 77

Figure 7.4: Photo-thermal ionization spectrum of phosphorus-doped Si measured by modulation spectroscopy. The inset shows schematically the photo-thermal ionization process for a donor atom.

78

impurity level. When these impurity bands lie close to a conduction or valence band extremum, the coalescence of these impurity levels with band states produces band tailing. This band tailing results in a smearing out of the threshold of the fundamental absorption edge as observed in absorption measurements. When the impurity band broadening becomes suciently large that the electron wavefunction extends to adjacent sites, metallic conduction can occur. The onset of metallic conduction is called the Mott metalinsulator transition.

7.4

Vacancies, Color Centers and Interstitials

Closely related to the impurity problem is the vacancy problem. When a compound semiconductor crystallizes, the melt usually is slightly o stoichiometry with respect to the concentration of anions and cations. As an example suppose we prepare PbTe with Pb and Te concentrations in the melt that are stoichiometric to 0.01%. This means that there will be a slight excess of one of the atoms or slight deciency of the other. This deciency shows up in the crystal lattice as a vacancy or the absence of an atom. Such a vacancy represents a strong local perturbation of the crystal potential which again cannot be modeled in terms of hydrogenic impurity models. Such vacancy centers further tend to attract impurity atoms to form vacancy-impurity complexes. Such defects are dicult to model theoretically because their spatial localization requires participation of energy states throughout the Brillouin zone. Such centers generally give rise to energy states within the band gap of semiconductors and insulators. Such defect centers are often studied by optical techniques. One important defect in ionic insulating crystals is the F-center (Farbe or color center). We see in Fig. 7.5 that the negative ion vacancy acts like a +ve charge (absence of a ve charge). This eective +ve charge tends to bind an electron. The binding of an electron to a ve ion vacancy is called an F-center. These F-centers give rise to absorption bands in the visible. Without F-centers, these crystals are usually clear and transparent. The F-center absorption band causes crystals with defects to appear colored, having the color of the transmitted light. When the crystals are heated to high enough temperatures, these defects can be made to anneal and the absorption bands disappear. This is called bleaching. Many other color centers are found in ionic crystals. For example, we can have a hole bound to a +ve ion vacancy. We can also have a defect formed by a vacancy that is bound to any impurity atom, forming a vacancy-impurity complex, which can bind a charged carrier. Or we can have two adjacent vacancies (one +ve and the other ve) binding an electron and a hole. Further generalizations are also found. These defect centers are collectively called color centers and each color center has its characteristic absorption band. In Fig. 7.6 we see an example of absorption bands due to F-centers in several alkali halides. In all cases the absorption bands are very broad, in contrast with the sharp impurity lines which are observed in the far infrared for shallow impurity level transitions (see Figs. 7.3 and 7.4). In the case of the vacancy there is a considerable lattice distortion around each vacancy site as the neighboring atoms rearrange their electronic bonding arrangements. A few comments are in order about the classication of point defects. In Fig. 7.7 various types of point defects are shown. Figure 7.7(a) illustrates a perfect ionic crystal. Figure 7.7(b) shows an ionic crystal with vacancies. This particular collection of vacancies is of the Schottky type (equal numbers of positive and negative ion vacancies). Schottky point 79

Figure 7.5: Diagram of a negative ion vacancy or Fcenter in an ionic crystal.

Figure 7.6: Examples of Fcenter absorption lines in various alkali halide ionic crystals.

80

defects also include neutral vacancies. Finally Fig. 7.7(c) shows both vacancies and interstitials. When a + () ion vacancy is near a + () ion interstitials, this defect conguration is called a Frenkel-type point defect. We will now use simple statistical mechanical arguments to estimate the concentration of Schottky defects. Let Es be the energy required to take an atom from a lattice site inside the crystal to the surface. If n is the number of vacancies, the change in internal energy resulting from vacancy generation is U = nEs . Now the number of ways that n vacancy sites can be picked from N lattice sites is N !/[(N n)!n!], so that the formation of vacancies results in an increase in entropy of S = kB ln and a change in free energy F = U T S = nEs kB T ln N! . (N n)!n! (7.8) N! (N n)!n! (7.7)

Using Stirlings approximation for ln x! when x is large, we write ln x! x ln x x. = Equilibrium is achieved when (F /n) = 0, so that at equilibrium we have Es = k B T from which we write or N! N n ln = kB T ln n (N n)!n! n n Es = e kB T N n Es n exp N kB T (7.10) (7.9)

(7.11) (7.12)

since n N . The vacancy density is small because for Es 1eV , T 300K, (n/N ) 40 1017 . e In the case of vacancy pair formation in an ionic crystal (Schottky defect), the number of ways to make n separated pairs is [N !/(N n)!n!]2 so that for Schottky vacancy pair formation np Ep exp (7.13) N 2kB T where np is the pair vacancy density and Ep is the energy required for pair formation. These arguments can readily be extended to the formation of Frenkel defects and it can be shown that if N is the density of possible interstitial sites, then the density of occupied interstitial sites is 1 Ei ni (N N ) 2 exp (7.14) 2kB T where Ei is the energy to remove an atom from a lattice site to form an interstitial defect site. 81

Figure 7.7: Schematic of various possible arrangements of both vacancies and interstitials. (a) a perfect ionic crystal, (b) an ionic crystal with positive and negative ion vacancies and (c) an ionic crystal with positive and negative ion vacancies and interstitials.

82

7.5

Spectroscopy of Excitons

An exciton denotes a system of an electron and a hole bound together by their Coulomb interaction. When a photon excites an electron into the conduction band, a hole is left behind in the valence band; the electron, having a negative charge will be attracted to this hole and may (provided the energy is not too large) bind to the positively charged hole forming an exciton. Thus, the exciton binding energy is attractive and represents a lower energy state than the band states. Excitons are important in the optical spectra of bulk semiconductors especially at low temperature. Exciton levels are important for device applications since light emitting diodes and semiconductor lasers often involve excitons. However, because of the connement of carriers in quantum wells, exciton eects become much more important in the case of quantum wells, superlattices and devices based on these deliberately structured materials (see Part I 8.3.1 of class notes). The topic of excitons in low dimensional semiconductor systems is discussed in 7.7. We will now use the eective mass approximation to nd the exciton spectrum near an interband threshold and we assume that the exciton was created by a photon with energy slightly less than the direct energy gap Eg . The Schrdinger equation for the two-body o exciton packet wave function is written in the eective mass approximation as: p2 p2 e2 e + h = E 2m 2mh 0 |re rh | e (7.15)

thereby including the Coulomb binding energy of the electronhole pair. For simplicity, we assume that the dielectric constant 0 is independent of re and rh corresponding to a large spatial extension of the exciton in a semiconductor. We introduce new coordinates for the spatial separation r between the electron and hole r = r e rh and for the center of mass coordinate given by = m r e + m r h e h . m + m e h (7.17) (7.16)

We now separate the Schrdinger equation (Eq. 7.15) into an equation for the relative o motion of the electron and hole in the exciton wave packet F (r) and an equation of motion for the center of mass G() (re , rh ) = F (r)G(). (7.18) Thus Eq. 7.15 becomes p2 p2 e2 r ) + 2m r F (r)G() = EF (r)G() 2(m + mh 0 e r where the reduced eective mass is given by 1 1 1 = + mr me mh 83 (7.20) (7.19)

to obtain an eigenvalue equation for G() p2 G() = G() 2(m + m ) e h which is of the free particle form and has eigenvalues (K) = 2K 2 h 2(m + m ) e h (7.22) (7.21)

where K is the wave vector of the exciton. The free particle solutions for the center of mass problem of Eq. 7.22 show that the exciton can move freely as a unit through the crystal. The momentum of the center of mass for a direct band gap exciton is small because of the small amount of momentum imparted to the excitation by the light. We thus obtain the Schrdinger equation in the coordinate system of relative motion: o e2 p2 r F (r) = En F (r) 2m 0 r r (7.23)

where Eq. 7.23 has the functional form of the Schrdinger equation for a hydrogen atom o with eigenvalues En for quantum numbers n (where n = 1, 2, . . .) given by En = and the total energy for the exciton is then E = (K) + En . (7.25) m e4 r 2 2 2, 2 0 n h (7.24)

The energy levels of Eq. 7.24 look like the donor impurity spectrum but instead of the eective mass of the conduction band m we now have the reduced eective mass m given e r by Eq. 7.20. Since m has a smaller magnitude than m as seen in Eq. 7.20, we conclude r e that the exciton binding energy is less than the impurity ionization energy for a particular solid. An example of a spectrum showing exciton eects is presented in Fig. 7.8. The points are experimental and the solid curves are a t of the data points to 2 () for excitons given by 8|(v|p|c)|2 m3 1 r 2 () = ( n ), (7.26) n3 3 2 3 0 n=1 where the sum is over all the exciton bound states. From Table 7.1 we see that the binding energy for excitons for GaAs is 4.9 meV and the eective Bohr radius is 112 , which is A many lattice spacings. The various exciton lines contributing to the exciton absorption proles in Fig. 7.8 are unresolved even for the data shown for the lowest temperature of 21 K. A material for which the higher exciton energy levels (n = 2, 3, . . .) of the Rydberg series are resolved is Cu2 O as can be seen in Fig. 7.9. The observation of these higher states is attributed to the forbidden nature of the coupling of the valence and conduction bands, giving rise to a strict selection rule that only allows coupling to exciton states with p symmetry. Since n 2 is required to have p exciton states the n = 1 exciton is forbidden, and the exciton lines start at n = 2. The transitions are sharp, and well resolved exciton lines up to n = 5 can be identied in Fig. 7.9. 84

Table 7.1: Exciton binding energy (E1 ) and Bohr radius (r1 ) in some direct bandgap semiconductors with the zinc-blende structure (from Yu and Cardona). Semiconductor GaAs InP CdTe ZnTe ZnSe ZnS E1 (meV) 4.9 5.1 11 13 19.9 29 E1 (theory) (meV) 4.4 5.14 10.71 11.21 22.87 38.02 r1 () A 112 113 12.2 11.5 10.7 10.22

Figure 7.8: Excitonic absorption spectra in GaAs near its bandgap for several sample temperatures. The lines drawn through the 21, 90 and 294 K data points represent ts with theory.

85

Figure 7.9: Low temperature absorption spectrum of Cu2 O (plotted as the log of the transmission) showing the excitonic p series associated with its dipole-forbidden band edge in Cu2 O.

The exciton spectrum appears quite similar to the impurity spectrum of shallow impurity states. These two types of spectra are distinguished through their respective dependences on impurity concentration. Suppose that we start with a very pure sample (10 14 impurities/cm3 ) and then dope the sample lightly (to 1016 impurities/cm3 ). If the spectrum is due to donor impurity levels, the intensity of the lines would tend to increase and perhaps broaden somewhat. If, on the other hand, the spectrum is associated with an exciton, the spectrum would be attenuated because of screening eects associated with the charged impurities. Exciton states in 3D semiconductors are generally observed in very pure samples and at very low temperatures. The criterion is that the average Bohr orbit of the exciton is less than the distance between impurities. For the sake of this argument, consider an excitonic radius of 100. If an impurity ion is located within this eective Bohr radius, A then the electronhole Coulomb interaction is screened by the impurity ion and the sharp spectrum associated with the excitons will disappear. A carrier concentration of 10 16 /cm3 corresponds to nding an impurity ion within every 100 from some lattice point. Thus A the electronhole coupling can be screened out by a charged impurity concentration as low as 1016 /cm3 . Low temperatures are needed to yield an energy separation of the exciton levels that is larger than kB T . Increasing the temperature shifts the absorption edge and broadens the exciton line in GaAs. At a temperature of 20 K we have k B T 1.7 meV which is nearly as large as the exciton binding energy of 4.9 meV found in Table 7.1, explaining why no well resolved exciton spectrum for higher quantum states is observed. For the case of Cu2 O, the exciton binding energy of the ground state (1s), were it to exist, would be 97 meV, neglecting central cell corrections. The large exciton binding energy in Cu 2 O also 86

helps with the resolution of the higher quantum exciton states.

7.6

Classication of Excitons

The exciton model discussed above is appropriate for a free exciton and a direct exciton. For the direct exciton, the initial excitation is accomplished in a k-conserving process without the intervention of phonons. In materials like silicon and germanium, the thermal band gap corresponds to an indirect energy gap. For these materials, the exciton is formed by an indirect phonon-assisted process and the exciton is consequently called an indirect exciton. Indirect excitons can be formed either with the emission or absorption of a phonon. Since excitons are more important at low temperatures, the emission process is much more likely than the absorption process. Because of the large dierence in crystal momentum k between the valence band extremum and the lowest conduction band minimum in these h indirect gap semiconductors, the exciton may acquire a large center of mass momentum corresponding to the momentum of the absorbed or emitted phonon q. For the indirect h exciton, a large range of crystal momentum k values are possible and hence the exciton h levels spread out into bands as shown in the lower dashed rectangle of Fig. 7.10. This portion of the gure also appears in more detail in the upper left-hand corner. In Fig. 7.10 we also show in the upper right-hand corner the direct exciton associated with the point conduction band for various temperatures. The shift in the absorption edge is associated with the decrease in band gap with increasing temperature. In Fig. 7.10, the individual exciton lines are not resolved a lower temperature would be needed for that. Addition of impurities to suppress the exciton formation does not help with the identication of bandgaps in semiconductors since the presence of impurities broadens the band edges. It is for this reason that energy gaps are best found from optical data in the presence of a magnetic eld, to be discussed in connection with magnetism (Part III of this course). For small distances from the impurity site or for small electron-hole separations, the eective mass approximation must be modied to consider central cell corrections explicitly. For example, central cell corrections are very important in Cu2 O so that the binding energy attributed to the 1s state is 133 meV, whereas the binding energy deduced from the Rydberg series shown in Fig. 7.9 indicates a binding energy of 97 meV. The kinds of excitons we have been considering above are called free excitons. In contrast to these, are excitations called bound excitons. It is often the case that an electron and hole may achieve a lower energy state by locating themselves near some impurity site, in which case the exciton is called a bound exciton and has a larger binding energy. Bound excitons are observed in typical semi-conducting materials, along with free excitons. Another category of excitons that occurs in semiconductors is the molecular excitons. Just as the energy of two hydrogen atoms decreases in forming molecular hydrogen H 2 , the energy of two free (or bound) excitons may decrease on binding to form a molecular state. More complicated exciton complexes can be contemplated and some of these have been observed experimentally. As the exciton density increases, further interaction occurs and eventually a quantum uid called the electron-hole drop is formed. Unlike other uids, both the negatively and positively charged particles in the electron-hole uid have light masses. A high electron-hole density can be achieved in indirect band gap semiconductors such as silicon and germanium because of the long lifetimes of the electron-hole excitations in these materials. In treating 87

Figure 7.10: Plot of the square root of the absorption coecient vs. for Ge for various h temperatures showing the eect of the excitons. Features associated with both indirect and direct excitons are found.

88

Figure 7.11: Recombination radiation (or photoluminescence spectrum) of free electrons (FE) and of electron-hole drops (EHD) in Ge at 3.04 K. The Fermi energy in the electronhole drop is F and the cohesive energy of the electron-hole drop with respect to a free exciton is s = 1.8 meV. The critical concentration and temperature for forming an electronhole drop in Ge are respectively 2.6 1017 /cm3 and 6.7 K. the electron-hole drops theoretically, the electrons and holes are regarded as free particles moving in an eective potential due to the other electrons and holes. Because of the Pauli exclusion principle, no two electrons (or holes) can have the same set of quantum numbers. For this reason like particles tend to repel each other spatially, but unlike particles do not experience this repulsion. Thus electron-hole pairs are formed and these pairs can be bound to each other to form an electron-hole drop. These electron-hole drops have been studied in the emission or luminescence spectra (see Chapter 8). Results for the luminescence spectra of Ge and Si at very low temperatures (T 2K) are shown in Fig. 7.11. Luminescence spectra for germanium provide experimental evidence for electron-hole drops for electronhole concentrations exceeding 1017 /cm3 . In insulators (as for example alkali halides), excitons are particularly important, but here they tend to be well localized in space because the eective masses of any carriers that are well localized tend to be large. These localized excitons, called Frenkel excitons, are much more strongly bound and must be considered on the basis of a much more complicated theory. It is only for the excitons which extend over many lattice sites, the Wannier excitons, that eective mass theory can be used. And even here many-body eects must be considered to solve the problem with any degree of accuracy already an electron bound to a hole is a twobody problem so that oneelectron eective mass theory is generally not completely valid. In studying the optical absorption of the direct gap, the presence of excitons complicates the determination of the direct energy gap, particularly in alkali halides where the exciton binding energy is large. Referring to Fig. 7.12(a), both -point and L-point excitons are identied in the alkali halide ionic crystal KBr. The correspondence of the optical structure with the E(k) diagram is shown by comparison of Figs. 7.12(a) and (b). Here it is seen that the Frenkel exciton lines dominate the spectrum at the absorption edge and we also see huge shifts in energy between the exciton lines and the direct absorption edge. These 89

Figure 7.12: (a) A spectrum of the optical density of KBr showing Frenkel excitons. The optical density is dened as log(1/T ) where T is the optical transmission. (b) The energy bands of KBr, as inferred from tight-binding calculations of the valence bands and the assignments of interband edges in optical experiments. We note that the spectrum is dominated by exciton eects and that direct band edge contributions are much less important and the binding energy is on the order of an electron volt.

90

gures show the dominance of strongly bound, localized Frenkel excitons in the spectra of alkali halides. Excitons involve the presence of an electron-hole pair. If instead, an electron is introduced into the conduction band of an ionic crystal, a charge rearrangement occurs. This charge rearrangement partially screens the electron, thereby reducing its eective charge. When an electric eld is now applied and the charge starts to move through the crystal, it moves together with this lattice polarization. The electron together with its lattice polarization is called a polaron. While excitons are important in describing the optical properties of ionic or partly ionic materials, polarons are important in describing the transport properties of such materials. The presence of polarons leads to thermally activated mobilities, which says that a potential barrier must be overcome to move an electron together with its lattice polarization through the crystal. The presence of polaron eects also results in an enhancement in the eective mass of the electron. Just as one categorizes excitons as weakly bound (Wannier) or strongly bound (Frenkel), a polaron may behave as a free particle with a relatively weak enhancement of the eective mass (a large polaron) or may be in a bound state with a nite excitation energy (a small polaron), depending on the strength of the electron-phonon coupling. Large polarons are typically seen in weakly ionic semiconductors, and small polarons in strongly ionic, large-gap materials. Direct evidence for large polarons in semiconductors has come from optical experiments in a magnetic eld in the region where the cyclotron frequency c is close to the optical phonon frequency LO .

7.7

Optical Transitions in Quantum Well Structures

Optical studies are extremely important in the study of quantum wells and superlattices. For example, the most direct evidence for bound states in quantum wells comes from optical absorption measurements. To illustrate such optical experiments consider a GaAs quantum well bounded on either side by the wider gap semiconductor Alx Ga1x As. Because of the excellent lattice matching between GaAs and Alx Ga1x As, these materials have provided the prototype semiconductor superlattice for study of the 2D electron gas. The threshold for absorption is now no longer the band gap of bulk GaAs but rather the energy separation between the highest lying bound state of the valence band and the lowest bound state of the conduction band. Since the valence band of GaAs is degenerate at k = 0 and consists of light and heavy holes, there will be two n = 1 levels in the valence band. Since En 1/m for the quantum well n = 1 bound state level, the heavy hole subband extremum will be closer in energy to the band edge than that of the light hole as shown on the left side of Fig. 7.13. Also the density of states for the heavy hole subband will be greater than that of the light hole subband by a factor of 2mhh /mlh . The optical absorption will thus show two peaks near the optical threshold as illustrated in the diagram in Fig. 7.13. These data are for a sample with a quantum well width of 50, which is small enough to contain A a single bound state (n = 1). Since the optical absorption from a single quantum well is very weak, the experiment is usually performed in superlattice structures containing a periodic array of many equivalent quantum wells. In forming the superlattice structure, it is important that the barrier between the quantum wells is not too small in extent, because for small spatial separations between quantum wells and low band osets at the interfaces, the eigenfunctions in adjacent wells become coupled and we no longer have a 2D electron gas in the quantum well. 91

Figure 7.13: Optical excitations in a quantum well where the valence band has light and heavy holes (as in GaAs). The optical density is dened as log(1/T ) where T is the transmission.

Figure 7.14: Frequency dependence of the absorption for GaAs/Al0.2 Ga0.8 As heterostructure superlattices of dierent thicknesses at optical frequencies.

92

For wider quantum wells containing several bound states (see Figs. 7.14 and 7.15), a series of absorption peaks are found for the various bound states, and the interband transitions follow the selection rule n = 0. This selection rule follows because of the orthogonality of wave functions for dierent states n and n . Thus to get a large n matrix element for coupling valence and conduction band states, n and n must be equal. As the width of the quantum well increases, the spectral features associated with transitions to the bound states become smaller in intensity and more closely spaced and eventually cannot be resolved. For the thickest lms, the quantum levels are too close to each other to be resolved and only the bulk exciton peak is seen. For the 210 quantum well (see A Fig. 7.14), transitions for all 4 bound states within the quantum well are observed. In A addition, excitonic behavior is observed on the n = 1 peak. For the 140 well, the transitions are broader, and eects due to the light and heavy hole levels can be seen through the distorted lineshape (see Fig. 7.14. To observe transitions to higher bound states, the spectra in Fig. 7.15 are taken for a quantum well width of 316 , for which A transitions up to (6,6) are resolved. For such wide quantum wells, the contributions from the light holes are only seen clearly when a transition for a light hole state is not close to a heavy hole transition because of the lower density of states for the light holes (see Fig. 7.15). Exciton eects are signicantly more pronounced in quantum well structures than in bulk semiconductors, as can be understood from the following considerations. When the width d1 of the quantum well is less than the diameter of the exciton Bohr orbit, the electronhole separation will be limited by the quantum well width rather than by the larger Bohr radius, thereby signicantly increasing the Coulomb binding energy and the intensity of the exciton peaks. Thus small quantum well widths enhance exciton eects. Normally sharp exciton peaks in bulk GaAs are observed only at low temperature (T 77K); but in quantum well structures, excitons can be observed at room temperature, as shown in Fig. 7.14, which should be compared with Fig. 7.8 for 3D bulk GaAs. The reason why the exciton line intensities are so much stronger in the quantum well structures is due to the reduction in the radius of the eective real space Bohr orbits, thereby allowing more k bandstates to contribute to the optical transition. This argument is analogous to arguments made to explain why the exciton intensities for the alkali halides are huge [see Fig. 7.12(a)]. In the alkali halides the excitons have very small real space Bohr orbits so that large regions of k space can contribute to the exciton excitation. In the case of the quantum well structures, two exciton peaks are observed because the bound states for heavy and light holes have dierent energies, in contrast to the case of bulk GaAs where the j = 3/2 valence band states are degenerate at k = 0. This property was already noted in connection with Fig. 7.13 for the bound state energies. Because of the large phonon density available at room temperature, the ionization time for excitons is only 3 1013 sec. Also the presence of the electronhole plasma strongly modies the optical constants, so that the optical constants are strongly dependent on the light intensity, thereby giving rise to nonlinear eects that are not easily observed in 3D semiconductors. Because of the small binding energy of these exciton states, modest electric elds have a relatively large eect on the photon energy of the exciton peaks and on the optical constants. Application of an electric eld perpendicular to the layers of the superlattice connes the electron and hole wave functions at opposite ends of the quantum well, as shown in Fig. 7.16. Because of this spatial separation, the excitons become relatively long lived and now recombine on a time scale of 109 sec. Also because of the quantum connement, it is 93

Figure 7.15: Transmission spectrum of a GaAs/AlGaAs multi-quantum well (well width = 316 ) measured as a function of photon energy at low temperature (right panel). The A peaks labeled (n, n) have been identied with optical transitions from the nth heavy hole (hh) and light hole (lh) subbands to the nth conduction subband as shown by arrows in the band diagram in the left panel. The values of the band osets used in the analysis are given in the diagram, but these are not the most recent values.

94

Figure 7.16: Excitonic wave functions in a GaAs quantum well without (left) and with (right) an applied electric eld. Because of the triangular potentials that are created by the electric eld in the zdirection, the quantum well retains the electron and hole in a bound state at electric elds much higher than would be possible in the bulk classical ionization eld.

Figure 7.17: The absorption spectra in GaAs/Ga1x Alx As heterostructures for various values of applied electric eld illustrating the large changes in optical properties produced by the quantum conned Stark shift. The electric elds normal to the layer planes are: (a) 104 V/cm, (b) 5 104 V/cm, and (c) 7.5 104 V/cm.

95

possible to apply much higher (50 times) electric elds than is possible for an ionization eld in a bulk semiconductor, thereby producing very large Stark red shifts of the exciton peaks, as shown in Fig. 7.17. This electric eld eect on the exciton levels is called the quantum conned Stark eect. This eect is not observed in bulk semiconductors. The large electric eldinduced change in the optical absorption that is seen in Fig. 7.17 has been exploited for device applications. The following mechanism is proposed to explain the quantum conned Stark eect when the electric eld is applied perpendicular to the layers. This electric eld pulls the electrons and holes towards opposite sides of the layers as shown in Fig. 7.16 resulting in an overall net reduction in the attractive energy of the electronhole pair and a corresponding Stark (electric eld induced) shift in the exciton absorption. Two separate reasons explain the strong exciton peaks in quantum well structures. Firstly the walls of the quantum wells impede the electron and hole from tunneling out of the wells. Secondly, because the wells are narrow (e.g., 100) compared to the threedimensional (3D) exciton size (e.g., 200 ), A A the electronhole interaction, although slightly weakened by the separation of electron and hole, is still strong, and well dened excitonic states can still exist. Thus exciton resonances can remain to much higher elds than would be possible in the absence of this connement, and large absorption shifts can be seen without excessive broadening.

96

Chapter 8

Luminescence and Photoconductivity

8.1 Classication of Luminescence Processes

Luminescence denotes the emission of radiation by a solid in excess of the amount emitted in thermal equilibrium. Since luminescence is basically a non-equilibrium phenomena, it requires excitation by light, electron beams, current injection, etc., which generally act to create excess electrons, holes, or both. The eects of electronhole recombination give rise to recombination radiation or luminescence. One classication of luminescent processes is based on the source of the excitation energy. The most important excitation sources are 1. photoluminescence by optical radiation, 2. electroluminescence by electric elds or currents, 3. cathodoluminescence by electron beams (or cathode rays), 4. radioluminescence by other energetic particles or high energy radiation. A second classication of luminescent processes pertains to the time that the light is emitted < relative to the initial excitation. If the emission is fast ( 108 sec is a typical lifetime for an atomic excited state), then the process is uorescent. The emission from most photoconductors is of the uorescent variety. For some materials, the emission process is slow and can last for minutes or hours. These materials are phosphorescent and are called phosphors. Let us rst consider luminescent processes of the uorescent type with fast emission times. The electronic transitions which follow the excitation and which result in luminescent emission are generally the same for the various types of excitations. Figure 8.1 shows a schematic diagram of the basic transitions in a semiconductor. These may be classied as follows: 1. Transitions involving chemical impurities or physical defects (such as lattice vacancies): 97

Figure 8.1: Basic transitions in a semiconductor for the luminescent process. After H.F. Ivey, IEEE J.Q.E. 2, 713 (1966). = electrons: = holes.

(a) conduction band to acceptor. (b) donor to valence band. (c) donor to acceptor (pair emission). 2. Interband transitions: (a) intrinsic or edge emission corresponding very closely in energy to the band gap, though phonons and/or excitons may also be involved (b) higher energy emission involving energetic or hot carriers, sometimes related to avalanche emission, where hot carriers refers to highly energetic carriers well above thermal equilibrium levels. 3. Intraband transitions involving hot carriers, sometimes called deceleration emission. It should be pointed out that the various transitions mentioned above do not occur in the same material or under the same conditions. Nor are all electronic transitions radiative. Phonon emission provides a non-radiative mechanism for the relaxation of an excited state in a solid to the lowest equilibrium ground state. An ecient luminescent material is one in which radiative transitions predominate over nonradiative ones. When electronhole pairs are generated by external excitations, radiative transitions resulting from the holeelectron recombination may occur. The radiative transitions in which the sum of electron and photon wavevectors is conserved are called direct transitions as opposed to indirect transitions which involve scattering agents such as phonons.

8.2

Emission and Absorption

For a given material the emission progagility will depend on the photon energy and on the temperature. The emission rate Rvc () for the transition from the conduction band (c) to 98

Figure 8.2: Luminescence emission spectrum in an n-type InSb crystal with an electron concentration of 5 1013 cm3 . The peak at 0.234 eV is due to interband recombinative emission. The peak at 0.212 eV (multiplied by 200) is due to phonon-assisted bandto-band transitions. (A. Mooradian and H.Y. Fan, Phys. Rev. 148, 873 (1966).)

the valence band (v) is related to the absorption rate Pvc () by the relation Rcv () = Pvc ()() where () is the Planck distribution at temperature T () = and the absorption rate is given by Pvc () = ()c nr (8.3) 2 n3 2 r c3 [exp( /kB T ) 1] h (8.2) (8.1)

so that the frequency and temperature dependence of the emission rate is given by Rcv () = 2 2 n2 r . c2 [exp( /kB T ) 1] h (8.4)

Basically, Rcv () shows high emission at frequencies where the absorption is large, so that emission spectroscopy can be used as a technique to study various aspects of the band structure. For example, Fig. 8.2 shows luminescence from the conduction band of InSb to the valence band (band-to-band process) at 0.234 eV from the conduction band to an acceptor impurity level at 0.228 eV and luminescence that is phonon assisted at 0.212 eV involving phonon absorption. For intrinsic or band-to-band transitions, the peak intensity occurs near the energy gap and the width of the spectral line (at half value of peak intensity) is proportional to 99

Figure 8.3: Direct and indirect intrinsic radiation recombination in Ge. The 70K spectrum is experimental and is in the energy range appropriate for indirect transitions assisted by longitudinal acoustic (LA) phonons. The circles are calculated from absorption data for both types of transitions. The free carrier densities at the direct and indirect conduction band minima and at the valence band maximum are denoted by n0 , n, and p, respectively.

100

the thermal broadening energy kB T . For extrinsic transitions, the peak emission intensity occurs near the transition energy, but the broadening is greater than for the intrinsic bandto-band emission shown in Fig. 8.3. The general problem of luminescence is not only to determine the luminescent mechanisms and the emission spectra, as discussed above, but also to determine the luminescent eciency. For a given input excitation energy, the radiative recombination process is in direct competition with the non-radiative processes. Luminescent eciency is dened as the ratio of the energy associated with the radiative process to the total input energy. Among the fastest emission luminescent processes, electroluminescence, or excitation by an electric eld or current, has been one of the most widely utilized for device applications. Electroluminescence is excited in a variety of ways including intrinsic, injection, avalanche, and tunneling processes. 1. Intrinsic process. When a powder of a semiconductor, (e.g., ZnS) is embedded in a dielectric (plastic or glass), and exposed to an alternating electric eld, usually at audio frequencies, electroluminescence may occur. Generally the eciency is low ( 1%) and such materials are used primarily in display devices. The mechanism is mainly due to impact ionization by accelerated electrons and/or eld emission of electrons from trapping centers. 2. Injection. Under forward-bias conditions, the injection of minority carriers in a p n junction can give rise to radiative recombination. The energy level band diagram for a Cd-doped GaP p n junction is shown in the Fig. 8.4. Several dierent transitions for electron-hole recombination are indicated. The relative intensity of the red and the green bands can be varied by varying the impurity concentrations. The brightness of the red-light emission from the GaP p n junction at room temperature is suciently high to merit electro-luminescent applications, as an example, GaP light-emittingdiodes (LEDs) for numeric displays in pocket electronic calculators. At the present time, the highest electro-luminescent eciency has been obtained experimentally in forward-biased GaAs diodes. This is expected because (1) the forwardbias injection is a very ecient method since electric energy can be converted directly into photons; (2) GaAs is a direct-gap semiconductor. Thus, the radiative recombination process is a rst-order transitions process (no phonon involved); and (3) GaAs has the most advanced materials technology of all the direct-gap semiconductors. 3. Avalanche. When a pn junction or a metal semiconductor contact is reverse-biased into avalanche breakdown, the electron-hole pairs generated by impact ionization may result in emission of either interband (avalanche emission) or intraband (deceleration emission) transitions, respectively. 4. Tunneling. Electroluminescence can result from tunneling into forward-biased and reverse-biased junctions. In addition, light emission can occur in reverse-biased metalsemi-conductor contacts. Fast emission luminescence also is of importance to semiconducting lasers. Luminescence is an incoherence emission process in contrast with laser action which involves the coherent emission of radiation in executing a radiative transition. The coherence is usually 101

Figure 8.4: Energy level diagram for a Cd-doped GaP p n junction where Cd-O denotes an cadmium-oxygen complex. Transitions between the exciton level of the Cd-O complex to the acceptor level of Cd give rise to red light emission. Transitions between the donor level (S) and acceptor level (Cd) give rise to the green light emission. (b) Measured emission spectrum from a GaP diode in which the color associated with the various luminescent peaks are shown. (After M. Gershenzon, Bell Sys. Tech. J. 45, 1599 (1966).)

102

Figure 8.5: Schematic of the phosphorescence process of the thallium+ activator in KCl. The emission is downshifted from the absorption. This is an illustration of the FranckCondon Principle.

enhanced by polishing the sample faces to form an optical cavity. Examples of solid state lasers are the ruby laser and the direct gap semiconductor lasers. Optical and electrical pumping are the most common methods of exciting laser action in solid state lasers. Finally, we conclude the discussion of electroluminescence in semiconductors with a short discussion of slow emission luminescence, i.e. phosphorescence. Phosphorescent materials exhibit afterglow eects and are consequently important in various optical display devices. These phosphors often do not exhibit large photoconductivities. That is to say, although the electrons that were produced survive for a long time, they are bound to particular defect centers and do not readily carry charge through the crystal. In Fig. 8.5 we show an example of how a phosphor works in an alkali halide such as KCl with a small amount of Tl impurities. The thallium defects act as recombination centers. If these recombination centers are very ecient at producing recombination radiation they are called activators; Tl in KCl acts as an activator. In this system, the excitation occurs at higher energy than the emission. The Franck-Condon principle states that the atoms in the solid do not change their internuclear separations during an electronic transition. We now explain how emitted light is downshifted in frequency from the exciting light. The Tl + ion in the ground or unexcited state occupies some conguration close to the symmetric center of a K+ ion which the Tl+ ion might be replacing. When excited, the Tl+ ion nds 103

Figure 8.6: Schematic diagram of the experimental arrangement for measuring the photoconductivity.

a lower energy state in a lower symmetry position near one of the Cl ions as shown on the top of Fig. 8.5. The absorption is made from the ground state energy (point A in Fig. 8.5) to an excited state with the same conguration. Phonon interactions then will bring the electron to the equilibrium position C. Achievement of equilibrium (B C) will take a longer time than the electronic transitions (A B). Emission from C D again occurs in accordance with the Franck-Condon principle and the readjustment to the equilibrium conguration A proceeds by phonon processes.

8.3

Photoconductivity

Photoconductivity is observed when light is incident on a poorly conducting material, (e.g., an insulator or semiconductor), and the photon energy is suciently high to excite an electron from an occupied valence state to an unoccupied conduction state. In such interband transitions both the electron and hole will contribute to the electrical conductivity if a voltage is applied across the sample as shown in the schematic experimental arrangement in Fig. 8.6. Since the threshold for photoconduction occurs at = Eg , measurement of the h photoconductivity can be used to determine the band gap for non-conducting materials. Photoconductivity is often the concept used for the design of practical optical detectors. The photoconduction process increases the electrical conductivity due to the increase in the density of electrons (n) and (p) resulting from photo-excitation: nn + pp = nn + pp (8.5)

in which n + p are respectively, the electron and hole mobilities. Since the carriers are generated in pairs in the photo-excitation process n = p. In preparing materials for application as photoconductors, it is desirable to have a high mobility material with a low intrinsic carrier concentration, and long electronhole recombination times to maximize the photo-excited carrier density concentration. CdS is an example of a good photoconductive 104

material. In CdS, it is possible to change the conductivity by 10 orders of magnitude through carrier generation by light. These large changes in electrical conductivity can be utilized in a variety of device applications such as: light meters, photo-detectors, electric eye control applications, optically activated switches, and information storage. To measure photo-currents, photo-excited carriers are collected at the external electrodes. In the steady state, free carriers are continually created by the incident light. At the same time, they annihilate each other through electron-hole recombination. To produce a large photocurrent, it is desirable to have a long free carrier lifetime or a slow recombination time. If G is the rate of generation of electrons per unit volume due to photo-excitation, then the photo-excited electron density in the steady state will be given by n = G . (8.6) The generation rate G will in turn be proportional to the photon ux incident on the photoconductor. Whereas slow recombination rates are essential to the operation of photoconductors, rapid recombination rates are necessary for luminescent materials. In the recombination process, an electron and hole annihilate each other, emitting a photon in a radiative process. In real materials, the recombination process tends to be accelerated by certain defect sites. When such defects tend to be present in relatively greater concentrations at the surface, the process is called surface recombination. In bulk, the density of recombination centers can be made low for a very pure and good crystal. A typical concentration in a high quality Si crystal would be 1012 cm3 . Photo-excited carriers can also be eliminated from the conduction process by electron and hole traps. These traps dier from recombination centers insofar as traps preferentially eliminate a single type of carrier. In practice, hole traps seem to be more common than electron traps. For example, in the silver halides which are important in the photographic process, the hole is trapped almost as soon as it is produced and photoconduction occurs through the electrons. Electron and neutron irradiation produce both recombination centers and traps in photoconducting materials. Thus, special precautions must be exercised in using photo-detectors in a high radiation environment or on satellites which are expected to pass through regions of high radiation uences. Trapped electrons can be released by thermal or optical excitation. For example, consider a p-type sample of Ge which has been doped with Mn, Ni, Co, Fe. At low temperatures EF will be near the top of the valence band and the acceptor impurity states will have very few electrons in them. Photons energetic enough to take an electron from the valence band to these impurity levels will result in hole carriers in the valence bands. The deep acceptor levels for these impurities are above the top of the valence band by 0.16 eV for Mn, 0.22eV for Ni, 0.25 eV for Co and 0.35 eV for Fe. The thresholds observed for photoconductivity in these p-type Ge samples are shown in Fig. 8.7 and the experimental results are in good agreement with this interpretation. The large increase in photoconductivity at 0.7eV corresponds to an interband transition and the threshold for this process is independent of the impurity species. The excess carrier lifetime can be measured by using light pulses and observing the decay in the photocurrent through measurement of the voltage across a calibrated load resistor R in the external circuit as shown in Fig. 8.8. Using a light chopper, light pulses can be generated as indicated in Fig. 8.9. For each light pulse, the carrier density will build 105

Figure 8.7: Photoconductance spectrum in bulk Ge with various dopants.

Figure 8.8: Schematic of a circuit used to measure the excess carrier lifetime through decay in the photocurrent.

Figure 8.9: Schematic experimental time dependence of light pulses and of the corresponding photoconductivity signal.

106

Figure 8.10: Experimental dependence of the photocurrent on light irradiation for CdS. A linear response is observed for low light levels.

up and then decay exponentially with a characteristic time equal to the lifetime of the excess carriers. In the interpretation of these experiments corrections must be made for surface recombination. To study a given material, the pulse repetition rate is adjusted to match approximately the excess carriers decay lifetime. For long lifetimes ( 10 3 sec), a mechanical chopper arrangement is appropriate. On the other hand, for short lifetimes a spark source can be used to give light pulse of 108 sec duration. For extremely short lifetimes, lasers with pulses well below 1010 seconds are available. To get an idea of the magnitude involved in the photoconduction process, we show in Fig. 8.10 some data for CdS, a common photoconductor. This plot of photoconductive response versus illumination level shows that the photocurrent is almost a linear function of the illumination intensity for low intensities but is non-linear at high illumination levels. The dark current refers to the background current that ows in the absence of incident light. Thus, the Fig. 8.10 shows that an incident power as small as 5 10 8 watts results in a photocurrent 50 times greater than the dark current.

107

Chapter 10

Optical Study of Lattice Vibrations

References:
Kittel, ISSP 6th edition, Ch 10. Ashcroft and Mermin, Ch 27. Yu and Cardona, Fundamentals of Semiconductors, pp. 251-258

10.1
10.1.1

Lattice Vibrations in Semiconductors

General Considerations

The lattice vibrations in semiconductors are described in terms of 3N branches for the phonon dispersion relations where N is the number of atoms per primitive unit cell. Three of these branches are the acoustic branches, and the remaining 3N 3 are the optical branches. The optical lattice modes at q = 0 are sensitively studied by infrared spectroscopy (optical reectivity or transmission) for odd parity modes, including those for which the normal mode vibrations involve a dipole moment. Raman spectroscopy provides a complementary tool to infrared spectroscopy, insofar as Raman spectroscopy is sensitive to even parity modes. Since the group IV semiconductors have inversion symmetry, the optical phonon branch is Raman active but is not seen in infrared spectroscopy. The IIIV compound semiconductors, however, do not have inversion symmetry, so that the optical modes for semiconductors such as GaAs are both infrared-active and Raman-active. A schematic optical absorption curve to a semiconductor is shown in Fig. 10.1. Since the wavevector for light is very much smaller than the Brillouin zone dimensions, conservation of momentum requires the wave vector for the phonon q phonon that is created or absorbed to be much smaller than Brillouin zone dimensions, so that the wave vectors for phonons that are observed in rst order infrared or Raman processes are close to q = 0. Since thermal neutrons can have a wide range of momentum values, neutron spectroscopy using thermal neutrons as a probe allows exploration of the phonon branches over a wide range of qphonon . Since heat in a semiconductor is dominantly carried by the acoustic phonons, information about the acoustic phonons is also provided by thermal conductivity studies. 108

Figure 10.1: Hypothetical absorption spectrum for a typical III-V semiconductor as a function of phonon energy. We now review the interaction of the electromagnetic eld with an oscillating dipole due to a lattice vibration. Crystals composed of two dierent atomic species (like NaCl) can have vibrating ions at nite temperatures. When these ions are vibrating in an optic mode , a vibrating dipole is created and this dipole can interact with the electromagnetic eld. In discussing this interaction, we wish to focus attention on the following points which are discussed more fully in the text below: 1. The existence of two characteristic frequencies for the vibrations in a solid in the presence of light: t = transverse optical phonon frequency (TO) = longitudinal optical phonon frequency (LO) The description of the LO and TO phonons is provided by the polariton model which accounts for the interaction between light and phonon excitations. Because of the very small wavevector of the incident photons, the phonons which are optically excited will also have very small wavevectors. Therefore, t and are taken as the phonon frequencies at q = 0 for the TO and LO phonon dispersion curves. 2. These two frequencies are observable experimentally either through an infrared absorption, transmission, or reection experiment (infrared activity) or through a scattering experiment (Raman activity). A transparent dielectric becomes lossy as 109

increases above t . The transverse optical phonon frequency t corresponds to a resonance in the dielectric function const (10.1) () = + 2 t 2 where is the high frequency dielectric function (appropriate to electronic excitation processes) and a resonance in () occurs at the TO phonon frequency = t . The strong frequency dependence of the dielectric function (large dispersion) near t is exploited in designing prisms for monochromators. The frequency t is also called the reststrahl frequency. 3. The frequency is the frequency at which the real part of the dielectric function vanishes 1 ( ) = 0. It will be shown below that is the longitudinal optical phonon frequency corresponding to q = 0 (zero wave vector). By group theory, it can be shown that the lattice modes at q = 0 for a cubic crystal are three-fold degenerate. This degeneracy is lifted by the electromagnetic interaction in polar materials to give a splitting between the LO and TO modes. An example of the reectivity of a normally transparent material in the region where phonon excitation processes dominate is shown in Fig. 10.2. From the diagram, we see that for t < < , the dielectric is both highly reective and lossy. This range between t and is also observed as an absorption band in infrared absorption studies. 4. The dielectric function () approaches the static dielectric constant 0 as 0. Also, () approaches the high frequency dielectric function as approaches frequencies that are large compared with t and . Even when we consider to be large, we are still thinking of as being very much smaller than typical interband electronic frequencies. Lattice modes typically are important in the wavelength range 10 100m or 0.01 0.1eV or 50 1000 cm1 . h 5. The quantities 0 , , t and are not independent, but are related by a very general relation called the LyddaneSachsTeller relation: 2 0 2 = t which is written here for a crystal with two atoms/unit cell. (10.2)

10.2

Dielectric Constant and Polarizability

p = Elocal . (10.3)

The polarizability of an atom is dened in terms of the local electric eld at the atom,

The polarizability is an atomic property, the the dielectric constant will depend on the manner in which the atoms are assembled to form a crystal. For a non-spherical atom will be a tensor. The polarization of a crystal may be approximated as the product of the polarizabilities of the atoms times the local electric elds, P =
j

N j pj =
j

Nj j Elocal (j),

(10.4)

110

Figure 10.2: Reectivity of a thick crystal of NaCl vs. wave length at several temperatures. The nominal values of and t at room temperature correspond to wavelengths of 38 and 61 microns, respectively. The additional structure seen in the reectivity spectrum near is associated with defects.

111

where Nj is the concentration and j the polarizability of atoms or ions j, and Elocal (j) is the local eld at atomic sites j. If the local eld is given by the Lorentz relation, then P =(
j

N j j ) E +

4 P . 3

(10.5)

Solving for the susceptibility = Using the denition P = E 1 4 1 = +2 3

j 4 3

N j j
j

N j j

(10.6)

= 1 + 4 one obtains the ClausiusMossotti relation N j j .

j

(10.7)

This relates the dielectric constant to the electronic polarizability, but only for crystal structures for which the Lorentz local eld relation applies.

10.3

Polariton Dispersion Relations

The statements 15 in 10.1 provide an overview on optical studies of lattice modes. In this section we discuss the polariton dispersion relations which describe the interaction of light with the electric dipole moment associated with infrared absorption, and the LOTO splitting of the normal mode vibration of the atoms in the solid arising from these dispersion relations. Consider the equation of motion of an ion in a solid using the normal mode coordinate r, so that harmonic motion yields m = r + eE = m 2 r r where E = E0 eit (10.9) and r represents a lattice restoring force while eE is the force due to the actual electric eld E at an ion site. Maxwells equations give us 1 1 i curl H = D = (E + 4 P ) = (E + 4 P ) c c c (10.10) (10.8)

i 1 H (10.11) curl E = H = c c We also have a constitutive equation which tells us that the total polarization arises from an ionic contribution N er where N is the number of optical modes per unit volume and from an electronic contribution nE, where n is the electron concentration and is the electronic polarizability: P = N er + nE. (10.12) Equations 10.8, 10.10, 10.11 and 10.12 represent 4 equations in the 4 variables E, H, r, and P. 112

In writing Eq. 10.12 for the polarization vector P , we have considered two degrees of freedom: the ion system and the electron system. We further assume that these polarizations are accomplished independently. In formulating this calculation, the electric eld in all equations is the applied electric eld, since it is assumed that the lattice polarization eects are weak. In more sophisticated treatments, we must consider the eect of local eld corrections when the dielectric function is large, as occurs for example in the case of ferroelectrics. We now seek plane wave solutions for transverse wave propagation: ( E, H) in the xy plane and perpendicular to the Poynting vector, S = [c/(8)]Re(E H), and the Poynting vector is taken along the z direction
0 Ex = Ex ei(tKz) 0 Hy = Hy ei(tKz) 0 Px = Px ei(tKz)

(10.13) (10.14) (10.15) (10.16)

rx =

0 rx ei(tKz) .

Here K is the wave vector for the light, K = 2/. Using values for typical for lattice modes in NaCl, we have 60m and K 103 cm1 . Substitution of the harmonic solutions in Eqs. 10.1310.16 into the 4 equations (Eqs. 10.8, 10.10, 10.11 and 10.12) for the four variables E, H, r, and P yields: iKHy i 4i Ex Px = 0 c c (10.17) (10.18) (10.19) (10.20)

i Hy = 0 c e 2 rx + rx Ex = 0 m m Px N erx nEx = 0. iKEx +

Equations 10.1710.20 form 4 equations in 4 unknowns. To have a non-trivial solution to Eqs. 10.1710.20, the coecient determinant must vanish. We arrange the coecient determinant following the order of the variables in Eqs. 10.1310.16: (E x Hy Px rx ) : /c K 4/c 0 K e/m n /c 0 0 0 0 1 2 0 = 0. /m (10.21)

Multiplying out the determinant in Eq. 10.21, we get a quadratic equation in 2 4 [1 + 4n] 2 c2 K 2 + 4N e2 4n + + + K 2 c2 = 0. m m m m (10.22)

N e

Equation 10.22 is more conveniently written in terms of the parameters , 0 , and T where these parameters are dened in Eqs. 10.23, 10.25 and 10.27 given below: 113

1. The high frequency dielectric constant is written as = 1 + 4P /E, and is the parameter normally used to express the optical core dielectric constant when discussing electronic processes studied by optical techniques. From the equation of motion (Eq. 10.8), we conclude that at high frequencies ( T and we show below that T is the transverse optical frequency), the ionic displacement is small, for otherwise the acceleration would tend to . Thus as the frequency increases, the ions contribute less and less to the polarization vector. We thus have the result P = nE, so that the electronic contribution dominates and = 1 + 4n. (10.23)

2. The low frequency ( T ) dielectric constant is written as 0 . At = 0 the equation of motion Eq. 10.8 yields r = eE/ so that the polarization vector at zero frequency is N e2 P0 = [ + n]E; (10.24) and N e2 0 = 1 + 4[ + n]. (10.25) At a general frequency , we must from Eqs. 10.8 and 10.12 write () = 1 + 4[ N e2 + n]. m 2 m (10.26)

3. Finally, we introduce a frequency T dened as

2 T

(10.27)

which depends only on the restoring forces and not on the externally applied eld. Of course, these restoring forces will depend on internal elds, since electromagnetic interactions are responsible for producing these lattice vibrations in the rst place. We will later identify T with t , the transverse optical phonon frequency. Substitution of , 0 , and T into Eq. 10.22 yields the polarization dispersion relation
2 2 4 2 [c2 K 2 + T 0 ] + T c2 K 2 = 0.

(10.28)

Equation 10.28 has two solutions 2 = c2 1 1 2 2 2 (T 0 + c2 K 2 )2 T K 2 (T 0 + c2 K 2 ) 2 42

1/2

(10.29)

which are shown graphically in Fig. 10.3. Each solution in Eq. 10.29 is twofold degenerate, since E can be chosen in any arbitrary direction perpendicular to the propagation vector. The coupled excitation of the transverse optical phonon to the electromagnetic radiation is called the polariton and the picture in Fig. 10.3 is called the polariton dispersion relation. There is also a longitudinal direction for both the light and the lattice vibrations; for this case there is no coupling between the light and the phonons 114

Figure 10.3: Polariton dispersion relations showing the coupling between the transverse lattice vibrations and the electromagnetic radiation. In this gure, we clearly see the splitting of the LO and TO modes (L T ) induced by the ionicity of the solid. and the frequency is the same as in the absence of light. We therefore obtain a total of 6 modes for the 3 coupled optical lattice modes and the three electromagnetic modes (two transverse modes representing photons and one longitudinal mode). It is of interest to examine the solutions of Eq. 10.29 for small and large K vectors where we must remember that the scale of the Kvectors for light is a scale of 10 3 104 cm1 rather than 108 cm1 which describes the Brillouin zone dimension, corresponding to Brillouin zone dimensions. Thus the whole picture shown in Fig. 10.3 occurs essentially at q = 0 when plotting phonon dispersion relations q (q) for wave vectors q in the Brillouin zone. At small K vectors (|K| positive solution is given by 104 cm1 ), we have two solutions to Eq. 10.29. The 2 = which gives
2 2 T 0 / L ,

1 ( 2 0 + c2 K 2 ) T

(10.30)

(10.31)

thus dening the frequency L . In writing this solution we neglected the term c2 K 2 as K 0. This solution corresponds to the phonon branch with nite frequency at K = 0 and hence is an optical phonon mode. We will call this frequency L and later we will identify L with the longitudinal optical phonon mode frequency, . We shall see that the above denition is equivalent to taking frequency as the frequency where the real part of the dielectric function vanishes 1 ( ) 0. We also remember, that the longitudinal optical (LO) phonon does not interact with the electromagnetic eld. For a phononelectromagnetic interaction, we require that the electric eld be transverse to the direction of propagation. 115

With regard to the negative solution of Eq. 10.29, we expand the square root term in Eq. 10.29 to obtain: 2 K 2 c2 (10.32) 2 = 2 T T 0 + c 2 K 2 or c2 K 2 2 (10.33) 0 yielding the photonlike mode with a linear K dependence cK = for 0 T . (10.34)

At large K values (|K| 105 cm1 ), we solve the quadratic equation given by Eq. 10.29 in the large K limit and obtain positive and negative solutions. Using a binomial expansion for Eq. 10.29, we obtain the following positive and negative solutions. For the positive solution, i.e., K large, we obtain 1 c2 K 2 2 2 (T 0 + c2 K 2 ) = . (10.35) This is clearly the photon-like mode, since cK for = T . (10.36)

This result is almost identical to Eq. 10.34 obtained in the low K limit, except that now we have instead of 0 . Correspondingly, the phonon-like mode for large K arises from the negative solution: 2 T K 2 c 2 2 T . (10.37) 2 2 T 0 + c 2 K 2 We have thus introduced two frequencies: T and L and from the denition of L we obtain the LyddaneSachsTeller relation 2 0 2 = . T (10.38)

Now, T and L have well-dened meanings with regard to the dielectric function as can be seen in Fig. 10.3. From Eq. 10.12, we have for the polarization due to ions and electrons: P = N er + nE (10.39) while the equation of motion, Eq. 10.8, (F = ma) gives m 2 r = r + eE yielding r= so that eE eE/m = 2 m 2 T 2 (10.41) (10.40)

P () 1 N e2 /m 1 + = = 2 , E 4 4 T 2 116

(10.42)

since the electronic polarizability term is n = ( 1)/4. We therefore obtain: () = + 4N e2 /m 2 T 2 (10.43)

where represents the contribution from the electronic polarizability and the resonant term represents the lattice contribution. Neglecting damping, we have the result |()| as T , where the transverse optical phonon frequency = T is interpreted as the frequency at which the dielectric function () is resonant. The name reststrahl frequency denotes that frequency T where light is maximally absorbed by the medium. We would now like to get a more physical idea about . So far has been introduced as the phonon mode of the polariton curve in Fig. 10.3 near k = 0. From Eq. 10.43 we have the relation 4N e2 0 = + (10.44) 2 mT where 0 is dened by 0 ( = 0), so that 4N e2 2 = T (0 ) m (10.45)

and t = T is the frequency where () is resonant. Thus from Eqs. 10.1 and 10.43, we can write (0 ) (0 ) = + (10.46) () = + 2 2 / 2 ) (1 (1 2 /T ) t so that T = t . We dene as the frequency at which the dielectric function vanishes ( ) 0 so that setting () = 0 in Eq. 10.46 yields = or ( 0 ) 2 (1 2 /t )

(10.47)

2 0 2 = . t

(10.48)

Thus, the frequency , which yields a zero in the dielectric function, also satises the Lyddane-Sachs-Teller relation (Eq. 10.48). We illustrate the properties of and t in Fig. 10.4 where we see that the frequency dependence of the dielectric function () has two special features: a zero of () occurring at an innity or pole of () occurring at t . For t < < , the dielectric function () is negative, so that losses must occur and transmission is consequently poor. The frequency dierence between the two characteristic frequencies and t depends on the ionicity of the crystal. Thus, predominantly covalent materials like InSb which have weak ionicity have a smaller t splitting than alkali halide crystals which are highly ionic. For weakly polar materials like InSb, the treatment of the electric eld given here is adequate. For highly polar materials, one must also consider the local elds, as distinct from the applied eld. These local elds tend to increase the 117

Figure 10.4: The dielectric function () plotted as a function of normalized frequency /T . When damping is included, the real part of the dielectric function remains nite at T .

118

separation between t and , pulling t to low frequencies. Since mechanically hard materials tend to have high Debye temperatures and high phonon frequencies, the passage of t toward zero for ferroelectric materials (extremely high dielectric function and capable of spontaneous polarization) is referred to as the appearance of a soft mode. The Lyddane-Sachs-Teller relation is more general than the derivation given here would imply. This relation can be extended to cover anisotropic materials with any number of optical modes. In this context we can write the frequency dependence of the symmetrized dielectric tensor function associated with symmetry as
2 f,j T,j () = () + 2 2 ij j=1 T,j p

(10.49)

where f,j is the oscillator strength, j is the damping of mode j, and p is the number of modes with symmetry . An example where this would apply is the case of tetragonal symmetry where could refer to the in-plane modes (Eu symmetry) or to the out-of-plane modes (A2u symmetry). Figure 10.5 shows the measured reectivity for the lattice modes of TeO2 which has 4 formula units per unit cell (12 atoms/unit cell) can be described by a model based on Eq. 10.49 for polarization of the electromagnetic eld parallel and perpendicular to the tetragonal axis. Setting the damping terms in Eq. 10.49 to zero, j = 0, we obtain the result
p 2 l,j 2 () = 2 () j=1 t,j 2

(10.50)

which leads to the generalized LyddaneSachsTeller relation

p 2 l,j 0 (0) = = . 2 () j=1 t,j

(10.51)

Equation 10.51 can be generalized for anisotropic crystals by writing Eq. 10.50 for each component, keeping in mind that the optical selection rules dier for each component. The dependence of the reectivity on polarization and on temperature is illustrated for the tetragonal crystal TeO2 in Fig. 10.5. To nd the LO and TO modes associated with Eq. 10.49, we would look for zeros and poles of the dielectric function for a general direction of light propagation. For example, in a tetragonal crystal we can write (, ) = () () . () cos2 + () sin2 (10.52)

The observation of LO and TO phonon frequencies by optical measurements is made using two basically dierent techniques. In one approach, we make absorption, reection or transmission measurements, while in the other approach, light scattering measurements are made. These are often complementary methods for the following reason. Many important crystals have inversion symmetry (e.g., the NaCl structure). In this case, the phonon modes are purely odd or purely even. If the odd parity modes have dipole moments and couple directly to the electromagnetic elds, then they are infrared active. On the other hand, 119

Figure 10.5: Reectivity in paratelluride, TeO2 , for (a) E parallel and (b,c) perpendicular to the tetragonal axis at 295 K (b). The polarization E the tetragonal axis has only the A2u modes allowed whereas for E the tetragonal axis has only the E2u modes allowed. The points are experimental and the solid line is a model based on Eq. 10.49. (After Korn, et al., Phys. Rev. B8, 768 (1973).)

120

the even parity modes are not infrared active but instead may be Raman active and can be observed in a light scattering experiment. Thus, by doing both infrared absorption and Raman scattering measurements we can nd both even and odd parity optical phonon modes, except for the silent modes which because of other symmetry requirements are neither infrared nor Raman active. These concepts are discussed in detail in the group theory course. In modeling the phonon and free carrier contributions to the dielectric function it can happen that these phenomena occur over a common frequency range. In this case, we write the complex dielectric function for an isotropic semiconductor as follows in analyzing optical data p 2 2 2 L,j T,j p () + (10.53) = 1 () ( + ip ) 2 2 ij j=1 T,j where the rst and second terms are, respectively, the free carrier and the infrared-active phonon contributions to the dielectric function. In Eq. 10.53, p is the screened electronic 2 plasma frequency (p = 4ne2 /m (), and () is the core dielectric constant used to approximate the higher frequency electronic polarizability). The phonon contribution to Eq. 10.53 depends on L,j and T,j which are the jth longitudinal and transverse optic mode frequencies, while j and p are the phonon and plasma damping factors, respectively. The model given by Eq. 10.53 can, for example, be used to model the optical properties of the anisotropic compound La2 CuO4 which becomes a high Tc superconductor, upon addition of a small concentration of Sr. In this case it is important to obtain polarized reectivity measurements on oriented single crystals, and to carry out the KramersKronig analysis of reectivity data for each of the polarization components separately.

10.4

Light Scattering

Light scattering techniques provide an exceedingly useful tool to study fundamental excitations in solids, such as phonons, because light can be scattered from solids inelastically, whereby the incident and scattered photons have dierent frequencies. Inelastic light scattering became an important tool for the study of excitations in solids in the mid-1960s with the advent of laser light sources, because the inelastically scattered light is typically only 107 of the intensity of the incident light. In the light scattering experiments shown schematically in Fig. 10.6, conservation of energy gives: = 0 q (10.54) and conservation of momentum gives: K = K0 q (10.55)

where the 0 subscript refers to the incident light, K refers to the wave vector of the light and q refers to the wave vector for the excitation in the solid. Since K 0 = 2/ is very small compared with the Brillouin zone dimensions, measurement of the angular dependence of q (q) can then be used to provide dispersion relations for the excitations near q = 0. If q 0 , then |K| |K0 |, and we have |q| 2|K0 | sin(/2) so that |qmax | = 2K0 . 121

Figure 10.6: Raman scattering of a photon showing both phonon emission (Stokes) and absorption (antiStokes) processes. The scattering process is called Brillouin scattering when an acoustic phonon is involved and polariton (Raman) scattering when an optical phonon is involved. Similar processes occur with magnons, plasmons or any other excitation of the solid with the correct symmetry. If the excitation is an acoustic phonon, the inelastic light scattering process is called Brillouin scattering, while light scattering by optical phonons is called Raman scattering. Raman and Brillouin scattering also denote light scattering processes due to other elementary excitations in solids. The light scattering can be understood on the basis of classical electromagnetic theory. When an electric eld E is applied to a solid, a polarization P results P = E

(10.56)

where is the polarizability tensor of the atom in the solid, indicating that the positive charge moves in one direction and the negative charge in the opposite direction under the inuence of the applied eld. In the light scattering experiments, the electric eld is oscillating at an optical frequency 0 E = E0 sin 0 t. (10.57)

The lattice vibrations in the solid modulate the polarizability of the atoms themselves = 0 + 1 sin q t. so that the polarization which is induced by the applied electric eld is: P =E0 (0 + 1 sin q t) sin 0 t (10.59) =E0 0 sin(0 t) + 1 1 cos(0 q )t 1 1 cos(0 + q )t . 2 2 Thus we see in Fig. 10.7 that light will be scattered elastically at frequency 0 (Raleigh scattering) and also inelastically, being modulated downward by the natural vibration frequency q of the atom (Stokes process) or upward by the same frequency q (anti-Stokes 122 (10.58)

Figure 10.7: Schematic diagram of light scattering spectrum showing the central unshifted Rayleigh line, the upshifted anti-Stokes line (emission process), and the downshifted Stokes line (absorption process). The ratio of the Stokes to anti-Stokes can be used to estimate the temperature of the phonon system.

process). The light scattering process can also be viewed from a quantum mechanical perspective. If the system is initially in a state E , then light scattering can excite the system to a higher energy state E shown in Fig. 10.8a by absorption of an excitation energy (E E ). Similarly, the system can initially be in a state E and light scattering can serve to bring the system to a nal state of lower energy E by emission of an excitation of energy (E E ) as shown in Fig. 10.8b. The matrix element of the polarization vector between initial and nal states is written (when expressed in terms of quantum mechanics) as Pnm = P m d3 r = E m d3 r (10.60) n n where the polarizability is a second rank symmetrical tensor. The Stokes and anti-Stokes processes arise from consideration of the phase factors in this matrix element: m has a h h phase factor eiEm t/ while has a phase factor e+iEn t/ . The polarizability tensor has a n phase factor eiq t so that the integration implied by Eq. 10.60 yields Em En q = 0. h (10.61)

We should remember that the optical absorption process is governed by the momentum matrix element which is a radial vector. Of particular signicance is the case of a crystal with inversion symmetry whereby the momentum operator is an odd function, but the polarizability tensor is an even function. This characteristic feature has an important consequence; namely electronic absorption processes are sensitive to transitions between states of opposite parity (parity meaning even or odd), while light scattering is sensitive to transitions between states of similar parity. For this reason, light scattering and optical absorption are considered to be complementary spectroscopies, and together form basic tools for the study of the optical properties of elementary excitations in solids. It is important to draw a clear distinction between Raman scattering and uorescence. In Raman scattering, the intermediate states shown in Fig. 10.8a,b are virtual states and 123

dont have to correspond to eigenstates of the physical system any optical excitation frequency will in principle suce. In uorescence, on the other hand, the optical excitation state must be a real state of the system and in this case a real absorption of light occurs, followed by a real emission at a dierent frequency. The major reason why these two processes are sometimes confused is that Raman scattering in solids often has a much higher intensity when 0 is equal to an energy band gap h and this eect is called resonant Raman scattering. In such cases, the uorescent emission diers from the Raman process because uorescent phenomena take a nite time to occur. Typical Raman traces are shown in Fig. 10.9 for several III-V compound semiconductors. The laser wavelength is 1.06 m (Nd:YAG laser) which is a photon energy below the band gap for each material. The scattered light is collected at 90 with respect to the incident light and both the LO and TO phonon modes at q = 0 are observed. For the case of the group IV semiconductors there is no LOTO splitting and only a single optical Raman-allowed mode is observed (at 519 cm1 for Si). What is measured in Fig. 10.9 is the frequency shift between the incident and scattered light beams. For the range of phonon wave vectors where Raman scattering can be carried out, this technique is the most accurate method available for the measurement of the dispersion relations near the Brillouin zone center. By doing the Raman scattering experiment with polarized light, it is possible to get information on the symmetry of the lattice vibrations by monitoring the polarization of both the incident and scattered radiation. This approach is important in the identication of phonon frequencies with specic lattice normal modes. The inelastic neutron scattering technique, though less accurate than Raman scattering, has the advantage of providing information about phonons throughout the Brillouin zone. By using neutrons of low energy (thermal neutrons), it is possible to make the neutron wavelengths comparable to the lattice dimensions, in which case the inelastic scattering by 124

(E(65320'%7DCB'@8 & 4 " 1 ) ( & $ " (A 9       

Figure 10.8: Schematic energy level diagram for the (a) Stokes and (b) antiStokes processes. In this gure the solid lines denote real processes and the dashed lines virtual processes.

7( 65320'%#! (&4 "1 ) (&$"      

Figure 10.9: Raman spectra of three zinc-blende-type semiconductors showing the TO and LO phonons in both Stokes and anti-Stokes scattering.

125

a lattice vibration can cause a large momentum transfer to the neutron.

10.5

Feynman Diagrams for Light Scattering

Feynman diagrams are useful for keeping track of various processes that may occur in an inelastic scattering process that absorbs or creates an excitation. The basic notation used in drawing Feynman diagrams consists of propagators such as electrons, phonons or photons and vertices where interactions occur, as shown in Fig. 10.10(g). The rules in drawing Feynman diagrams are: Excitations such as photon, phonons and electron-hole pairs in Raman scattering are represented by lines (or propagators) as shown in Fig. 10.10(g). These propagators can be labeled with properties of the excitations, such as their wavevectors, frequencies and polarizations. The interaction between two excitations is represented by an intersection of their propagators. This intersection is known as a vertex and is sometimes highlighted by a symbol such as a lled circle or empty rectangle. Propagators are drawn with an arrow to indicate whether they are created or annihilated in an interaction. Arrows pointing towards a vertex represent excitations which are annihilated. Those pointing away from the vertex are created. When several interactions are involved they are always assumed to proceed sequentially from the left to the right as a function of time. Once a diagram has been drawn for a certain process, other possible processes are derived by permuting the time order in which the vertices occur in this diagram. The basic diagram for the Raman process is given in Fig. 10.10(a) taken from the Yu and Cardona book on Fundamentals of Semiconductors. The other permutations of (a) obtained by dierent orders of the vertices are given in Figs. 10.10(bf). We then use the Fermi Golden rule for each diagram, multiplying the contributions from each vertex. For example, the rst vertex in Fig. 10.10(a) contributes a term to the scattering probability per unit time of the form n|HeR (i )|i [ i (En Ei )] h where the sign (+) corresponds to absorption and () to emission and H eR (i ) denotes the interaction between the electron and the electromagnetic radiation eld. The interaction for the second vertex Heion (i ) between the electron and the lattice vibrations of the ion (or the electron-phonon interaction) and the corresponding energy denominator is i (En Ei ) q (En En ) = [ i q (En Ei )] h h h h and for the third vertex the denominator becomes [ i q s (En Ei )] but since h h h the initial and nal electron energies are the same, energy conservation requires ( i h 126

Figure 10.10: Feynman diagrams for the six scattering processes that contribute to one-phonon (Stokes) Raman scattering. (Taken from Yu and Cardona.) (g) Symbols used in drawing Feynman diagrams to represent Raman scattering.

127

 q s ) to yield the probability per unit time for Raman scattering for diagram (a): h h Pph (s ) =
2 h n,n i|HeR (s )|n n |Heion |n n|HeR (i )|i [ i (En Ei )][ i q (En Ei )] h h h 2

(10.62)

( i q s ). h h h Then summing over the other 5 diagrams yields the result Pph (s ) = +
2 h n,n i|HeR (i )|n n|Heion |n n |HeR (s )|i [ i (En Ei )][ i q (En Ei )] h h h

i|HeR (i )|n n|HeR (s )|n n |Heion |i [ i (En Ei )][ i s (En Ei )] h h h

i|HeR (s |HeR + [ s (E)|n n|Heion |nhn(E (i )|i h h Ei )] n Ei )][ s q

n

i|HeR (s )|n n|HeR (i )|n n|Heion |n [ s (En Ei )][ s + i (En Ei )] h h h

(10.63)

i|Heion |n n|HeR (i |HeR + [ q (En Ei )][ )|n hn(E (s )|i Ei )] h h q + i

n

i|Heion |n n|HeR (s )|n |HeR + [ q (En Ei )][ q n(E (i )|i h h h s Ei )]

n

( i s q ). h h h Although Eq. 10.63 is not generally used to calculate scattering intensities directly, Feynman diagrams similar to those in Fig. 10.10 are widely used in physics.

10.6

Raman Spectra in Quantum Wells and Superlattices

Raman spectroscopy has also been used to study quantum well and superlattice phenomena. One important example is the use of Raman spectroscopy to elucidate zone folding phenomena in the phonon branches of a superlattice of quantum wells. Since the Raman eect is highly sensitive to phonon frequencies, this technique can be used to characterize quantum wells and superlattices with regard to the composition of an alloy constituent (e.g., the composition x of an alloy such as Six Ge1x ). The Raman eect can then be used to determine the amount of strain in each constituent from measurement of the phonon frequencies. Zone folding eects in the phonon dispersion relations are demonstrated in a superlattice of [GaAs (13.6)/AlAs (11.4)] 1720 periods. The observed Raman spectra are shown A A in Figs. 10.11(a) and (b), demonstrating the zone folding of the LA branch. The dierence in the force constants between the GaAs and AlAs constituents causes splittings of the zonefolded phonon branch, as shown in Fig. 10.11(c). The peaks in the Raman spectrum at 64 cm1 and 66 cm1 are identied and labeled with the zone folded modes of the (1) (1) LA branch with symmetries A1 and B2 , consistent with the polarization of the incident and scattered photons. At higher frequencies the Raman spectrum of Fig. 10.11(a) shows additional structure related to the zone folded LO phonon branch. Here we note that the normally threefold levels of T symmetry of the cubic crystal are split into E and B 2 symmetries in the superlattice because of its lower tetragonal symmetry. The twofold level of E symmetry can be further split by the LOTO splitting which occurs in ionic solids. 128

Figure 10.11: (a) Raman spectra of a superlattice consisting of 1720 periods of a 13.6 A GaAs quantum well and a 11.4 AlAs barrier. The polarizations for the incident and A scattered light are arranged so that only longitudinal phonons are observed. (b) Dispersion of the LA phonons in the superlattice. (c) An expanded view of the 65 cm1 region of the zone folded LA branch near k 0. (C. Colvard, T.A. Grant, M.V. Klein, R. Merlin, R. Fischer, H. Morkoc and A.C. Gossard, Phys. Rev. B31, 2080 (1985).)

129

As another example, Raman spectroscopy can be used as a compositional characterization technique to conrm the chemical composition of a semiconductor alloy. This characterization is based on the identication of the Raman-active modes and the measurement of their frequency shifts and their relative intensities. The strain induced by the lattice mismatch at the interface between Si0.5 Ge0.5 and a GaAs (110) surface is responsible for the dependence of the frequency shifts of the Ge-Ge, Si-Si and Si-Ge phonon lines on the thickness of the quantum wells in the spectra shown in Fig. 10.12 for Si0.5 Ge0.5 layers of various thicknesses on a GaAs (110) surface. Since phonon frequencies depend on (K/M ) 1/2 (where K is the force constant and M is the ion mass) the mode frequencies of the GeGe, GeSi and SiSi optical mode vibrations are very dierent as seen in Fig. 10.12. Therefore the amount of interface strain can be sensitively monitored by Raman scattering. Note the disappearance of the GaAs Raman lines (associated with the substrate) as the thickness of the Si0.5 Ge0.5 overlayer increases.

130

Figure 10.12: Raman spectra for various thicknesses of Si0.5 Ge0.5 on an GaAs (110) substrate. Here the dependence of the Si Si, GeGe, and SiGe bond lengths on the thickness of the Si0.5 Ge0.5 layer can readily be seen. The samples were grown at 720 K and the measurements were made at 80 K using a laser with a wavelength of 457.9 nm. (G. Abstreiter, H. Brugger, T. Wolf, H. Jorke and H.J. Herzog, Phys. Rev. Lett. 54, 2441 (1985).

131

Chapter 11

Non-Linear Optics
References:
Bloembergen, Non-Linear Optics, W.A. Benjamin, 1965. Loudon, Quantum Theory of Light, 2nd Ed., Oxford, 1983. Yariv, Quantum Electronics, 3rd Ed., Wiley, 1989. Yariv, Optical Electronics, 3rd Ed., Holt, Rinehart and Winston, 1985. Yariv and Pearson, Progress in Quantum Electronics, Vol. I. Pt. I.

11.1

Introductory Comments

Non-linear optics became an important eld of activity in the 1960s with the advent of high power laser sources such as: CO2 YAG:Nd3+ ruby argon at at at at 10.6 m 1.06 m 6943 A 5145 A

The topics we will consider under heading of non-linear optics are: 1. harmonic generation 2. paramagnetic oscillation 3. frequency conversion. The signicant point in non-linear optics is that, when the electromagnetic elds become strong enough, the dielectric function becomes dependent on the electric eld E. We write the total dielectric function as T = + NL E (11.1) 132

where is the linear response and NL E represents the non-linear term. Thus, nonlinear eects become more important as the magnitude of the electric eld is increased. To observe non-linear eects, we require elds of magnitude |E| 106 volt/cm. Fields of this magnitude are readily available with high power laser sources. Nonlinear eects are of great importance in quantum well structures. It is sometimes more convenient to express these non-linear eects in terms of the polarization per unit volume P P = 0 E+ 1 E E+ 2 E E E +

(11.2)

where 0 is the linear susceptibility tensor, 1 is the lowest order non-linear susceptibility important in non-linear materials with no center of inversion; and 2 is the second order non-linear susceptibility which comes into play for non-linear eects in cubic crystals with a center of inversion for which the rst order term vanishes by symmetry. For the present discussion we will only consider the lowest order non-linear term: P N L = 1 E E

(11.3)

where N L denotes nonlinear and the non-linear susceptibility is in reality a 3rd rank tensor in the same sense that 0 for the linear response is a second rank tensor in an actual solid state medium. Likewise 2 in Eq. 11.2 is a 4th rank tensor. We will not make any use of the tensorial properties of 1 here, because we want to keep things simple. Maxwells equations in this case become E+ 0 H c t =0 H 0 E c t = 4 c t

1 E E

(11.4)

where the term in 1 E E is the non-linear term. We thus obtain the non-linear wave equation 0 0 2 E 40 2 2 1 E E. E 2 (11.5) = 2 2 c t c t2 The wave equation without the non-linear term has plane wave eigen functions. That is, if we have an incident eld with more than one frequency E = E1 ei(K1 z1 t) + E2 ei(K2 z2 t) , the wave equation
2

(11.6)

0 0 2 E c2 t2

=0

(11.7)

is valid for each of the waves in Eq. 11.6. This insures that no mixing occurs and each frequency propagates independently through the linear medium. Now what does the non-linear term do? Here we have to take a product E E. Clearly we will get terms at 21 and 22 (frequency doubling or second harmonic generation) and also at [1 + 2 ] (frequency mixing). This means that we now no longer have plane wave solutions Ei = Ei ei(Ki zi t) (11.8) 133

for Ei polarized along x, where Ei is a constant amplitude. Instead we must look for a more generalized form. For example, we could seek a solution of the form of a modied plane wave Ei = Ei (z)ei(Ki zi t) (11.9) where Ei (z) now has a weak z dependence. This approach is in the spirit of perturbation theory. Now in taking spatial derivatives of Ei we will get two terms Ei Ei = + iKi Ei ei(Ki zi t) z z (11.10)

where the term in iKi Ei is the large term and the term in Ei /z represents the small perturbation. Thus in taking two derivatives we will get 2 Ei Ei = iKi 2 + iKi Ei ei(Ki zi t) , 2 z z (11.11)

retaining only the lowest order term in the perturbation. We will now show that coupling to frequency 3 is possible for waves at frequencies 1 and 2 provided that 3 = 1 + 2 . From the wave equation Eq. 11.5, we see that if the incident eld has two frequencies 1 and 2 , we will get a perturbation driving term on the right hand side of the non-linear wave equation and also a perturbation term on the left hand side of this equation due to the z dependence of Ei (z). Assuming a solution at some frequency 3 (to be determined from the wave equation, Eq. 11.7) we can write an equation for the right and left hand perturbation terms 2iK3 E3 i(K3 z3 t) 40 e = (1 + 2 )2 1 E1 E2 ei(K1 z1 t) ei(K2 z2 t) z c2 (11.12)

To satisfy the right hand side of Eq. 11.5 we need to match the time phase terms on the left hand side, yielding 3 = 1 + 2 , (11.13) which indicates that mixing has occurred. Because of the dispersion properties of crystals, the wave vector for light K will be a function of and we cannot in general cancel the phases for all frequencies. Thus, some mismatch K = K3 (K1 + K2 ) will generally occur. Phase matching is achieved when K 0 and in this case the three waves will be coherent. In free space, phase matching is automatically satised since there is no nonlinear response in free space. In a solid, K = n/c where n is the index of refraction. Solids have the property that the optical constants are frequency-dependent so that in general n1 + n2 = n3 . If, however, K is small, phase matching is approximately satised. In fact, provided that the phase changes by less than , some coherence will be achieved. We thus introduce a phase coherence length c dened as c = /K over which some degree of coherence is achieved.

11.2

Second Harmonic Generation

For the nonlinear process corresponding to second harmonic generation we have 1 = 2 = and 3 = 2. The non-linear contribution to the polarization will be proportional 134

to (E )2 and the power generated at 2 will be proportional to (E )4 . If phase matching is achieved, the power produced at the second harmonic will be maximized. For phase matching we require 3 n 3 = 1 n 1 + 2 n 2 (11.14) or 2 3 = 2 1 , n n (11.15) which can be written more conveniently as n2 = n , which says that we require the index of refraction at 2 to be equal to the index at frequency for phase matching. For solids, the index of refraction n will generally be frequency dependent, so phase matching would seem dicult to achieve. By using anisotropic materials and selecting particular directions of propagation, it is sometimes possible to arrange matters so that n for one polarization direction is equal to n2 for another polarization direction. Ecient harmonic generation has been achieved using the semiconductor Te, which crystallizes in a hexagonal structure and has a rather dierent index of refraction for the polarization E c than for E c. To achieve frequency doubling in Te, it is convenient to use a CO2 laser source, since Te which has a band gap of 0.344 eV is quite transparent at both and 2 for the CO2 laser line at 10.6m. With frequency doubling, it is possible to convert infrared light to visible radiation and thus to utilize the highly developed technology for the detection of visible light signals. The process of second harmonic generation need not be considered a small or weak eect. High conversion eciencies (> 50%) can be achieved.

11.2.1

Parametric Oscillation

Here laser power is applied to a non-linear crystal at a pump frequency 3 . Oscillations are induced in the crystal at frequencies 1 and 2 , the signal and idler frequencies respectively. The signal and idler frequencies are determined by the frequency condition 3 = 1 + 2 and the phase matching condition K = 0, K3 = (K1 + K2 ). (11.17) (11.16)

It is only when phase matching occurs that the two waves will interact suciently to produce any measurable nonlinear eects. The phase matching condition can also be written as 3 n 3 = 1 n 1 + 2 n 2 (11.18)

where n1 and n2 are refractive indices at frequencies 1 and 2 and are determined by the propagation direction and polarization of the modes at 1 and 2 . In non-cubic materials, the index of refraction depends on the direction of the E eld relative to the crystallographic directions. Thus, by changing the propagation direction of the pump frequency relative to the crystal optic axis (e.g., the c axis in a hexagonal crystal like Te), it is possible to tune the signal and idler frequencies 1 and 2 . Furthermore, since the indices are temperature dependent, tuning can also be accomplished by varying the temperature of the non-linear crystal; tuning with application of uniaxial stress can also be accomplished. Parametric oscillation need not be a small eect. Using a Q-switched ruby laser as a pump at 6943 135

 on a non-linear LiNbO3 crystal, a signal at 1.04m and an idler at 2.08m have been A achieved with 60 kW of signal power generated for 270 kW of input power or a conversion eciency of 20%. Tuning with a parametric oscillator between 0.54m and 3.7m has also been achieved.

11.2.2

Frequency Conversion

For frequency conversion, two frequencies are applied to a non-linear crystal and the sum frequency (up converter) or dierence frequency (down converter) is generated. In this experiment, we might impose a high power signal at a pump frequency 3 and a lower power signal at 2 . These signals mix in the non-linear medium to produce a signal at 1 = 3 2 or at 1 = 3 + 2 (up converter). (11.20) Phase matching determines whether up-conversion or down-conversion actually occurs. The power from the signal at 3 drives the system at frequencies 1 and 2 , and as a function of length of the non-linear crystal, the amounts of power at 1 and 2 can be varied. Thus, by choosing the length properly the conversion of power to frequency 1 can be maximized. Frequency conversion is attractive for practical applications because up-conversion can be exploited to convert an infrared signal into the visible region where detectors are fast and sensitive. Down-converters can be exploited to create a dierent frequency in the far infrared where high power sources have been unavailable until about 1965, when some far infrared lasers were rst built. For example, the two strong CO2 laser lines at 10.6m and 9.6m can be mixed to get a far infrared signal. (down converter) (11.19)

136

Chapter 12

Electron Spectroscopy and Surface Science

References:
Demuth, Physics in a Technological World, French, Editor, AIP, NY 1988. Burns, Solid State Physics, Academic Press 1985, Chapt. 7.

12.1
12.1.1

Photoemission Electron Spectroscopy

Introduction

Photoemission is one of the most important of the electron spectroscopy techniques. In photoemission, the excitation particle is the photon and the excited particle is the electron. What we measure is the dielectric response function which gives us information about elementary excitations and the electronic structures of the solid. Depending on the excitation energies, photoemission measures the density of states, energy distribution of the joint density of states, and probes the valence bands and the core levels. With the advent of the use of synchrotron radiation and angle-resolved spectroscopy, the mapping of the electron energy bands has become possible. Other electron spectroscopies or related surface techniques include: 1. ESCA (Electron Spectroscopy for Chemical Analysis) which measures the chemical shift and thereby probes the local environment and oxidation state of the compound also called x-ray photoelectron spectroscopy (XPS) 2. AES (Auger Electron Spectroscopy) is a two electron process mainly used for elemental analysis of surface constituents 3. X-ray Fluorescence again is mainly used for elemental (chemical) analysis 4. ELS (Electron Loss Spectroscopy) which like optical spectroscopy gives the dielectric function of the material also called electron energy loss spectroscopy (EELS) 5. LEED (Low Energy Electron Diraction) mainly used for structural analysis. 137

6. RHEED (Reection High Energy Electron Diraction) mainly used for in situ characterization of superlattices during layer-by-layer growth. 7. STM (Scanning Tunneling Microscopy) which is used to obtain atomic resolution of atoms and molecules on surfaces. All of the above, except LEED, involve an inelastic scattering mechanism. In photoemission the photoelectric current (I) can be written in the form I = I(E, e , e , ; , p , p , p ) h (12.1)

where E, e , e , are respectively the kinetic energy, polar angle, azimuthal angle, and spin of the electron, and , p , p , p are respectively the frequency, polarization, polar angle, and azimuthal angle of the photon. For the various experimental measurements, dierent variables are held constant. The most common quantities that are measured in photoemission experiments include 1. EDC (Energy Distribution Curves) where the photoelectron current is observed as a function of electron energy I = I(E) with all other parameters held constant 2. CIS (Constant Initial State Spectroscopy) where E is kept constant, and the h photoelectron current is measured as a function of electron energy and photon energy I = I(E, )|E =const h h 3. CFS (Constant Final State Spectroscopy) where I = I( ) is measured h 4. ARPS (Angular Resolved Photoemission Spectroscopy) where the angles are allowed to vary. Historically, photoemission was rst used to study the work function of solids, and to study core levels in molecules. Both UPS (Ultraviolet Photoelectron Spectroscopy) and XPS (X-ray Photoelectron Spectroscopy) were used to probe core levels. In the photoemission process light incident on a sample is absorbed in a length characterized by the optical skin depth. In this optical skin depth, electrons can be excited to ionization states and are eventually emitted. Because of the much stronger interaction of electrons with matter (in contrast to the case of photons), the characteristic absorption length for the photo-excited electrons is much smaller than that for the exciting photons and as a consequence, only electrons that are excited close to the surface will be emitted. By applying an electric eld to the sample, some of the electrons generated in the optical excitation process are collected and their kinetic energy is measured. The following two quantities are observed in standard photoemission studies: 1. the photoelectric yield (dened as the number of electrons that are produced per unit of incident photon ux) as a function of photon energy, 2. the energy distribution of the emitted electrons, for various values of the incident photon energy. These measurements provide information on interband transitions through analysis of structure in the photoelectric yield curves, and on the density of valence states through the shape 138

Figure 12.1: The potential barrier at a surface showing the work function.

of the electron distribution curves. Because this technique provides one of the few methods for studying the density of states (particularly for low lying valence states), this has become an important measurement technique. Furthermore because of its surface sensitivity, ultraviolet photoemission spectroscopy provides a useful tool for contrasting electronic states characteristic of the surface relative to states characterizing the bulk. In describing the photoemission process in a metal, we make use of the model for the potential barrier at a surface shown in Fig. 12.1. In sodoing, we illustrate one of the classical applications of the photoemission process in measuring the work function of a solid. In this model we assume that our surface is in the xy plane, and that pz is the electron momentum (in the direction normal to this surface) which the electron acquires through the photo-excitation process. From the diagram, we see that it will be possible for this electron to escape from the surface (by photoemission) provided that p2 z + E0 h 2m (12.2)

h where E0 is called the vacuum level, located at an energy 0 above the band extremum. The work function e represents the minimum kinetic energy that an electron must be given by the light in order to escape from the surface (see Fig. 12.1). For the electron to retain any information about its initial state the mean free path (see Fig. 12.2) must be greater than the penetration depth of the exciting radiation. The photocurrent I will then be proportional to the number of electrons escaping from the surface of the metal I=e pz n(pz )D(pz )dpz (12.3)
0

where D(pz ) is the escape probability for an electron of momentum pz and n(pz ) is the corresponding electron density, which can be expressed through the Fermi distribution function n(pz ) = 2 3 h

dpx dpy e

1
p2 ( 2m EF )/kB T

(12.4) +1

The discussion given above is appropriate to photoemission from a metal. In the case of semiconductors there is an energy gap and the Fermi level lies in this energy gap, in 139

Figure 12.2: The mean free path for an electron in various materials as a function of energy.

Figure 12.3: Schematic diagram of the states near the Fermi energy of a semiconductor.

contrast to the situation in metals where the Fermi level lies at the top of the occupied electron states within an energy band. For semiconductors, the work function e is still dened relative to the Fermi level but the threshold energy is now increased to e + where is the energy dierence between the highest lying valence band maximum and the Fermi level as seen in Fig. 12.3. For semiconductors it is customary also to refer to the electron anity, denoted by E a on the diagram and representing the energy dierence between the vacuum level E 0 and the bottom of the conduction band. We can see how interband transitions are detected in the photoemission process by the following argument. Suppose that Ea < Eg . The threshold for photoemission requires the incident photons to have an energy of at least: > e + = Ea + Eg . h (12.5)

The photoemission process will dominate until the photo-emitted electrons are themselves energetic enough to make electron-hole pairs through collisions with other electrons. The threshold for this secondary interband transition process (whereby the photo-excited electron has enough energy to produce a second photoelectron) is > (Ea + Eg ) + Eg = Ea + 2Eg . h 140 (12.6)

Figure 12.4: Quantum yield vs. photon energy above the threshold energy.

The threshold for this interband process gives rise to structure in the photoemission distribution curves identied with interband transitions. We call this general eld of study photoemission spectroscopy. The advent of the theoretical development of the electronic structure of solids in the 50s made the microscopic understanding of photoemission possible. Furthermore, instrumental advances in high vacuum technology in the late 60s extended the photon range of ultraviolet photoelectron spectroscopy (UPS) to greater than 6 eV. Simultaneously the use of synchrotron radiation as a light source gives tunability of the excitation frequency and high power densities from the visible to hard xray frequencies. Angle resolved techniques are now commonly used to gain an understanding of the electronic structure of solids. In the photoemission process, three basic things must happen: (3-step model) 1. optical excitation of an electron from an occupied state 2. transport of the photo-excited electron to the surface 3. the electron must escape from the surface and into the vacuum region In order for the electron to escape from the surface into the vacuum region where the electron is collected, it must have sucient kinetic energy. Measurements of the photoelectric yield exhibit a threshold energy and thereby provide a measure of the work function. Photoelectric yield data are plotted in terms of the quantum yield (dened as the number of electrons emitted per incident photon) vs. photon energy as shown in Fig. 12.4.

12.1.2

Energy Distribution Curves

Of greater interest however is the energy distribution of the photo-emitted electrons N (E) which is dened as the number of electrons emitted with energy E in the range E relative to the total number of electrons produced per photon. We show that the probability that an electron of energy E is produced is proportional to the initial density of states at energy (E 141

 ) written as g(E ). The intensity prole of the electrons emitted in a photoemission h h experiment will contain both the primary electrons which suer no inelastic collisions, and the secondary electrons that suer at least one inelastic collision. The photoelectric current can then be written as I(E, ) = Ip (E, ) + Is (E, ) (12.7) where Ip , Is are respectively identied with primary and secondary electrons. Ip depends on three factors according to the three-step model: Ip (E, ) = P (E, ) T (E) D(E) (12.8)

where P (E, ) is the probability that a photoelectron of energy E is excited by a photon of energy , T (E) is the transmission function of the excited electrons and D(E) is the h escape function of the excited electron. We can write T (E) as T (E) = e (E)/ph () 1 + e (E)/ph () (12.9)

where e is the mean free path of the electrons and ph is the attenuation length of photon. Likewise, we can write D(E) as D(E) =

1 2 [1

( EF + ) 2 ] E

for E > EF + e (12.10) otherwise

where EF denotes the Fermi level and is the work function. If we consider only bulk states, and direct transitions, the energy distribution takes the form P (E, ) n,n h d3 k(En (k) En (k) )(En (k) E) (12.11)

The rst function represents the joint density of states for optical absorption and the second function selects out the energy that is set by the energy analyzer. Thus the structures of the EDCs (energy distribution curves) mimic those of the joint density of states and thus give information concerning the joint density of states. The experimental data for the photo-emitted electron energy distribution are taken for a variety of photon energies as shown in the curves in Fig. 12.5 and N (E) is plotted as a function of (E ) in h order to relate the electron energy distributions to the density of states at the same initial energy. Peaks in the density of initial states give rise to peaks in N (E) at the same value of E , independent of the energy of the photons involved in the excitation process. h Each curve in Fig. 12.5 for aluminum is labeled by the incident photon energy. The dashed curve is a density of states curve for the occupied electron states in aluminum obtained from the interpretation of these data. Note the threshold appearing at the Fermi level. Since the onset of interband transitions corresponds to discontinuities in the density of states spectrum, the EDC curves can also be used to identify interband transitions. To interpret valence band states we make use of the fact that d-bands have low dispersion and therefore a high density of states over a narrow energy region, while s- and p-bands have a low density of states over a wide energy region (see Fig. 12.5). Since laboratory ultraviolet sources are weak and dicult to work with, it is common to use a monolayer of cesium on the surface to lower the work function and the threshold 142

Figure 12.5: Energy distribution curves for photoelectrons in aluminum for various photon energies. The dashed curve shows the density of states over a wide energy region.

energy for the photoemission process. This allows the photoemission experiments to be carried out at somewhat lower photon energies where laboratory sources are more intense. More recently, intense synchrotron radiation ultraviolet sources have become available at a few of the national accelerator facilities and this has accelerated the development of photoemission spectroscopy research. If we allow nondirect transitions to occur, then the energy distribution function P (E, ) takes the form P (E, ) nn d3 kd3 k | n |p|n |2 (En (k ) En (k) )(En (k ) E). h (12.12)

If we rewrite P (E, ) as P (E, ) n d3 k[E En (k)] h d3 k [En (k ) E]| n |p|n |2 (12.13)

then P (E, ) is expressed as a weighted average of the initial and nal density of states. Thus indirect transitions can be invoked to explain stationary structures in the EDCs as we scan the photon frequency. Modications to the three step model have been made to include the possibility of an energy dependent electron mean free path e (E) = vg Te (E) = 1 | h
k E(k)|Te (E)

(12.14)

and the possibility that the transmission function Te (E) is described by more than one core state that couples to Bloch states. In Eq. 12.14 vg denotes the group velocity. 143

12.1.3

Angle Resolved Photoelectron Spectroscopy

Advances in angle-resolved photoemission have made photoemission an even more powerful experimental technique especially for the study of electronic band structure. If one measures the kinetic energy and the propagation direction of the electron by the conservation of wave vector parallel to the surface, we obtain K|| = k|| + G|| (12.15)

where K and k are respectively the wave vectors in vacuum and in the solid and G is a reciprocal lattice vector of the solid. Wave vector conservation together with the energy conservation E = Ef (k) (12.16) where Ef (k) denotes the energy of the electron and the zero of energy is taken at the vacuum level gives 2 h 2 E= (K 2 + K|| ) (12.17) 2m 2 (k|| + G|| )2 h 2 K h 2 = Ef (k) 2m 2m (12.18)

allowing the determination of Ef (k) as a function of k from photoemission data. Thus with a good band structure calculation the energy for all the directions of the bulk photoemission are determined. The functional form of Eq. 12.18 is especially suitable for layered materials due to the fact that the k dispersion is very small. Thus for layered materials the band structure is approximately 2-dimensional. Each peak in the EDC will give rise to a point on the E vs. k|| plot and thus E vs. k can then be mapped uniquely. For a three-dimensional system, a knowledge of the energy band structure is needed since k is not determined if Ef (k) is not known.

12.1.4

Synchrotron Radiation Sources

Before the availability of synchrotron radiation, photoemission was carried out using a few strong discrete line sources. Synchrotron radiation has provided us with a strong tunable source from the infrared to the x-ray region of the electromagnetic spectrum. Synchrotron radiation is emitted by electrons in circular accelerators, such as synchrotrons and storage rings. This radiation is a consequence of the centripetal acceleration of the particle moving in a circular path at relativistic velocities (close to the velocity of light). The energy to produce this radiation is supplied by particle accelerators. Synchrotron radiation has a number of properties which make it extremely useful. First, this photon source can be extremely intense, several orders of magnitude more intense than other broad-band sources. Second, it can have a very broad frequency spectrum, including the ultraviolet and x-ray regions where there are no other intense, tunable sources. The center of the spectrum is near the energy = h 3 c h R (12.19)

144

Figure 12.6: Radiation emission pattern of electrons in circular motion: Case I, nonrelativistic electrons. Case II, relativistic electrons. Synchrotron radiation sources operate under case II.

E mc2 is the ratio of the accelerator energy to the particles rest energy, and = R= mc2 eB

where

(12.20)

(12.21)

is the radius of the circular path. For example, a 1 GeV accelerator with a 1 tesla magnetic eld and a radius of 3 meters gives of about 1 keV. The third important property of h synchrotron radiation is that it is highly collimated, being conned to a narrow searchlight beam in the direction tangent to the orbit with angular spread 1 radians as shown in Fig. 12.6. A fourth useful property is the high degree of polarization of the radiation in the plane of the orbit. Fifth, devices called wigglers and undulators have recently been developed which enhance the intensity of the radiation in a particular part of the spectrum, adding one or two orders of magnitude to the already high intensity in the desired region. The rst experiments using synchrotron radiation were carried out in the so-called parasitic mode, at synchrotrons being used for particle physics research where the accelerators are optimized for the particular set of particle-physics experiments being carried out. Furthermore these synchrotrons operate in a pulsed mode, where the electrons are accelerated in bunches up to the maximum energy desired. The desired maximum energy may vary from experiment to experiment. An example of the spectral distribution available from the Stanford SLAC facility is shown in Fig. 12.7. Dedicated sources of synchrotron radiation are becoming increasingly available, such as the National Light Source at Brookhaven National Laboratory, for both the UV and x-ray regions. These are operated as storage rings, where the electron beam is maintained at a constant energy for long periods of time. Some synchrotron radiation work is still being done in the parasitic mode at storage rings being used for colliding-beam high-energy physics experiments, but here the beam currents are lower and hence the radiation is weaker. The new synchrotron radiation sources have made possible many new experiments, not only advances in photoemission such as angle-resolved experiments, but also advances in crystal structure determination, microlithography, x-ray uorescence, and the determination of local environments on surfaces using x-ray absorption ne structure (EXAFS). 145

Figure 12.7: Photon intensity vs. photon energy for various maximum accelerator energies E0 showing the spectral distribution of a synchrotron radiation source. The photon energy at which the maximum photon intensity occurs is denoted by c on the gure.

12.2
12.2.1

Surface Science
Introduction

Many electron devices depend on the electronic properties of surfaces. Because of the geometrical eect of a two-dimensional surface, atoms at a surface have fewer neighbors than similar atoms in the bulk. Therefore, the electronic energy levels at surfaces are dierent from what they are in the bulk. For example, a silicon atom in bulk silicon is surrounded by four tetrahedral bonds. On the surface, a silicon atom will have fewer bonds, and the surface valence electrons that do not participate in bonding are described as dangling bonds. These surface valence electrons give rise to new electronic states called surface states. When the surface states are located in the band gap of a semiconductor or insulator, they are more readily detected. A probe of electronic energy states with a skin depth that is large a compared with a lattice constant a is sensitive to the bulk electronic states because the surface atoms comprise a small fraction of the total number of atoms that are probed. On the other hand, a probe with a short skin depth (such as electrons in the ten and hundred eV range) is especially sensitive to the surface atoms. In addition, impurity atoms are preferentially adsorbed on the surface. These impurity atoms also give rise to surface states. Since adsorbed impurity atoms are important in catalyzed chemical reactions, there is considerable interest in studying these surface states. At the present time there is a great deal of work being done on the study of surfaces and on their electronic surface states. One reason is the availability of new experimental probes, such as the photoemission experiments already discussed and the STM probes discussed in this chapter, using ultra-high vacuum equipment. A second reason is the recent improvement in the calculation of surface states and of the total energy of dierent surface structures. A third reason is related to the smaller dimensions of semiconductor electronic devices and the increasing importance of surfaces in these devices. A number of dierent experimental techniques are used to study surfaces. The photoemission process discussed in 12.1 is sensitive to surface states especially for incident photon energies close to the photoemission threshold. In this limit, only electrons near the surface will have enough kinetic energy to escape because of the strong electron-electron interaction. As discussed above, photoemission provides information on the occupied valence states. 146

12.2.2

Electron Diraction

Another common technique is low energy electron diraction (LEED). This technique is especially sensitive to atomic arrangements on the surface and is analogous to the x-ray diraction techniques that are used to establish the crystal structure in the bulk solid. LEED experiments can be carried out to study the structure of clean surfaces or of adsorbed species on surfaces. The positions of the LEED spots on a photograph establish the periodicity of the intrinsic surface structure. Because of the small penetration depth for low energy electrons (E < 100eV), the LEED technique emphasizes the surface structure. The LEED spot pattern that is formed is due to the constructive interference of reections of the electron beam through scattering of rows of atoms rather than planes of atoms as occur in three-dimensional x-ray diraction illustrated in Fig. 12.8. Since the Bragg law in this case corresponds to rows of atoms, the surface structure that is probed is indexed by a two-dimensional lattice. In many cases the surface structure forms a superlattice relative to the substrate. This rearrangement of the surface atoms takes place because of the dangling bonds, which would otherwise occur at the surface. The rearrangement partially satises the bonding requirements. Within the surface, rows of atoms may move closer together or farther apart, and the surface atoms may move in or out relative to the inner layers of atoms. Such a rearrangement of the surface atoms is called reconstruction. Surface reconstruction is illustrated in Fig. 12.9 for the case of a (110) GaAs surface. The corresponding change in the surface density of states is shown in Fig. 12.10. The surface is highly sensitive to the presence of adsorbed atoms. Figure 12.11 shows the modication to the surface density of states of a (110) surface of GaAs upon exposure to oxygen, as measured in this case by photoemission (see 12.1). The notation used to describe the surface structure is in terms of (1) the length of the lattice vectors in the superlattice relative to those of the substrate and (2) the angle of rotation of the superlattice coordinate system relative to that of the substrate. Illustrated in Fig. 12.12 examples of (2 2)R0 and ( 3 3)R30 superlattices. Note in the case are of the ( 3 3)R30 superlattice that the coordinate system of the superlattice makes an angle of 30 with respect to that of the substrate. The LEED technique is used to study the structure of pristine surfaces. In many cases, the surface structure may be dierent from the bulk structure because the surface atoms 147

0 ) 1(' %# !  &$"

    

Figure 12.8: Bragg condition for x-ray diraction from rows of atoms.

Figure 12.9: A schematic diagram of the electronic and spatial congurations of the GaAs (110) surface. The As atoms have moved outward and the Ga atoms inward compared to the positions in the bulk of the crystal.

Figure 12.10: The local density of surface states (solid line) for the bond relaxation model of the GaAs (110) surface. Electronic states located on the rst two layers of Ga and As atoms are shown.

148

Figure 12.11: Energy distribution curves from the upper part of the valence band of two dierent GaAs (110) crystals. The eects of a small oxygen exposure on the Fermi level pinning and valence band structure are also shown.

Figure 12.12: This gure illustrates two possible commensurate structures for adsorbed atoms on a honeycomb triangular lattice substrate. (a) The (2 2)R0 structure and (b) the ( 3 3)R30 structure.

149

Figure 12.13: Schematic diagram of an MBE system.

have fewer nearest neighbors. In general, one would thus expect the bond lengths (normal to the surface) for the surface atoms to be slightly shorter than in the bulk. In addition, the LEED techniques can be used to determine the structure of adsorbed species in the coverage range from below one monolayer to perhaps two monolayers. The analysis of LEED patterns in general is more complex than for x-ray diraction because of multiple electron scattering phenomena. For many surface structures, other evidence in addition to LEED data is needed to determine the surface structure unambiguously. The standard in situ characterization technique for MBE growth of semiconductor superlattices (see Fig. 12.13) is RHEED (reection high energy electron diractometry). The RHEED measurements are carried out at almost glancing angles of incidence to accentuate the surface sensitivity. The RHEED measurements provide a diraction pattern on a uorescent screen which is used to monitor the growth, providing information on the: structure of the growing surface and smoothness of surface surface reconstruction growth dynamics through observation of intensity oscillations evolution of surface impurities. Some examples of RHEED patterns from the growth of InSb on a CdTe (001) substrate, looking down a [110] direction (glancing angle) are seen in Fig. 12.14. After 5 sec of InSb 150

Figure 12.14: Reection highenergy electron diraction patterns at 20KeV and at a glancing [110] angle characterizing the growth of InSb on CdTe (001): (a) after 5 sec; (b) after 75 sec; (c) after 115 sec; (d) after 160 sec, to yield an InSb quantum well of 160 . (After A L. A. Kolodziejski, Gunshor, Otsuka, Datta, Becker, and Nurmikko, IEEE J. Quantum Electronics, vol. QE22, 1666 (1986)).

151

growth ( 1/sec growth rate), a spotty diraction pattern is seen (Fig. 12.14a), indicative A of the formation of islands of InSb on the CdTe substrate. As the growth proceeds over a 75 sec time period, the islands get large enough to join up and form a smooth surface, yielding the more uniform streak pattern. This streak pattern develops further with increasing deposition time. The diraction patterns for later times (75 sec and beyond) show a central diraction line characteristic of the bulk material, and some weak sidebands characteristic of the surface reconstruction. The surface reconstruction sidebands indicate a surface structure that is also periodic but with a dierent structure than the bulk, due to the relaxation of the solid to tie up the dangling bonds at the surface.

12.2.3

Electron Energy Loss Spectroscopy, EELS

Electron energy loss spectroscopy (EELS or ELS) is another commonly used technique in surface science. In electron energy loss spectroscopy, a primary electron with an energy of perhaps 100 eV will excite an electron in a lled initial state to an empty excited state. The electronic structure of the valence band states is determined by examination of the energy spectrum of the emitted secondary electrons. In the interpretation of these energy loss studies, no correction need be made for the work function for the electron, since the same potential energy drop is experienced for both the primary and secondary electrons at the surface potential barrier. The EELS technique is conceptually the same as Raman scattering (10.4 of notes Part II) or inelastic neutron scattering. An incident electron of energy Ei is scattered by an electron in the solid, imparting (or absorbing) an energy to (from) the electron in the h solid to achieve an energy Ef for the scattered electron using conservation of energy: Ei Ef = . h Likewise momentum is conserved to yield the relation ki kf = q. (12.23) (12.22)

Unlike the case of Raman scattering, the incident electrons in the EELS experiment can have a large range of wave vectors ki so that the change in momentum for the electron in the solid can be comparable to Brillouin zone dimensions. Since the incident electrons typically have energies up to 100eV with wave vectors up to ki 51 , the EELS technique can A probe a wider wave vector range in the E(k) diagram than is commonly probed in an optical reectivity measurement. EELS is dierent from optical absorption and Raman scattering in that it is sensitive to dierent aspects of the electronic structure of solids because the probe is a charged particle rather than a photon. An incident light wave is characterized by its electric eld E. The rate of absorption of the light or the power loss is proportional to Im(E D) Im()|E|2 2 ()|E|2 (12.24)

where the imaginary Im() = 2 () and D = E is the displacement vector while = 1 +i2 is the complex dielectric constant. Optical absorption occurs preferentially at peaks in 2 (). On the other hand, an incident electron sets up a free charge density (r) which determines the displacement vector D through the Maxwell equation D = 4 152 (12.25)

For an incident electron beam, the electron energy loss rate is proportional to 2 1 |D|2 Im(E D) = Im( )|D|2 = 2 1 + 2 2 (12.26)

Thus in the EELS experiment peaks in the energy loss rate occur both near peaks in 2 () (if |1 ()| |2 ()| in this frequency region), and also near zeros in 1 () (where 2 () often remains small), corresponding to longitudinal modes. The most important longitudinal mode is the plasma mode, which is usually a prominent feature in EELS data. The standard EELS technique is limited by the small penetration depth of the electrons relative to that for photons. EELS is thus primarily a surface technique. By using electron beams of higher energy and near normal incidence, EELS can be applied to study the electronic structure of the bulk. To emphasize the electronic structure near the surface, low energy electrons are used at grazing angles of incidence.

12.2.4

Auger Electron Spectroscopy (AES)

To determine the chemical species present on a surface, Auger Electron Spectroscopy (AES) is commonly used. In this technique an electron beam of several keV is incident on the surface. An electron in this primary incident beam will excite an inner core electron, creating a hole in this inner core electron state, which we will label as state . An electron in some higher-lying state in the same atom will quickly fall into the hole state . At the same time, a second electron in a state also in the same atom will be ionized. This second electron will acquire kinetic energy E such that energy is conserved in the total Auger process. Thus in the Auger process (see Fig. 12.15) there are three quantum states involved: , , . What is measured is the energy spectrum of the emitted secondary electrons. In this spectrum, peaks in the intensity prole are identied with specic core states of the atom participating in the Auger process. Since each atomic species has its own characteristic core state spectrum, the Auger spectrum provides an excellent tool for the identication of atomic species. Furthermore, the intensity of the Auger lines provides a measure of the concentration of each atomic species. Because of the strong interaction 153

!







 

Figure 12.15: Schematic diagram of the Auger process. E =E E E

B @ 0 7 5 (2 CA98643( 0( & $ " 1)'%#     

of electrons with matter, electrons will be emitted only from atoms near the surface and for this reason Auger electron spectroscopy preferentially studies the chemical species at or near (within 40) the free surface. Typical Auger electron spectroscopy equipment A contains an Argon ion sputtering gun, permitting the removal of surface atoms so that the Auger experiments can be carried out as a function of depth into the surface by a method called depth proling. Use of the depth proling technique is destructive to the sample, leaving a tiny hole behind. X-ray uorescence measurements are also used to identify the chemical species present in a given species. In the x-ray uorescence technique, x-rays (energies 50keV) are incident on a sample and eject an electron from a core state in the atom of the solid. An electron in a level (see Fig. 12.16) will fall into the state , releasing a photon with energy E E which is measured. The emitted x-rays thus have characteristic energies corresponding to the core level for each atom, thereby allowing identication of the chemical species. The intensity of the characteristic emission lines are related to the concentration of each of the chemical species. X-ray uorescence (see Fig. 12.16) is a non-destructive method for chemical analysis and has a penetration depth of 1m. This technique is usually available when doing scanning electron microscopy (SEM) studies and is called EDX.

12.2.5

EXAFS

Another powerful experimental method for studying solids is Extended X-ray Absorption Fine Structure (EXAFS), which has been applied to the study of surfaces. Figure 12.17 shows a typical x-ray absorption spectrum, in the region of an inner-shell ionization energy. The photon energy at the absorption edge is equal to the minimum of threshold energy required to excite an electron from an inner tightly-bound atomic level to an unbound or continuum state. The region of higher photon energies, which is about 1-2 keV above the absorption edge, is called the EXAFS region; small oscillations in the absorption strength are produced by interference between the wave function for the outgoing electron state and the wave-functions reected from neighboring atoms in the solid or on the surface. (See 154

$#

! "

  

Figure 12.16: Schematic diagram for the x-ray uorescence process.

  HF D IG1 EC A @37531 ) ' B6986420(&

@WG8V6U3 ) 1 A 7 3T D 8S) RP )A Q   

Figure 12.17: Schematic of xray absorption spectrum showing the threshold region (including pre-edge and edge region) and the EXAFS spectrum.

155

EXAFS structure is shown in Fig. 12.17). To analyze the EXAFS region, the nal state of the electron f (r) is written as f (r) = f, (r) + kj SC (r Rj ) (12.27)

where f, (r) is the nal state of the absorption process, centered at the particular atom with angular momentum quantum number and which corresponds to a particular energy E above the threshold, and SC is a reected wave corresponding to the scattering of the electron in state f, (r) from a neighboring atom at position Rj . If one knows, from atomic physics experiments or theory, the phase shifts j corresponding to scattering from each of the neighboring chemical species, then the position Rj of these neighbors can in principle be determined, since the EXAFS amplitude is proportional to: (see a chapter on scattering theory in a quantum mechanics text such as Schi, Chapter 5 or Sakurai, Chapter 7.) (E) | i |H |f |2 F (|Rj |)Im(f ()e2ik|Rj | e2ij ) (12.28)

where i is the initial atomic state, H is the optical perturbation Hamiltonian, F is a 1 smooth function of |Rj |, and k (2mE/ 2 ) 2 is the wave vector of the ejected electrons, h while f () is the amplitude for scattering the ejected electron at 180 , back towards the emitting atom. Because of the factor exp(2ik|Rj |), the EXAFS amplitude is essentially the Fourier transform of the probability distribution of nearest-neighbor separations | Rj |. The Fourier transform of the probability distribution, is then compared to the function calculated for a given model of the structure to test the validity of that model. In Eq. 12.28, f is the wavefunction for the nal state and j is phase shift for an atom at position j. Figure 12.18 gives an example of the power of the EXAFS technique. Here EXAFS spectra are shown for a clean ruthenium surface and for surfaces exposed to two dierent oxidation conditions. The analysis of each trace is shown on the right to extract the pertinent nearest neighbor distances. Equipment to carry out the various surface science experiments mentioned above is very expensive and the techniques are generally useful in many areas of solid state research. Therefore we have a number of experimental systems available through the Central Facilities of the Center for Materials Science and Engineering (Building 13). The LEED, ESCA (Electron Spectroscopy for Chemical Analysis) and AES equipment is in the Surface Analytical Laboratory (4th oor), EELS measurements can be done with the transmission electron microscope (basement 13-1027) and x-ray uorescence measurements can be made with the scanning electron microscope using the KEVEX attachments (2nd oor) or with an electron microprobe (basement level). A range of scanning electron microscopes (SEM), transmission electron microscopes (TEM), and scanning transmission microscopes (STEM) are also available in the Building 13 Central Facilities.

12.2.6

Scanning Tunneling Microscopy

The scanning tunneling microscope (STM) provides a unique and powerful new tool for the direct determination of real space surface structure at the atomic level, including nonperiodic structures. In this microscope, a small metal tip is brought close enough to the surface to permit electron tunneling between the tip and the surface. The tip scans the surface in twodimensions (hence the name scanning tunneling microscope). By adjusting the height 156

Figure 12.18: Extended absorption edge ne structure (EXAFS) is shown in (a) for thinly dispersed pure ruthenium metal, (b) for ruthenium partially covered with O2 at 25 , and (c) for ruthenium mostly converted to RuO2 at 400 C. Curves d, e and f are the Fourier transforms of a, b and c, from which the number and bond distances of nearestneighbor atoms may be derived.

157

Figure 12.19: Schematic of a scanning tunneling microscope and of its operation.

of the tip above the surface to maintain a constant tunneling current, it is possible to obtain a contour map of the surface. The announcement of the successful observation of surface structure on an atomic scale with the STM in 1982 created tremendous excitement in the solid state community. Some useful references and reviews are listed below. IBM Journal of Research and Development 30, #4 & #5 (1986) G. Binnig and H. Rohrer, IBM Journal of Research and Development 30, 355 (1986); Scientic American, August 1985, p. 50. J.E. Demuth, R.J. Hamers, R.M. Tromp and M.E. Welland, IBM Journal of Research and Development 30, 396 (1986) J.E. Demuth, Physics in a Technological World, French, Editor, AIP, NY 1988. Inelastic Electron Tunneling Spectroscopy, T. Wolfram, editor, Springer Series in SolidState Sciences 4, Springer, Berlin (1977). Tunneling in Solids, C.B. Duke, ed. by H. Ehrenreich, F. Seitz, and D. Turnbull, Solid State Physics, Supplement #10 Academic Press, New York (1969). M. Amrein, A. Stasiak, H. Gross, E. Stoll, and G. Travaglini, Science 240, 514, (1988). Referring to Fig. 12.19, the piezodrives Px and Py scan the metal tip over the surface. The control unit (CU) applies the appropriate voltage to the piezodrive P z to maintain a constant tunneling current JT at constant tunnel voltage VT . For a constant work function, the voltages applied to the piezodrives Px , Py , and Pz yield the topography of the surface directly, whereas modulation of the tunnel distance z by z gives a measure of the work function. The dotted line in Fig. 12.19 indicates the z displacement in a y scan at a surface step. The very high resolution of the STM depends on the exponential dependence of the tunneling current on the distance z between the tip and the scanned surface. If is the average barrier height for tunneling (the average work function = ( 1 + 2 )/2 between the tip and the surface), then the tunneling current is given by JT = J0 exp (2z) 158 (12.29)

where 2 2 = 2m h and m is the free electron mass, so that JT = J0 exp (A1/2 z) where A = 2(2m/ 2 )1/2 = 1.025 h and J0 = (e2 / )(/4 2 z). h (12.33) The tunneling probe tip in a typical instrument may be prepared to have a radius between a few thousand angstroms to 1m, but containing some sharp mini-tips close to the atomic limit as shown schematically in Fig. 12.19. The extreme sensitivity of the tunneling current on the gap width selects the extremal mini-tip protuberance ( 10) for operation of the A STM. The very ne tips of the STM can resolve monatomic steps within 10 lateral resolution, A as indicated in Fig. 12.20 for the case of steps on a clean Au (100) surface (G. Binnig, H. Rohrer, Ch. Gerber and E. Stoll, Surface Science 114, 321 (1984)). To give the order of magnitude of the sensitivity of the STM due to the exponential dependence of the current on distance z, an increase in distance of 1 results in an order of magnitude decrease in A tunneling current. From Eq. 12.31, we see that scanning the tunneling tip at constant tunneling current implies that 1/2 z = constant. Thus for constant work function of the substrate, the displacement in the z direction is adjusted to yield z = constant during the scan. The voltage applied to the piezodrive Pz to achieve z = constant thus provides a record of the surface topography along the scan. Figure 12.20 shows scanning tunneling micrographs of a gold (100) surface, exhibiting atomically at terraces with monolayer steps. The tunneling current is sensitive not only to the topographical features but also to the local electronic structure. A good deal of eort has been devoted to disentangling the contributions to the tunneling current from each of these two eects. Electronic and chemical surface properties manifest themselves primarily in the voltage dependence of the tunneling current. They appear as specic features in the local I V , V z or I z characteristics, where z is the distance between the tip and the surface. In practice, electronic or chemical images are obtained by recording dI/dV or dI/dz while scanning and controlling the gap width, and while keeping the average current constant. Depicted in Fig. 12.21 is an example of such scanning tunneling spectroscopic (STS) imaging on a Ni (100) surface (R. Garcia, J.J. Saenz and N. Garcia, Phys. Rev. B33, 4439 (1986)). In trace (a) showing a plot of dI/dV vs V , the strong peak at 0.8 V is attributed to surface nickel oxide. In (b), the surface is imaged with respect to that spectroscopic feature (top) by taking scans of dI/dV at 0.8 V in the y direction. On the bottom of (b), topographic I = constant scans are taken at the indicated voltages. Whereas the STS images dier dramatically, the STM images remain essentially unchanged. Because of the usually close relation of the tiptosample spacing at I = constant with the topography, of dI/dV with the local density of states, and of dI/dz with the local barrier height (or work function), it is possible to separate the various eects observed with STM. Associated images are 159 1 eV1/2 A (12.31) (12.30)

(12.32)

Figure 12.20: STM image of a clean Au (100) surface obtained at a constant tunneling current of 1 nA, showing the terraces and monolayer step lines. The wavy structure can be resolved into individual atomic rows. The divisions on the axes correspond to spacings of 5 (G. Binnig, H. Rohrer, Ch. Gerber and E. Stoll, Surface Science 114, 321 (1984)). A

160

Figure 12.21: Spectroscopic and structural imaging of NiO on a Ni (100) surface. Shown in (a) is dI/dV vs. V from an oxidecovered region. The strong peak at 0.8 V is characteristic of NiO. The STS and STM images shown in (b) were obtained at the indicated bias voltages of 0.8 and 1.3 V. An oxide island to the left is evident in the STS image obtained at 0.8 V. Spatial separations in units of the NiO lattice spacing are indicated at the left (bottom). The oxide island is hardly noticeable in the STM images. The divisions on the y axis correspond to spacings of 5 (R. Garcia, A J.J. Saenz and N. Garcia, Phys. Rev. B33, 4439 (1986)).

161

Figure 12.22: STM relief of a (111) Si surface showing the (7 7) unit cells with a superposed model for the (7 7) structure. The diagram covers a scan area of 6060. A A Original work by G. Binnig, H. Rohrer, Ch. Gerber and E. Weibel, Phys. Rev. Lett. 50, 120 (1983).

often referred to as topographical or STM images, spectroscopic or STS images, and workfunction proles, respectively. An important aspect of the STM is its apparent nondestructive nature. In normal operation, no perceptible irreversible damage to the sample surface occurs as a result of its use. On the other hand, the STM can be used to intentionally induce permanent local structural or chemical modications. One of the most noteworthy achievements of the STM has been the direct observation of surface reconstruction in silicon (G. Binnig, H. Rohrer, Ch. Gerber and E. Weibel, Phys. Rev. Lett. 50, 120 (1983); G. Binnig and H. Rohrer, IBM Journal of Research and Development 30, 355 (1986); J.E. Demuth, R.J. Hamers, R.M. Tromp and M.E. Welland, IBM Journal of Research and Development 30, 396 (1986)). Surface reconstruction pertains to a dierent surface structure relative to that of the bulk due to the broken bonds at the surface discontinuity. It has been known for some time that the surface structure of semiconductors diers from that of the bulk because of the dierent number of nearest neighbors available for bonding. It had been conjectured that the (111) surface of Si when heated above 900 C exhibits a (7 7) surface reconstruction, though many uncertainties remained about whether the (7 7) reconstruction was correct and where the atoms were located within the unit cell. With the STM, the (7 7) surface reconstruction has been vividly demonstrated, as shown in Fig. 12.22. This picture of the (7 7) reconstruction was obtained on a sample previously heated at 900 C in high vacuum to remove any surface oxide layer. The STM micrograph was taken at a 2.9 V positive tip potential. The (7 7) rhombohedral unit cell is clearly seen in the scan in Fig. 12.22, bounded by lines on minima with deep corners. Each unit cell contains 12 maxima and the diagonals are determined to be 46 1 and 29 4 , in good agreement with the crystallographically determined A A and 26.88 , respectively. values of 46.56 A A 162

Figure 12.23: Schematic diagram of the Atomic Force Microscope (AFM). An STM is used to measure the displacements of the probe tip as it scans the surface of an insulating sample.

Figure 12.24: AFM images of grooves etched in Si for three increased magnications. The highest magnication scans show the details of the groove substructure which is only 50 A high. (Y. Martin, C.C. Williams, and H.K. Wickramasinghe, J. Appl. Phys. 61, 4723 (1987)). Following the successful development of the scanning tunneling microscope, several related instruments have emerged. One of the more important of these instruments is the atomic force microscope (AFM). The forces measured by this instrument are the interatomic forces between the surface atoms under investigation and the apex atoms of a very sharp diamond tip xed on a conducting cantilever (see Fig. 12.23). Bending of the cantilever by the interatomic forces is monitored by the tunneling current between the cantilever and an STM tip. Scanning the diamond tip across the conducting or insulating surface under investigation at constant interatomic force yields a topographical image of the surface. The AFM image is composed of contours of constant force between the imaged surface and a probe tip, and permits measurement of surface contours of both conductors and insulators; the present versions of the STM are not able to make topographical maps of insulators. Although atomic resolution has been achieved on graphite surfaces, this technique will be more widely exploited in the study of larger-scale features on real surfaces under normal laboratory conditions. Recent work at IBM (see Fig. 12.24) shows an application of the AFM to examine the groves in a reactive-ion-etched silicon wafer after oxidation in air. The dierent microstructures on top of and in the grooves are associated with the etching method and are clearly resolved by the AFM over a wide range of scales, even though the A surface is insulating. At the highest resolution, features as small as 50 high are easily 163

seen.

164

Chapter 13

Amorphous Semiconductors
References
Kittel, Introduction to Solid State Physics Ziman, Models of Disorder, Cambridge, 1979.

13.1

Introduction

There are many materials which are of scientic and technological interest which are not single crystals, or even microcrystalline. The general category of amorphous materials is dened as including materials which have no crystalline order; that is, their xray diraction patterns consist of thick diuse rings or halos instead of sharp spots. Usually excluded from this denition are polycrystalline materials which consist of small crystallites in random orientations. Since, as we shall discuss below, amorphous materials often do possess considerable shortrange order, the distinction between polycrystalline and amorphous materials blurs as the crystallite size is reduced. There are a number of substances, such as Ge and Si, which can be prepared in all three forms: singlecrystal, polycrystalline and amorphous. Amorphous materials can be prepared in several ways, depending on the material. Materials called glasses can be formed by cooling from the liquid state. These materials are not in thermal equilibrium and can be classied as super-cooled liquids with an extremely high viscosity. Some materials cannot be cooled fast enough to avoid crystallization. Splat cooling (or rapid solidication) is often used to prepare metallic glasses. In this process a liquid stream is shot onto a thermally conducting substrate. Some of the splatcooled metals exhibit microcrystalline ordering. There are also a number of techniques for deposition from the vapor state, such as sputtering, or for condensation from a chemically reactive vapor such as silane (SiH4 ) to prepare (hydrogenated) amorphous Si. Many amorphous materials can be called semiconductors in the sense that they are neither good conductors nor good insulators, but instead are poor conductors. Many are also similar to their crystalline counterparts in that they possess an optical gap. The reason for this general behavior seems to be that, even though the amorphous structure is quite random over long distances, there still seems to be considerable shortrange order with local bonding requirements generally satised. Although these materials are full of defects, 165

Figure 13.1: Distribution of neighbor distances G(r) for Ge, from EXAFS data, shown for both crystalline and amorphous germanium.

there are not many electrons which are involved in the process to carry electric current. In addition, because of the spatial disorder resulting in strong carrier scattering, the carriers mobilities are low. A further consequence of the tendency for bonding to be satised is that the electrical properties of many amorphous semiconductors tend to be insensitive to the presence of large concentrations of impurities. There are two major categories of amorphous semiconductors. The rst consists of the tetrahedrally bonded materials, primarily Ge and Si but also including amorphous III V semiconductors. The second major category consists of the chalcogenide or lonepair semiconductors, which means the elements Se, S or Te (column VI) and compounds and alloys containing these elements.

13.1.1

Structure of Amorphous Semiconductors

The major attribute of the structure of amorphous materials is the lack of longrange order or the absence of a periodic lattice. In spite of this, there is considerable similarity in the local environments of amorphous and crystalline materials. For example, the EXAFS measurements for Ge shown in Fig. 13.1 indicate that the rst and secondnearestneighbor distances are the same and that dierences only appear at the thirdand higherneighbor distances. Not only are these shortrange bond lengths usually preserved, but also bond angles tend to be the same. Some amorphous materials have been modeled successfully as continuously 166

Figure 13.2: The diamond structure showing the atoms 1 and 2 are in a staggered conguration and are crystallographically distinct in the perfect crystal.

perturbed from their crystalline form. However there is also some evidence from xray diraction for more drastic changes. For example, one can view crystalline germanium or silicon as containing distorted 6fold rings (see Fig. 13.2 for the diamond structure). Amorphous germanium and silicon also seem to contain some 5fold rings.

13.1.2

Electronic States

It is evident that Blochs theorem no longer holds in amorphous materials; hence electronic states cannot be characterized by a k vector conned to a single Brillouin zone. In other words, k is no longer a good quantum number. Thus one can no longer use the powerful energyband theory which predicts bands of extended electronic states with forbidden gaps, and which we use to dierentiate conductors from insulators. In practice the electronic properties of an amorphous material do not dier as drastically as one might expect from those of the crystalline material. Consider the data for the resistivity of a number of materials at temperatures near their melting points, as shown in Fig. 13.3. Where small jumps do occur at the melting point, these are correlated with small discontinuous volume changes. When the volume does not change, as for HgTe or CdTe, neither does the resistivity. Evidently the conduction properties are remarkably similar even though the structure has changed drastically. We note here that for the column IV semiconductors (Si, Ge, Sn), the molten material is metallic and octahedrally coordinated. Without the powerful simplication of Blochs theorem, it is extremely dicult to calculate the electronic states in amorphous materials. Without a kvector one cannot calculate E(k). Instead one attempts to calculate directly a density of states (E), and also an average energydependent mobility (E). One starts by noting the similarity of amorphous and crystalline electronic properties and especially their similarity with respect to short range order. Recently much progress has been made using various types of cluster models. The general result, as shown in Fig. 13.4, is that the crystalline density of states, which has sharp features called Van Hove singularities at critical points where E/ k = 0, is smoothed out and broadened at the critical points (see 4.2). This is shown experimentally 167

Figure 13.3: Resistivity as a function of temperature for several materials in both the crystalline and liquid states. The melting temperature in each case is indicated by an arrow.

Figure 13.4: (a) Density of electronic states as a function of energy for a single band of a crystalline solid. The sharp behavior at the band edges and in the interior represents the eects of Van Hove singularities. (b) Density of electronic states as a function of energy for a single band of an amorphous solid. All Van Hove singularities have disappeared. (The Van Hove singularities are the M0 , M1 , M2 , M3 singularities in the joint density of states discussed in 4.2).)

168

Figure 13.5: Ultraviolet photoemission results (top) for the density of states for trigonal (solid line) and amorphous (dashed line) Se. Photoemission results (bottom) on trigonal (solid line) and amorphous (dashed line) Te.

in the photoemission results in Fig. 13.5 for the valence bands of trigonal (crystalline) and amorphous Se and Te. In order to nd the transport properties of an amorphous material, one needs to know not only the density of states (E) at each energy E, but also the mobility (E) which may be a function of (E). Broadly, one distinguishes extended states, similar to states in periodic crystals which have nite amplitude throughout the material, and nite mobility, from localized states which have a signicant amplitude only in a small region of the material and have extremely small mobility. An example of such a localized state, in a nearly perfect crystal, is an impurity state in which the electron is localized in a hydrogenlike orbit around a donor ion. Such states have sharp features in a densityofstates diagram near the conductionband minimum or, for acceptors near, the valenceband maximum, as shown in Fig. 13.6(b). If there is a large enough impurity concentration, these states can broaden into impurity bands which can merge into the conduction or valence bands. However, if the material itself is disordered or amorphous, the states near the band edges are themselves localized. This subject of localized vs. extended states has been treated extensively, most notably by Mott and by Anderson who shared the Nobel Prize in Physics in 1977 for this work. Mott developed the concept of the mobility edge, postulating a relatively sharp demarcation between localized and extended states, giving rise to a mobility gap which is considerably larger than the forbidden gap in the density of states, as illustrated in Fig. 13.7. The mechanism for localization in states near the energy band edges is illustrated in Fig. 13.8, and is due to Fritzsche. According to this model disorder produces a spatial variation in the conduction and valence band edges, giving rise to local conductionband minima or valenceband maxima which trap electrons or holes. 169

Figure 13.6: Density of states n(E) for (a) a perfect crystal, (b) a crystal with a few donors and acceptors, and (c) a crystal with a larger number of imperfections where the impurity levels have broadened into impurity bands.

Figure 13.7: Sketch of the MottCFO (CohenFritzscheOvshinsky) model for covalent disordered semiconductors having a threedimensional crosslinked network structure. The critical energies Ec and Ev dene the mobility gap. For T > 0, the mobility (E) may be nite in the gap because of thermally assisted tunneling. Here EF denotes the Fermi energy. The distribution of localized gap states may be non monotonic when defect states of a certain energy are prevalent.

170

Figure 13.8: Potential uctuations of the initial and nal electron states for the optical transitions corresponding to the optical gap E0 . The left hand side shows the density of states. The region of localized states lies between Ec and Ev . Note that the short range potential wells which give rise to many of the localized states are not shown here. This gure shows only that part of the long wavelength potential uctuations which cause a parallel shift of the valence and conduction band states. The part which causes a spatial variation of E0 is omitted for clarity (after Fritzsche). An important dierence between the tetrahedrallybonded amorphous semiconductors and the chalcogenide materials is that the former have large numbers of unpaired spins, as observed in electron spin resonance experiments, and the chalcogenides have no measurable density of spins. A large number of unpaired spins is expected in a material containing a large number of broken or dangling bonds, each bond being occupied only by one electron instead of two electrons of opposite spin. The lack of unpaired spins has been explained by the valencealternationpair model of Kastner, Adler and and Fritzsche. The chalcogen atoms have 4 electrons in an outer (unlled) p shell. In the lowestenergy bonding conguration, two of these electrons form bonds with neighboring atoms, and two are in nonbonding or lonepair states. Thus both crystalline and amorphous Se, for example, contain chains of atoms, each bonded to two neighbors. In the amorphous material a Se atom can also be triplybonded in a trigonal conguration, leaving the fourth p electron in a higherenergy, nonbonding state (with unpaired spin). However, the total energy can be reduced if this extra electron migrates to another triplybonded Se atom nearby. First, two of the bonds on this second atom break, leaving only one electron in a bonding orbital. The two electrons from the broken bonds join the single electron in the lonepair orbitals. Then the new electron can enter this atom as a fourth lonepair electron. Kastner, Adler and Fritzsche argued that this is an energetically favorable conguration. The result, as shown in Fig. 13.9, is a large density of equal number of positive and negative ions but with all electron spins paired. Structurally, this picture implies that chalcogenide glasses contain a large number of linked chains (by triplybonded atoms) as well as nearby broken chains (ending in singlybonded atoms), which provides an explanation for the fact that these materials are more resistant to crystallization (better glasses) than the tetrahedrally bonded materials. At the left of Fig. 13.9 two selenium atoms, each of which is triply bonded, serve to 171

Figure 13.9: Valencealternation pairs can form in a neutral chalcogenideglass matrix without any major displacement of its atoms, leading to a sharp reduction in total energy. crosslink two molecular chains of doubly bonded atoms. A valencealternation pair can be produced (right) by a spontaneous break of the crosslinkage, combined with the simultaneous transfer of an electron from one of the triply bonded selenium atoms to an atom near the one where the crosslink was broken. Since such electronic transfers reduce the total energy of the solid, nearly all the trigonally bonded selenium atoms become members of a valencealternation pair. Important physical consequences follow, including the almost complete disappearance of electrons with unpaired spins and the appearance of large but equal concentrations of positively and negatively charged traps in chalcogenide glasses. The consequences of this high disorder (in the chalcogenides) for the electronic density of states is shown in Fig. 13.10. On the left is shown the broadened conduction band and valence bands, with the mobility edges, for an amorphous tetrahedrally bonded semiconductor. Cohen, Ovshinsky and Fritzsche postulated that (whatever the details of the structure), the high disorder in the chalcogenides produces overlapping densities of states, as shown in the right side of the gure, so that electrons will lower their energy by migrating to new localized states, creating large charge separation as in the valencealternation model. The distinguishing feature of the bands in amorphous solids is the replacement of the sharp band edges present in crystals by what are called band tails or localized states, that extend into the energy gap. The localized states are separated from the extended states in the main part of the bands by mobility edges. The region that lies between the mobility edges of the valence and conduction bands is the mobility gap (see Fig. 13.10). It plays the same role in amorphous semiconductors that the energy gap plays in crystalline semiconductors. Chemical impurities or defects in the conguration of local bands can lead to sharp structural changes (not shown) in the mobility gap. The result of the large density of localized states in the mobility gap is a high density of positively and negatively charged traps, which decrease the mobility of the carriers and make the material less sensitive to eorts to control its conductivity by doping. This density of states model explains the fact that the chalcogenide glasses are much less sensitive to doping than the tetrahedrally coordinated materials. Impurity states, in172

Figure 13.10: Amorphous semiconductors that are not strongly disordered (left) have valence and conduction bands similar to those in the corresponding crystalline semiconductor. If the disorder is large, as is expected in multicomponent glasses (right), the band tails of the valence and conduction bands can overlap in the mobility gap. This leads to a redistribution of electric charge as electrons move from one localized state to another in order to lower their energy. troduced in or just outside the band tails, make only a negligible change in the already appreciable density of states in this region. If, on the other hand, the bands do not overlap, impurity states in or just outside the band tails can make a large change in the density of states. Thus they can, at nite temperatures, become a source of carriers in the conducting or extended states.

13.1.3

Optical Properties

Amorphous semiconductors have optical spectra similar to their crystalline counterparts in that they possess an optical gap or absorption edge. However, all sharp features, including the band edge absorption, are considerably broadened, as shown for the case of the chalcogenide semiconductor As2 S3 in Fig. 13.11. The reectance spectra for crystalline, amorphous and liquid Ge are given in Fig. 13.12. These data show that the amorphous material more closely resembles the crystalline material than the liquid which shows metallic behavior at low frequencies. As was the case for the photoemission measurements, the broadened optical spectra result from the broadened density of states for the amorphous materials. In fact, since k is no longer a good quantum number, one can expect that transitions would be allowed between any pair of valence and conduction band states. The absorption coecient in this picture is proportional to () = (const/) dE (E)(E + )| M (E) |2 h (13.1)

where (E) is the density of states, is the optical frequency, and M (E) is a generalized momentum matrix element. Mott has argued that M = 1 for transitions involving two extended states and for one extended and one localized state, but that M is negligible for transitions involving two localized states (which will have negligible spatial overlap). Thus 173

Figure 13.11: Absorption edge of crystalline cAs2 S3 for 2 directions of light polarization relative to the caxis compared with the absorption edge of amorphous a As2 S3 .

Figure 13.12: Fundamental reection spectra due to electronic transitions in crystalline, amorphous and liquid Ge. The results are consistent with the metallic transport properties of liquid Ge.

174

the optical gap should be at a dierent energy = Eopt than the mobility edge. Mott h argued that the density of states (E) (E E0 ) is a linear function of E near the band edges E0 , giving () = const ( Eopt )2 h . (13.2)

Unfortunately this argument is not of general validity, but is nevertheless used to estimate 1 Eopt by plotting ( ) 2 vs. and extrapolating the straightline behavior to zero frequency. h The amorphous chalcogenides exhibit remarkable luminescence behavior, with a large shift to lower energy of the luminescence peak relative to Eopt , by as much as 1/2 Eopt . This is attributed to a large electronlattice interaction: the excited state of the optical transition produces an atomic or bond rearrangement sucient to cause a large shift in the energies of both the excited and ground states.

13.1.4

Transport Properties

Transport measurements on amorphous semiconductors have proved dicult to interpret, partly because of dierences in the measured transport results arising from dierences in methods of sample preparation. Because of the low mobilities, Hall data have been dicult to obtain, and thermopower data have been dicult to interpret. Attempts to measure mobilities using transitiontime methods resulted in the discovery of nondispersive transport: a pocket of charge injected at one side of the sample does not propagate to the other side with xed velocity but instead spreads out in time because of the large number of traps which have a large distribution of release times. These measurements have been exploited by Professor Kastners group at MIT to give new data on the electronic density of states in amorphous As2 Se3 .

13.1.5

Applications of Amorphous Semiconductors

The most successful application of amorphous semiconductors has been the use of amorphous Se lms for Xerography. In this process, one surface of the lm is charged. When light reected from the white area of the original page strikes the Se lm, electronhole pairs are formed, which then migrate to the surface and neutralize the charge. These areas on the Se lm do not attract the small charged black toner particles, resulting in areas which do not print (white on the copy). The unaected (black) areas retain their charge, do attract toner, and do print on the copy. Another important application is for solar cells and thin lm transistors, where the substantial reduction in cost of producing large areas of amorphous rather than crystalline lms has the potential to oset their lower eciency. An eect which caused some excitement several years ago was the observation by Ovshinsky and others of reversible switching behavior in chalcogenide semiconductors. This has been shown to be due to an electronic mechanism, the lling of traps with carriers above a threshold current, producing a sharp drop in the resistance. A second type of switching is associated with the formation of small crystalline regions. These eects have been exploited in computer memory devices. 175

13.2

Amorphous Semiconductor Superlattices.

The extraction of quantitative information from the study of amorphous semiconductor superlattices oers considerable challenge, because the number of variables is large (band osets, masses, band gaps, mobility edges, chemistry, etc.) On the other hand, superlattices introduce one element of order (zaxis periodicity) in an otherwise disordered system; the superlattice periodicity may perhaps be exploited to learn new physics about this class of materials. Early achievements in the eld of amorphous semiconductor superlattices (B. Abeles and T. Tiedje, Phys. Rev. Lett. 51, 2003 (1983)) indicated that superlattices can be synthesized with alternate layers of amorphous semiconductors such as aSi:H, aGe:H, aSiN x :H and aSi1x Cx :H where the a denotes amorphous and the :H denotes the addition of hydrogen to tie up the dangling bonds in the amorphous semiconductor. Unlike the case in crystalline materials, lattice matching is not an issue in the synthesis of amorphous semiconductor superlattices. The amorphous superlattices are prepared by a plasmaassisted chemical vapor deposition (CVD) method in which the composition of the reactive gases is changed periodically in the reaction chamber. This process has some similarities to the MOCVD technique discussed in connection with heterojunction crystalline semiconductor superlattices. The plasma assisted technique allows deposition to occur at lower substrate temperatures, thereby achieving sharper interfaces. The amorphous lms can be deposited on quartz substrates. The residence time of the gases in the reactor (SiH4 for preparing aSi:H, and SiH4 + NH3 for preparing aSi1x Nx :H) can be as short as 1 sec while the time to grow a monolayer is 3 sec. (see Fig. 13.13). Thus the gases in the reactor can be exchanged rapidly enough to achieve sharp interfaces. The plasma discharge is maintained continuously while the gases are changed. The superlattice periodicity is monitored during the growth process, and the periodicity is conrmed after the lm is deposited by xray diraction, as shown in Fig. 13.13. Because of the random atomic arrangements in the layer planes of the two constituents aSi:H and aSi1x Nx :H, there is no periodicity within the layers d1 and d2 , so that the only periodicity found with the xray characterization experiment is that due to the periodicity d = d 1 + d2 . From the width of the xray peaks, the authors deduce an rms uctuation in the layer thickness of d 5 where d = 41 + 27 = 68 . A A Although no direct observation has been made of bound states in the quantum wells of amorphous semiconductors, the optical absorption measurements of Fig. 13.14 show an increase in the optical bandgap Eg as the quantum well width decreases, where the optical gap was determined from the energy dependence of the absorption coecient using the relation ( Eg )2 /, which normally is valid for bulk amorphous semiconductors. h The temperature dependence of the photoluminescence of the amorphous semiconductor superlattice is similar to that of the bulk, showing an exp(T /T0 ) dependence, except that for the superlattice the characteristic temperature, T0 increases as the width of the quantum well decreases, as shown in Fig. 13.15. Also, the energy width E0 of the localized state distribution in the Urbach tail for amorphous semiconductors shows a similar increase as the width of the quantum well decreases (see Fig. 13.15), indicating that the distribution of localized states broadens as the layer thickness decreases. It is expected that the superlattices will have a small eect on localized states that are deep in the band tail, but 176

Figure 13.13: Xray (1.54 ) diraction A pattern vs. scattering angle (lower scale) and d spacing (upper scale) of a aSi:H (41 )/aSiNx :H (27 ) superlattice with 41 A A periods on a quartz substrate. The inset shows the energyband diagram assumed for the superlattice. The conduction band oset U = 1.05 eV is indicated on the gure.

Figure 13.14: Opticalabsorption coecient vs. photon energy E for aSi:H/a SiNx :H superlattices with varying aSi:H layer thickness L and a constant aSiNx :H layer thickness of 27 . Also given in the A gure is the absorption coecient for a SiNx :H lms prepared under the same conditions.

177

Figure 13.15: Dependence of the optical gap on the aSi:H sublayer thickness L with the aSiNx :H thickness held xed at 27. The solid line is a calculated curve A based on the band diagram in Fig. 13.13, assuming eective masses of unity. Dependence of the Urbach slope parameter E0 (left scale) and photoluminescence quenching parameter T0 (right side) on the aSi:H layer thickness.

a large eect on the energy of the weakly localized shallow states.

178

Appendix A

Time Dependent Perturbation Theory

References
Eisberg, Fundamentals of Modern Physics, Ch. 9 Schi, Quantum Mechanics, Ch. 8

A.1

General Formulation

To proceed further with the formal development of the optical properties of solids, we need to consider how to handle the eect of time-dependent electromagnetic elds quantum mechanically. The most important case of interest is the one where the external eld is a sinusoidal function of time. For most practical applications, the external elds are suciently weak, so that their eect can be handled within the framework of perturbation theory. If the perturbation has an explicit time dependence, it must be handled by timedependent perturbation theory. Practical problems which are handled by time-dependent perturbation theory include such subjects as magnetic resonance (nuclear and electronic spin), cyclotron resonance and optical properties of solids. We give here a brief review of the subject. In doing time-dependent perturbation theory we write the total Hamiltonian H as: H = H0 + H (t) (A.1) where H0 is the unperturbed Hamiltonian and H (t) is the time-dependent perturbation. We assume here that we know how to solve the unperturbed time independent problem for its eigenvalues En and corresponding eigenfunctions un . H 0 un = E n un . (A.2)

Since H (t) has an explicit time dependence, then energy is no longer a constant of the motion. Since we no longer have stationary, time-independent solutions, we must use the time-dependent form of Schrdingers equation, which is: o i h = H = (H0 + H ). t 179 (A.3)

Now, if we didnt have the perturbation term H (t) to contend with, we would set
h (r, t) = un (r)eiEn t/

(A.4)

where un (r) is independent of time and satises Eq. A.2. Thus all the time dependence of h (r, t) is contained in the phase factor eiEn t/ . For H (t) = 0, it immediately follows that i h = En t (A.5)

which yields the time-independent Schrdinger equation. With the perturbation present, o h we expand the time dependent functions (r, t) in terms of the complete set un (r)eiEn t/ (r, t) =
n h an (t)un (r) eiEn t/

(A.6)

where the an (t) are the time-dependent expansion coecients. Substituting Eq. A.6 in the time-dependent Schrdinger equation (Eq. A.3) we obtain: o i h n an (t)un e
iEn t/ h

n an (t)un En e

iEn t/ = h

n an (t)[H0 n an (t)[En

h + H (t)]un eiEn t/

(A.7)

+ H (t)]un

h eiEn t/

where an (t) denotes the time derivative dan (t)/dt. We note that because of Eq. A.2 the second term on the left hand side of Eq. A.7 is canceled by the rst term on the right hand side. We now multiply on the left hand side of Eq. A.7 by u (r) and integrate over all space. k If we make use of the orthogonality of the eigenfunctions u (r)un (r)d3 r = n,k k we obtain from Eq. A.7 i h
n h an (t)un eiEn t/ = n h an (t)H (t)un eiEn t/

(A.8)

(A.9)

the result:
h i ak eiEk t/ = h n h an k|H (t)|n eiEn t/

(A.10)

where we have written the matrix element k|H (t)|n = u (r)H (t)un (r)d3 r. k (A.11)

Since H (t) is time-dependent, so is the matrix element time-dependent, even though, the matrix element is taken between stationary states. We thus obtain the result i ak (t) = h
n h an (t) k|H (t)|n ei(Ek En )t/ .

(A.12)

If we set kn = Ek En h 180 (A.13)

where kn is the Bohr frequency between states k and n, we have ak (t) = 1 i h an (t)eikn t k|H (t)|n
n

(A.14)

in which the indicated matrix element is taken between eigenstates of the unperturbed Hamiltonian H0 . So far, no perturbation theory has been used and the result given in Eq. A.14 is exact. We notice that the unperturbed Hamiltonian is completely absent from Eq. A.14. Nevertheless, its energy eigenvalues appear in kn and its eigenfunctions in the matrix element k|H (t)|n . In applying perturbation theory, we consider the matrix element k|H(t)|n to be small, and we write each time-dependent amplitude as an expansion in perturbation theory

an = a(0) + a(1) + a(2) + = n n n

a(i) n
i=0 (0)

(A.15)

where the superscript gives the order of the term. Thus an is the zeroth order term and (i) an is the ith order correction to an . From Eq. A.14, we see that ak (t) changes its value with time only because of the time dependent perturbation. Thus, the unperturbed situation (0th order perturbation theory) must give no time dependence in zeroth order a(0) = 0 m and the rst order correction yields: a(1) = 1/i m h
n

(A.16)

a(0) m|H (t)|n eimn t . n

(A.17)

In the application of perturbation theory we assume, for example, that if we start in an (0) eigenstate n = , only the coecient a will be appreciably large. Then all other terms in the sum can be neglected. This gives us in 1st order perturbation theory: a(1) = m
(0)

1 (0) a m|H | eim i h

(A.18)

where a is approximately unity. For many cases of interest, this integration over the time variable can be performed and (1) am rather than its time derivative is obtained. The two simple cases that can be integrated easily are: 1. The perturbation H is constant but is turned on at some time (t = 0) and we look at the amplitudes of the wave function in the various states after the perturbation has been acting for some time t > 0. 2. The perturbation H has a sinusoidal time dependence with frequency . This is the situation for all resonant phenomena. Let us rst consider case (1). Then a(1) (t) = m 1 i h
t 0

m|H | eim t dt = 181

m|H | i h

[eim t 1] . im

(A.19)

Figure A.1: Plot of sin2 ( t/2)/ 2 vs. , a function which enters the calculation of timedependent perturbation problems.

Similarly, for case (2), we can write H (t) = H (0)eit (A.20)

to show the explicit time dependence, so that upon integration we obtain for the amplitudes (1) am (t) a(1) (t) = m 1 m|H (0)| i h
t 0

ei(m

)t

dt =

1 ei(m )t 1 m|H (0)| . i h i(m )

(1)

(A.21)

We interpret the time dependent amplitudes |am (t)|2 as the probability of nding the system in a state m after a time t has elapsed since the perturbation was applied; the system was initially in a state = m. We thus obtain for case (1) given by Eq. A.19 |a(1) (t)|2 = m |a(1) (t)|2 = m | m|H | |2 2 h | m|H | |2 2 h |eim t 1|2 2 m 4 sin2 (m t/2) 2 m (A.22)

(A.23)

Clearly for case (2), the same result follows except that m is replaced by (m ) where is the applied frequency and a resonant denominator results for the transition probability amplitude. It is clear from the above arguments that for both cases (1) and (2), the explicit time dependence is contained in an oscillatory term of the form [sin2 ( t/2)/ 2 ] where = m for the case (1) and = m for case (2). This function was previously encountered in diraction theory and looks like that shown in Fig. A.1. Of special interest here is the fact that the main contribution to this function comes for 0, with the height = 182

of the main peak proportional to t2 /4 and the width proportional to 1/t. This means that the area under the central peak goes as t. If we think of |am (t)|2 as the probability of nding the system in a state m, then for case (2), where we have a perturbation with frequency , the system attempts to make a transition from a state to a state m with a transition probability proportional to the time the perturbation acts. If we then wait long enough, a system in an energy state will make a transition to a state m, if photons of the resonant frequency m are present.

A.2

Fermi Golden Rule

Since the transition probability is proportional to the time the perturbation acts, it is therefore useful to deal with a quantity called the transition probability per unit time and the relation giving this quantity is called the Golden Rule (named by Fermi and often called Fermis Golden Rule). In deriving the Golden Rule from Eq. A.19, we must consider the system exposed to the perturbation for a time suciently long so that we can make a meaningful measurement within the framework of the Heisenberg uncertainty principle: Et h (A.24)

so that the uncertainty in energy (or frequency) during the time that the perturbation acts is E h/t (A.25) or
m

2/t.

(A.26)

But this is precisely the period of the oscillatory function shown in Fig. A.1. In this context, we must think of the concept of transition probability/unit time as encompassing a range of energies and times consistent with the uncertainty principle. In the case of solids, it is quite natural to do this anyhow, because the wave vector k is a quasi-continuous variable. That is, there are a large number of k states which have energies close to a given energy. The quantum states labeled by wave vector k are close together in a solid having about 1022 atoms/cm3 . Since the photon source itself has a bandwidth, we would automatically want to consider a range of energy dierences . From this point of view, we introduce h the transition probability/unit time Wm for making a transition to a state m Wm = 1 (1) | a (t) |2 t m m m (A.27)

where the summation is carried out over a range of energy states consistent with the uncertainty principle; mm 2/t. (1) Substituting for |am (t)|2 from Eq. A.23, we have |a(1) (t)|2 = m 4| m|H | |2 2 h 183 sin2 ( t/2) 2 (A.28)

and the summation is replaced by an integration over a narrow energy range weighted by the density of states (Em ) which gives the number of states per unit energy range. We thus obtain 1 sin2 (m t/2) Wm = 2 |4Hm |2 (A.29) (Em ) dEm 2 m t h where we have written Hm for the matrix element m |H | . But, by hypothesis, we are only considering energies within a small energy range Em around Em and over this range the matrix elements and density of nal states will not be varying. However, the function [sin2 ( t/2)/ 2 ] will be varying rapidly, as can be seen from Fig. A.1. Therefore, it is adequate to integrate Eq. A.29 only over the rapidly varying function [sin 2 (t/2)]/ 2 . Writing dE = d , we obtain; h Wm 4|Hm |2 (Em ) t 2 h sin2 2 t d . 2 (A.30)

The most important contribution to the integral in Eq. A.30 comes from values of close to . On the other hand, we know how to do this integral between and +, since

(sin2 x/x2 )dx = .

(A.31)

Therefore we can write an approximate relation from Eq. A.30 by setting x = t/2 Wm (2/ )|Hm |2 (Em ) h = (A.32)

which is often called Fermis Golden Rule. In the subsequent sections, we will apply the Fermi Golden Rule to calculate the optical properties of solids. If the initial state is a discrete level (such as donor impurity level) and the nal state is a continuum (such as conduction band), then the Fermi Golden Rule (Eq. A.32) as written yields the transition probability per unit time and (Em ) is interpreted as the density of nal states. Likewise if the nal state is discrete and the initial state is a continuum, W m also gives the transition probability per unit time, only in this case (Em ) is interpreted as the density of initial states. For many important applications in solid state physics, the transitions of interest are between a continuum of initial states and a continuum of nal states. In this case the Fermi Golden Rule must be interpreted in terms of a joint density of states, whereby the initial and nal states are separated by the photon energy inducing the transition. These issues h are discussed in Chapter 4.

A.3

Time Dependent 2nd Order Perturbation Theory

H = H0 + H (A.33)

This second order treatment is needed for indirect optical transitions, where

and 1. Here H0 0 = i 0 /t. Expand , the solution to Eq. A.33 in terms of the h complete set of functions denoted by 0 |n, k =
n,k
an (k, t)e h En (k)t |n, k i

(A.34)

184

where |n, k is a Bloch function describing the eigenstates of the unperturbed problem H = i h
nk
an (k, t)En (k)e h En (k)t |n, k + i h En (k)t i

(A.35)
i

n,k

an (k, t)e h En (k)t H |n, k i h En (k)t

(A.36)

= i h which gives

n,k

an (k, t)e

|n, k +
i

n,k

an (k, t)e

|n, k (A.37)

am (k , t) = We expand

1 i h

n,k

an (k, t)e h (Em (k )tEn (k)t) m, k |H |n, k

am (k , t) = a(0) + a(1) + 2 a(2) + . . . m m m and let aj (k, 0) = 1, and all others an (k, 0) = 0 where n = j. To rst order, as before, 2 a(1) (k , t) = m or a(1) (k , t) = m To second order 2 a(2) (k , t) = m or am (k , t) = 12 h
t 0 dt (2) n,k

(A.38)

i 1 (0) an (k, t) exp [Em (k ) En (k)]t m, k |H |n, k i h h 1 i h

t

(A.39)

dt exp
0

i [Em (k ) En (k)]t m, k |H |n, k h i [Em (k ) En (k)]t m, k |H |n, k h

(A.40)

1 i h

a(1) (k, t) exp n

n,k

(A.41)

an (k, t) exp

(1)

i [Em (k h

) En (k)]t

m, k |H |n, k

(A.42)

exp

i [En (k h

) Ei (k)]t

n, k |H |i, k

We write the time dependence of the perturbation Hamiltonian explicitly as H =

Hei t

(A.43)

and then Eq. A.42 can be written, after integrating twice |af (kf , t)|2 = 2 t h
(2)

| f |H |m, k |2 | m, k|H |i |2
m,k,,

(Em (k) Ei )2 h

(Ef Ei ) (A.44) h h

This second-order time-dependent perturbation theory expression is used to derive the probability of an indirect interband transition.

185

Appendix B

Harmonic Oscillators, Phonons, and the Electron-Phonon Interaction

B.1 Harmonic Oscillators

In this section we review the solution of the harmonic oscillator problem in quantum mechanics using raising and lowering operators. The Hamiltonian for this problem is written as: p2 1 H= + x2 . (B.1) 2m 2 Classically, we know that the frequency of oscillation is given by = H= 1 p2 + m 2 x2 . 2m 2 /m so that (B.2)

We dene the lowering and raising operators a and a , respectively, by a = and a = Since [p, x] = /i, then it follows that h [a, a ] = 1 so that H = 1 (p + imx)(p imx) + m h 2m = [a a + 1/2]. h 186 (B.6) (B.7) (B.5) p + imx . 2 m h (B.4) p imx 2 m h (B.3)

Let N = a a denote the number operator and we denote its eigenstates by |n , so that N |n = n|n where n is any real integer. However n|N |n = n|a a|n = y|y = n 0 (B.10) (B.9) (B.8)

where |y = a|n implies that n is a nonnegative integer. We note with regard to Eq. B.10 that the absolute value square of any wavefunction cannot be negative, because quantum mechanically, this quantity signies a probability. Hence n is positive number or zero. The action of the lowering operator is found from consideration of N a|n = a aa|n = (aa 1)a|n = (n 1)a|n . Hence we nd that a|n = c|n 1 . However from Eq. B.10, we have n|a a|n = |c|2 , and also from Eq. B.10 we have n|a a|n = n, so that c= and a|n = n (B.14) (B.15) (B.16) (B.13) (B.12) (B.11)

n|n 1 .

Since the operator a lowers the quantum number of the state, a is called the annihilation or lowering operator. From this argument you can also see that n has to be an integer. The null state is obtained for n = 0. To obtain the raising operator consider, N a |n = a aa |n = a (1 + a a)|n = (n + 1)a |n . Hence we obtain a |n = n + 1|n + 1 (B.17)

(B.18)

so that a is called a creation operator or a raising operator. Finally, for the Hamiltonian in Eq. B.7 we write H|n = [N + 1/2]|n = (n + 1/2)|n h h (B.19) so that the eigenvalues for the harmonic oscillator are written as: E = (n + 1/2) h 187 n = 0, 1, 2, . . . . (B.20)

Figure B.1: 1D spring model

B.2

Phonons

In this section we relate the lattice vibrations to harmonic oscillators and identify the quanta of the lattice vibrations with phonons. Consider the 1-D model with springs shown in Fig. B.1 The Hamiltonian for this case is written as
N

H=

s=1

p2 1 s + (qs+1 qs )2 2ms 2

(B.21)

This equation doesnt look like a set of independent harmonic oscillators since q s and qs+1 are coupled. To obtain normal mode solutions we write qs = 1/ N ps = 1/ N
k

Qk eiksa (B.22) Pk eiksa .

These Qk s and Pk s are called phonon coordinates. It can be veried that [ps , qs ] = ( /i)ss h implies that h [Pk , Qk ] = ( /i)kk . The Hamiltonian for 1D lattice vibrations in phonon coordinates is H= 1 1 2 P k P k + k Q Q k k 2 2 (B.25) (B.24) (B.23)

and gives rise to the 1D phonon dispersion relation (see Fig. B.2)
1/2

2(1 cos ka) = 4/m

| sin(ka/2)|.

(B.26)

This is all in Kittel ISSP, see pp 611-615 (Sixth edition). Again let ak = a = k
iPk + k Qk , 2 k h

(B.27)

iPk + k Q k 2 k h 188

(B.28)

Figure B.2: Phonon dispersion relation. represent the annihilation and creation operators. The Hamiltonian written in terms of the creation and annihilation operators becomes H= yielding energy eigenvalues E=
k

k (a ak + 1/2) h k
k

(B.29)

(nk + 1/2) k h

(B.30)

The quantum excitation in this case is called a phonon, and the state vector of a system of phonons is written as |n1 , n2 , . . . , nk , . . . . To annihilate or create a phonon in mode k we then write ak |n1 , n2 , . . . , nk , . . . = = nk |n1 , n2 , . . . , nk 1, . . . (B.31) (B.32)

a |n1 , n2 , . . . , nk , . . . k

nk + 1 |n1 , n2 , . . . , nk + 1, . . .

from which the probabilities nk and (nk + 1) are obtained for the annihilation and creation processes.

B.3

Phonons in 3D Crystals

We give some examples of the phonon is 3D crystals. The rst example is the zone center atomic displacements in graphite shown in Fig. B.3. Graphite has 4 carbon atoms per unit cell, thus 12 zone center modes. There are 3 acoustic modes and 9 optic modes. The next example is the phonon dispersion curves for diamond shown in Fig. B.4. Diamond has 2 carbon atoms per fcc unit cell, thus 6 branches. The zone center optic modes are Raman active. There are 3 acoustic branches and 3 are optic modes. The next example is the phonon dispersion curves for silicon shown in Fig. B.5. Silicon 189

Figure B.3: Zone center optical phonon modes in graphite.

190

Figure B.4: Phonon dispersion curves in diamond.

Figure B.5: Phonon dispersion curves in silicon.

191

Figure B.6: Phonon dispersion curves in silicon. like diamond has 2 atoms per fcc unit cell, thus 6 branches. The zone center optic modes are Raman active. There are 3 acoustic branches and 3 are optic modes. The next example is the phonon dispersion curves for GaAs shown in Fig. B.6. GaAs like diamond has 2 atoms per fcc unit cell, thus 6 branches. However the two atoms are dierent and GaAs lacks inversion symmetry. The zone center optic modes are both infrared and Raman active. There are 3 acoustic branches and 3 are optic modes.

B.4

Electron-Phonon Interaction

The basic Hamiltonian for the electron-lattice system is H= where H = Helectron + Hion + Helectronion . (B.34) The electron-ion interaction term can be separated into two parts: the interaction of electrons with ions in their equilibrium positions, and an additional term due to lattice vibrations:
0 Helion = Helion + Helphonon k,i

p2 1 k + 2m 2

kk

e2 + |rk rk |

Pi2 1 + 2M 2

ii

Vion (Ri Ri ) +

k,i

Velion (rk Ri ) (B.33)

(B.35) (B.36)

k,i

Velion (rk Ri ) =

0 Velion (rk (Ri + si )) 0 Velion (rk Ri )

=
k,i

k,i 0 Helion

si

0 Velion (rk Ri )

(B.37) (B.38)

+ Helphonon .

192

In solving the Hamiltonian H of Eq. B.33 we seek a solution of the total problem in the form = (r1 , r2 , R1 , R2 , )(R1 , R2 , ) (B.39) such that H = E. (B.40) We then use an adiabatic approximation, which solves the electron part of the Hamiltonian by 0 (Helectron + Helion ) = Eel . (B.41) Neglecting the Helphonon term, which we consider as a perturbation, we write: Hion = (E Eel ) = Eion (B.42) and we have thus decoupled the electron-lattice system. Equation B.42 gives us the phonon spectra and harmonic oscillator like wave functions, as discussed in the previous section (B.2). The term that was left out in the above discussion is the electronphonon interaction Helphonon =
k,i

si

0 Velion (rk Ri )

(B.43)

which we now treat as a perturbation. We rewrite Eq. B.42 by introducing the normal coordinates 1 0 si = Qq,j eiqRi ej (B.44) N M q,j where j is polarization index and ej is a unit displacement vector for mode j. Hence we obtain 1 0 0 (B.45) Helphonon = Qq,j eiqRi ej Velion (rk Ri ) N M q,j k,i where Qq,j h = 2q,j
1 2

(aq,j + a ). q,j

(B.46)

Writing the time dependence explicitly for the raising and lowering operators aq,j (t) = aq,j eiq,j t
aq,j (t)

(B.47) (B.48)

aq,j eiq,j t

we obtain Helphonon = h 2M N q,j

0 1 2

(aq,j eiq,j t + aq,j eiq,j t )

0

q,j

which can be written as Helphonon =

k,i

e (eiqRi + eiqRi )j

0 Velion (rk Ri )

(B.49)

q,j

h 2N M q,j

1 2

aq,j

e
k,i

0 i(q Ri q,j t)

ej

0 (rk Ri ) + c.c. (B.50)

193

If we are only interested in the interaction of one electron and a phonon on a particular branch, say the longitudinal acoustic branch, then we drop the summation over j and k and write Helphonon = h 2N M q
1 2

aq
i

ei(qRi q t) e

0 Velion (r Ri ) + c.c.

(B.51)

where the 1st term in the bracket corresponds to phonon absorption and the c.c. term corresponds to phonon emission. With Helphonon in hand, we can solve transport problems (e.g., due to phonon scattering) and optical problems (e.g., indirect transitions) directly, since all these problems involve matrix elements f |Helphonon |i coupling initial and nal states i and f , respectively.

194