Appl. Phys.

A 29, 125-132 (1982)

Applied so.,..
Physics A Surfaces
9 Springer-Verlag 1982

Impact Ionisation Probability in InSb

J. T. Devreese 1'2 and R. G. van Welzenis 1
t Eindhoven University of Technology, Department of Physics, P.O. Box 513,
NL-5600 MB Eindhoven, The Netherlands
2 University of Antwerpen (RUCA and UIA), Belgium

R. P. Evrard
Universit6 de l'Etat ~t Liege, Institut de Physique, Sart Tilman, B-4000 Libge, Belgium

Received 16 February 1982/Accepted 18 August 1982

Abstract. An exact analytical expression for the impact ionisation probability for the main
ionisation process in a semiconductor with a Kane-type bandstructure is derived. In a
parabolic approximation the integrals involved can be evaluated and the result reduces to
the simple relation vi(E) = C(E-Ei)20 (E-Ei) where C is a material constant related to the
transition matrix elements. From experimental results on the photo-quantum efficiency we
can estimate the value of C for InSb near 77 K, i.e. hC~0.012 eV- 1.-It is concluded that the
impact ionisation probability cannot be described with a unit step function at the threshold
energy as was done in many theories on the avalanche effect. The (effective) threshold is
smooth and a more detailed description, as derived here, is necessary.

PACS: 72.20

Impact ionisation is a process of still growing impor- is the threshold energy for the dominant impact ioni-
tance in semiconductors. When a free carrier is allowed sation process [2, 3]. This follows the original idea of
to gain sufficient energy it may collide with a bound Shockley [4]. The actual impact ionisation probability
carrier, freeing the latter in the process. This results in was calculated by Kane for Si [5] but this procedure
excess free electron hole pairs, in addition to the was not applied to other important semiconductors
original free carrier which looses most of its kinetic like the III-V compounds. An attempt by Curby and
energy. There is a whole class of semiconductor devices Ferry [6] for InSb has been criticized by Borodovskii
which operate on this principle [1], and in a large et al. [7]. A better approach is to replace Shockley's
number of other semiconductor phenomena in which unit step function by the intuitive formula given by
highly energetic carriers are involved it manifests itself Keldysh [8], as was recently done in Monte Carlo
either as a disturbance or as an important factor. calculations for GaAs [9].
Consequently there is fundamental interest to under- We shall give a rigorous derivation of the impact
stand and describe these processes. Usually there is not ionisation probability for semiconductors in which the
just one electron undergoing impact ionisation scatter- band structure, at the energies of interest, can still be
ing, but there will be a number of events resulting in an reasonably approximated by Kane's formulae
avalanche of electron-hole pairs. In order to be able to (Sects. 1-3). Next we shall derive a parabolic band
calculate the generation rate of these extra carriers one approximation of the general formula that leads to a
needs the (dynamical) distribution function and the simple expression which is comparable to that of
ionisation probability for a single event. Keldysh. Finally, we shall apply our results in a semi-
In this paper we shall concentrate on the latter aspect. empirical fashion to InSb at 77 K. From experimental
In most theories on the avalanche effect one still data we are able to calculate the impact ionisation
assumes a unit step probability U(E-Ei), in which E~ energy relationship vi(E) for this case.

0721-7250/82/0029/0125/$01.60
126 J. T. Devreese et al.

Table 1 where fi,.k,~ is an annihilation operator for an electron

with quantum numbers n, k, s. The operator describing
J L the Coulomb interaction of two electrons then takes
Conduction band (c) 1/2 + 1/2
the following form;
Valence band heavy hole (vi) 3/2 + 3/2
Valence band light hole (v2) 3/2 _ 1/2
V=e2 S d3r ~ d 3 r ' g - 1 q~+(r)q~+(r')@r')q~(r)
[r-r'l ' (3)
9 50

where e-1 is the inverse dielectric function describing

the medium. The operator (3) represents the interac-
tion between an electron in the conduction band
0
(labelled 1 in Fig. 1) and an electron in a valence band
(e.g. 4 in Fig. 1), which will lead to the excitation of the
0 : i - valence electron to the conduction band.
O)
f_
0
Using (2) for the field operators, the interaction oper-
c-
o
ator (3) becomes:
-.2S
c
0
L ~+ ~+ ~ A 2
4~
0
0
-. 50
0

.re - 1 T*2,k~,sz(r)~*3,k'z, s3(r')~nl,kl,s,(r')~n4,ka,

Ir-r'l
s4 (r)

- . 7S - - f J
-1O -S S lt3 9d3rd3r ' . (4)
k ( 10 8 m- I )
Fig. 1. Band structure of InSb at 77 K. A possible impact ionisation The sum in (4) is over all quantum numbers {ni, k'i, si}.
process, initiated by the conduction band electron in state 1, is
To calculate the probability per unit time for impact
indicated. The arrows show electron transitions
ionisation one needs to calculate the following matrix
elements :

1. Derivation of the Cross Section for Impact Ionisation

Vii ~ ( f l ~/v+,,k4, s4 ac, k3, s3 ac, k2, s2 [/r ac+,kl, ~ li>, (5)
for a Kane-Type Bandstructure
In this section we shall derive the general form of the
where the indices i,..., 4 correspond to the labels for
matrix elements which contribute to the cross section
the electrons in the different bands in Fig. 1, for all the
for impact ionisation by conduction electrons in semi-
conductors with a Kane-type bandstructure. possible initial and final states: li), If). The spin state
As an example, the conduction band and the 2 highest of the electron 1 is labelled e. The spin quantum
valence bands of InSb are presented in Fig. 1. These numbers s2, s3, and s~ can be either ~ or ft.
bands can be characterized by the quantum numbers j We will concentrate on ionisation of heavy holes,
(angular momentum) and j~ (z-component of the called the main process in [11], where it was also
angular momentum) as given in Table 1 [10]. shown that ionisation of light holes can be neglected in
The wave function for an electron at r with wave vector InSb. The non-zero matrix elements in (5) correspond
k in a given band is in general: to the following choice of indices in fz:

1 k ~ = k l, n,~=c, s4=e
T,,k,~(r) = u~AU,,k, ~(r) e ik'r, (1)
V ~Z with k'l = k 4 , nl=vl, s i {=~ }fl , (6a)
where n = c , v l, v z is the band index, f2 the crystal
or
volume, and s the spin variable. We introduce the field
operators : k'i=ki, nl=c, sl=~

#~(r)-- ~ ~.,k,~T.,k,~(r), (2) with k~=k,, n4=vi, s4='{~ }, (6b)

n,k,s
Impact Ionisation Probability in InSb 127

and The other terms, for which the spin indices are the
same, give zero.
k 2 =k 2 , Similarly, the choice of indices (6a) and (6d) results
in the following contributions.
For the r-dependence;
with k3=k3, n3=c, s3: , (6c)
~rtc*k3, a (r) ~tttc,kl ' ~(r) = 1 e-i(k3 -kl)'rS2(r) . (9a)
or

k~ = k2, For the r'-dependence;

* k2, ~(r ') }t~vl, k4, p( r' )

~trJc,
with k 2' = k 3 , n2=c, $2= {~}
fl , (6d)
-- & e -- i ( k 2 -- k 4 ) ' r ' S ( r ' ) [X(r') - i Y ( r ' ) ] (9b)
eve
{~} means either spin state ~ or/L
and
The conditions (6) then allow four contributions if
different possibilities to choose s 2, s3, and s 4 are * ! !
t/tc, k2,/~(r ) ~vl,k4,~( r )
excluded.
It is assumed that for the energies of interest here the = ~2 e- i(k~- k~)-~'S(r,)[X(r') + i Y(r')]. (9c)
electrons and holes can be adequately described by
Kane's wave functions [10]
1 Note that the possibilities (6b) with (6c), and (6b) with
T.,u, s(r) = - ~ u., 0, ~(r) e i k ' r , (7) (6d) lead to the same results [as in (8) and (9)] where
now r and r' are interchanged.
and the k dependence is in e ~k'~ only. As stated before
only one valence band has to be taken into account.
Therefore, following the standard Kane notations we
will use the functions: 2. Calculation of M a t r i x E l e m e n t s
uc,~=iSt~), u~,p =iSI]'), (7a) Consider first the choice of indices (6a) with (6c). The
interaction operator V then takes the form :
u~l,~=~2(X+iY)l$), u~,p= ~22(X-iY)I$). (7b) A ~ A
V - a~k2,~a:k3,~av~,k4,pac, kl ' ~M1,
A
(10a)
First the choice of indices (6a) and (6c) is considered. with
This choice leads to only one non-zero contribution in
-ie 2 [d3_ f 33_,^_ 1
the numerator of the integrand in the R.H.S. of (4) as M1- ~2-J- 9 9~
far as the r-dependence is concerned:
e -i(u2-kl)'rSe(r)e-i(k~-k4)'~'S(r')[X(r')-i Y(r')]
~c,, k2,~(r) ~c, kl, ~(r) _-- 1~ e- i ( k 2 - x,)-,S2(r ) (8a)
Ir- r'l
The term involving both ~ and fl leads to a matrix (lOb)
element proportional to ($]'~) and therefore is zero. The following matrix elements should now be
From the/-dependence two non-zero terms arise: calculated:
Iff* k3, ~(r' ) tffvt, k4, a(r') VYi=Ml(Olavl,k4,s4ac,
~+ ~ k .... ac,
A kz . . . .fi+
. k2,~
--i , " t
= 0 1 / ~ e-"k~-k")" S(r') [X(r')- i Y(r')] (8b) c,k . . . . ,,k4,#aC, k2,eac, k~,o~lO) (IOc)
V ~. One easily finds
and
Vfl =M1, (10d)
T* k~, p(r') ~/31, k4,,(r') for the following spin indices (Fig. 2)
-i
= ~1/2f2 e - i(k~- k~). ~'S(r') [X(r') + i Y(r')]. (8c) electron 1 2 3 4,
spin state c~ a c~ fl (10e)
128 J.T. Devreese et al.

and where the spin states are now

electron 1 2 3 4, electron 1 2 3 4
(10t)
spin state e c~ fi c~. spin state c~ c~ ~ ft.
The matrix element M~ can now be simplified as M 2 then becomes
follows ~ :
4xie 2 1 Mz:-- ice 3 3 t s - l e - i ( k 3 - k D ~ r -i(k2-k4)"r'
f272~_~-S d r S d r e
M1 - Oa ]//2 q~ e(co, q)q2 it, R'
9 e--i(kz--kl-q).l~e-i(k3--k4 +q)-R' S2(r) S(r') [X(r') - i Y(r')]
(16)
Ir-r'l
9 I daP. I d a p . ' S 2 ( e ) S ( e ' ) [ X ( Q ' ) - i Y(Q')]
9 e -i(k2-k~-@~e -i(k3-k4-q)'r , (11) Using arguments similar to those following (10b) this
expression is transformed to
where r = R + Q and r' = R' + Q' ; both r and 0' belong to
the unit cell containing the origin and R and R' are 4nie/N 2
M2- ~3~/~ 3kl+k4'k2+k3~e-l(c0'lk3--kl+K[)K
translational vectors of the InSb lattice, e(co, q) is the
Fourier transformed dielectric function. The sums over 1
R and R' are zero except if Ik3 - k 1 + KI / -[d 3Qsz(Q) eiK'e
q=k2-kl+K and q=k4-k3+K', (12) 9~ d 3 Q' S(Q') [X(Q')- i Y(Q')] e-iK. r (17)
where K and K' are reciprocal lattice vectors. Because
all k vectors have their endpoint close to the origin of The remaining spin configuration also contributes a
the Brillouin zone it follows that K = K' and also matrix element equal to M 2.
Furthermore one should consider the combination
k 1 +k 4 =k 2 +k 3 . (12b) (6b) with (6c). This combination is similar to (6a) and
t t t
This result, which is not obvious a priori [cf. (12)], (6d) but now k't and k 4 as well as k 2 and k 3 are
means that Umklapp processes are not important in interchanged 9 This leads to the commutation of
the present approximation. a,,1, kl, ~1 and a,,4, k'~,s,, and to the commutation of a,,+~,k'~,s2
Therefore M 1 simplifies to and a,3,+ k~,~ in (10c) so that the sign is conserved9 The
conditions on the spins remain the same for (6b) with
4nie2N 2 (6c) as for (6a) with (6d). In the matrix elements r is
M1-- ~3V~ 6kl+k4,k2+k3~ e - l ( c ~ +Kt)
replaced by r' and vice versa so that (6b) with (6c) and
(6a) with (6d) contribute the same matrix elements. In
1
Ik2 - k t + K [ 2 ~ d 3 ~ S Z ( ~ ) e iK'Q an analogous fashion it is found that (6a) with (6c) and
(6b) with (6d) also contribute the same matrix
9~ d3 Q, S(Q') [X(Q') - i Y(0')] e - i ~ ' r (13) elements9
In conclusion, there are four contributions two of
Now consider the possibility (6a) with (6d): The in- which contribute 2M t and the others contribute
teraction term then takes the form - 2M 2. For M 1 the energy hco is transferred during the
transition k 1--*k2. For M 2 the energy ha) corresponds
V = M 2 a +ka,eac+kz,~av,,k4,11aC,kl,~ (14a)
to the transition k l ~ k 3. Because relatively large
and subsequently K-vectors dominate M 1 and M 2 (i.e. any non-zero
reciprocal lattice vector) the dielectric function
Yfi = M 2 <0l a+,k4, fl aC,k3, c~ac,k2, a ac+,k3, a ac+k2, c~ e-1(co, K) describes short-distance screening over dis-
avl,k4,11ac, kb~ac,kb~l ) (14b) tances comparable to the unit cell, between charges. At
such short distances the screening is very weak and it is
From which we have reasonable to assume e-1(co, K)= 1.
Vfi = - M 2 , (15) We are therefore led to the conclusion that the total
amplitude for pair creation ~ is of the form:

1 The reader may have noticed that at this stage our expressions
bear some resemblance to those derived by Anton6ik [E. Anton6ik :
~ ~___M~kl +k4,kz+k3 , (18)
Czech. J. Phys. B 17, 735 (1967)]. However, our approximation of the
wavefunctions is different. Moreover, the present work leads to
where M is a sum over reciprocal space, cf. (13) and
analytical final expressions (17).
Impact IonisationProbabilityin InSb 129

3. The Cross Section for Electron-Hole Pair Creation

The cross section for electron-hole pair creation for a
specific transition takes the form:
2n
w(kl, k4; k2, k3) = ~ - 6(Et + E4 - E 2 - E3)
(kz,,~3) (k/.,,ca)
~. i(Ola~k,..fi..k,.,,
a,.k,..,[Za+k,..lO)i9 (19)
$2, S3, $4

We now denote by /}t, /)z, /3~, and /)4 the four

operators which realise the corresponding transitions
in Fig. 2. The operator fz then is

f~=bl+b~+b3 +b,, (20)

and it follows that

2~ 6
w(kl,k4;kz, k3) = ~ - ( E 1 - ] - E 4 - - E 2 - - E 3 ) ~ k ~ + k 4 , k 2 + k 3 ,,k
Fig.2. The four differentcombinationsof quantum numbers for the
9 IMI~(<olD~B~ IO>- <OlbJ-b~ IO>+ <OlbJ-b~ IO> electron states that contributematrix elements M 1 and M~
+ <OlbJ-b~ IO>- <Olb~-b~Io> + <Olb~-b,,lO>). (21)
.~0
The minus sign for the second and fifth term are /
associated with the matrix elements - 2 M 2 for the / bv-
"crossed" diagrams @ and (~) in Fig. 2. Terms of the
type (0[/)~-/)4[0) vanish because the electrons in the
conduction band have opposite spins for the inter-
' I,
mediate state.
The cross section for electron-hole pair creation then
o)
L
B A
o
becomes: C
o
-. 2S
C
4re o
w(ka, k 4 ; k2, k3) : ~ - [M[2 6(E~ + E 4 - E 2 - E3) L
o
o
-.SO
95kl +k4,k2 +k3 9 (22) o

In the expression a factor of 2 is due to the inclusion of

the probability for transitions of valence electrons with -. 7S I I

both spin directions. The probability per unit time that -11 -5 0 S 10
an electron with energy k2~/2 will create an electron- ( 10 8 m-1 )
hole pair equals Fig. 3. The two possible photon absorption processes between the

w(k~)= ~ (~ ~dak2~d3k3~d3k4
conduction band and the two highestvalencebands

expression for w(kl). Combining (22) and (23) we then

9w(kl, k•; k2, k3). (23) have to calculate the following expression:
4~ 2/~2 \3
w(kl)= ~-IMI I ~ ) J'dk21 dk3j'dk 4
4. Parabolic Approximation
Although the band structure of hardly any semicon-
ductor at higher electron energies can be described by
a parabolic law it makes sense to evaluate the forego- 96(k 1 + k 4 - k z - k 3 ) . (24)
ing expressions in the parabolic approximation 9 The Here m is the heavy hole mass expressed in units of the
resulting mathematical simplificatiofl allows us to cal- (band) mass of the conduction electron. Some details of
culate the integrals and we can find a closed analytical the evaluation of (24) are presented in Appendix A.
130 J.T. Devreeseet al.

The integrals involved can be evaluated analytically the ratio of the number of created electron-hole pairs
and the following expression results: to the number of absorbed photons - remains equal to
unity. As soon as the electron energy rises above Ei,
w(kl) = W m + 2 Eg O (25) and this will happen first for process A only, there is
m+l m+i 9, the possibility of creating secondary electron hole pairs
where the Heaviside function O ( x ) = l if x > 0 and by impact ionisation and t/ may rise above 1. Let us
O (x) = 0 for x < 0. It should be pointed out that (25) has consider the region near this critical photon energy E,..
been obtained as a simple approximation to (17). The If we introduce:
expression (17) is rather general and allows in principle N(A)
to evaluate the matrix elements needed here if the 7 - N(A) + N(B)' (27)
dielectric function ~(co,K) is known accurately. A suf- as the relative chance that a photon is absorbed by
ficiently accurate expression for w(k0 is a prerequisite process A and not by process B, one can readily derive
for any fundamental study of avalanche or photo the following expression for the quantum efficiency for
quantum yield in the avalanche regime. process A only

t/A_ number of electron- heavy hole pairs created by process A

= 1 + (~1- 1)/7. (28)
number of absorbed photons

Any conduction band electron created via process A

5. Derivation of IMI 2 from Experiment
has in principle two possibilities to loose its energy: it
In the previous section an expression for the probabili- can either impact ionise - frequency h(E) - or it can be
ty of impact ionisation with parabolic bands was scattered by a polar optical phonon with a frequency
derived. This expression can be written in a simpler VLo(E ). We shall only account for phonon emission.
form if it is recognized that for parabolic bands The phonon absorption can be neglected because of
(m+2)/(m+ 1)E~ is just the threshold energy E i [13]. the low phonon occupation number at 77K. This
We also transfer from k to E as the independent variable means that the relative probability for the creation of
und use the impact ionisation frequency vlinstead of w, so secondary electron hole pairs is h/(vi + VLO),thus
we write C for W per second:
vi(~) = c ( E - E~)2 0 ( ~ - ~i). (26) t/A= 1 + v~(E) (29)
v~(E)+ Vio(E ) '
Now it is known that for the ionisation process which
we consider here, there is only a small difference Strictly speaking this simple approach only holds in an
between the parabolic band estimate for E~ (233 meV) electron energy interval from E i to Ei+hC%o where
and a calculation taking the real bandshape into hc0i.o is the energy quantum of a longitudinal polar
optical phonon (24.2 meV in InSb at 77 K). Above
account (243meV) [11]. This is because the end pro-
this energy interval it becomes more complicated,
ducts of the ionisation process are all quite near k = 0.
because the electron could first be scattered by a
On the other hand, one does not expect (26) to change
polar optical phonon and then ionise or emit another
its functional form drastically on going from parabolic
phonon. For a first estimate we can take (29) as the
to actual bandshapes. Only the values of C and Ei will
main contribution in subsequent intervals nfiO)Lo as
be changed. So we will use (26) also for the non-
well, however. At these energies VLo is only weakly
parabolic case, inserting the actual i.e. the non-parabo-
energy dependent 2. Also ?, is only very slightly energy
lic, E~ value. This enables us to find a numerical
estimate for the value of C from photo-quantum 2 The polar optical emission frequency was calculated using the
efficiency experiments [14]. These measurements can standard FrShlich formula, see e.g. [E. Conwell: In Solid State
be interpreted in terms of the number of created Physics, Suppl.9 (AcademicPress, New York 1967)]. The value of
electron-hole pairs per absorbed photon. For photon the high frequencydielectric constant was taken from T. S. Moss
energies not much larger than the bandgap there are [Proc. Phys. Soc. B70, 778 (1957)] as ~o=15.68, which is the
generally accepted value. The best value for the longitudinal optical
two possible absorption processes in InSb (Fig. 3); phonon frequency still seems to be that of Stradling and Wood
either an electron heavy hole pair is created [Proc. Phys. Soc. 1, 1711 (1968)], i.e. C%o=3.67x 1013rad/s. This
(process A), or an electron light hole pair (process B). results in hegLo=24.2meV. The value of the transverse optical
In the latter case the energy of the created conduction phonon frequency~Orois found by interpolation from the 4 K value
of Hass and Henvis [J. Phys. Chem. Solids 23, 1099(1962)]and that
electron is lower for a given energy of the incident of Sanderson at 100K [J. Phys. Chem. Solids 26, 803 (1965)]. Finally
photon than in case A. As long as the energy of the the value of the low frequencydielectricconstant is calculated from
created electrons is below the impact ionisation thresh- the Lyddane-Sachs-Tellerrelation as ~s= 17.64. The Fr~Shlichcou-
old E i, the quantum efficiency t / - which is defined as pling constant has a value of 0.02
Impact Ionisation Probability in InSb 131

l. IS
dependent over this energy range, a value of 0.82 can
90K
be calculated from Kane [10]. //
v
We are not aware of experimental data at 77 K so we
use the results at 90 K of Volkov et al. [14] (Fig. 4). 1.10
233
The apparent photon threshold energy E'c~0.54eV, O
r

is 2.5h~oLo larger than the absolute threshold E~=E~

1.05
+ E g + E h = 0 . 4 8 eV. From the experimental results it
is not possible to decide whether the quantum ef- O
ficiency from E~ to E'c is indeed constant and equal to E
1 or that the curve is actually slowly rising. New and :3
.,.)
1.00
c" ,/'i
more detailed measurements would be necessary. CO
D
CY I Ecj
Just above E'~ the experimental curve shows a well
95
defined slope, i.e. dO/= 1)/dE= 1.24 eV -1. We can use 9

.45 .S0 9 EE .60 . GS

this to find an estimate for C. By equating (28) to (29)

and differentiating and using (26) we find the following photon e n e r g y ( eU )
expression for C: Fig. 4. The quantum efficiency of the internal photo-effect in n-type
InSb at 90 K (solid line), according to Volkov et al. [Ref. 14, Curve 2
C- Ve~ 7 d(rI - 1) (30) from Fig.2]. Also indicated are the absolute photon threshold
2(E-E,) [ 7 - ( r / - l ) ] 2 dE energy E c and the apparent photon threshold energy E'c

If we choose the electron energy E=Ei+3hO,)LO,

which corresponds to a photon energy of 0.55 eV, I I I I I I l
this result in
C,~7x 105~ - 2 s -a or hC~0.012eV -1 10 12 "v't~
Using this value for C we have plotted (26) in Fig. 5.
We have also included vLo(E) in this figure for com-
parison. Note that even at 400 meV, which is 6hO)LO
above E~, the two curves do not intersect.

N 1011
"r"
6. Discussion
v

The impact ionisation frequency rises only slowly

from the threshold E i and remains much smaller than
VLo in a large energy range (in units of hCOLO)above E i.
Thus, the avalanche process can not be described I010
with a simple unit step function at E i. This would only
be reasonable if the vi(E) curve would intersect the
vLo(E) curve within hC0LOfrom Ei. Moreover, it will
be clear from Fig. 5 that it is not possible to sub-
stitute some energy E > E ; for the threshold at which
a unit step function description would hold. There is 109
not a sharp threshold for avalanche but instead a 240 280 320 360 400
smooth transition.
electron energy ( meU )
The physical explanation for the apparent photon
threshold energy E'c in Fig. 4, is not clear at present. Fig. 5. The impact ionisation frequency for InSb at 77 K calculated
It corresponds with an electron energy El= 300 meV according to (26) (solid line) with C = 7 x 10s~ The broken
line is the polar optical phonon scattering frequency. The absolute
which is smaller than an effective threshold that can (E,), apparent (El) and effective (E'[) electron threshold energies are
be calculated by imposing equal energy losses, i.e. indicated as explained in the text
vi(Ei')E i VLo(E~t)~'/(2)LOor:
=

(E;' - El) z = (hCOLo/Ei) x (Veo(E~')/C) ,

The ideas developed in this paper and applied to InSb
resulting in E;' = 332 meV,,~ E~+ 4hCOLO.It was already may have a more general applicability. They could be
pointed out by Kane [5] however, that the definition used for other semiconductors having a Kane-type
of a "'threshold" would depend on the process that bandstructure, provided that the latter approximation
one considers. is still valid at the required energies [9].
132 J.T. Devreese et al.

Our estimate for C will be too low, because we evaluat- with

ed it at 3hCOLO above the threshold using (29). This kl +mk2 +u
kv- m+l '
means that we neglected the possibility that an
electron first emits one or two phonons and sub- k 1 +ink 2 --U (A.6)
sequently ionises, this could introduce a factor of the kL: m-I- 1
or
order of two. However, we feel that it is not worthwhile (2TO2 /k 2 m a_2 \ kv, kr~
to try to improve our estimate along the same lines. I(kl)=t~'~[~ O[Z~---Eg '" ' ~} ~ dk2k 2 ~ dk
2k 1 \2 mq-1/kL, kL
A better way to arrive at a value of C would be to
9[ - (m + 1)k2 + 2k(k 1 + mk2) + k~(m- 1 ) - 2mE a - 2mk~]
solve the Boltzmann transport equation for the with
quantum efficiency experiment, given vLo(E) and an k - kl +v
expression like (26) for v~(E), and fitting the resulting ~' m+2 '
t/(E) curve to the experimental results. k 1- v (A.7)
k L,
Acknowledgements. It is a pleasure to thank M.J. Steenland for his m+2 '
continuous interest and stimulating interaction. We have benefitted where
from fruitful discussions with F. Brosens and P. van Camp.
v 2 = k~(m + 1)2 - 2(m + 1)(m + 2)E o . (A.8)
Appendix A I(kl) then becomes
In this appendix a brief outline is presented of a method to evaluate I"" (2n)2[k 2 m+2 ~ 1 4 k .v' 3
(24) for w(k 0 ttq)= ~ t~ I V - m + 1 E~ ~ 3k!, dk2kzs (A.9)

I(kl)=Idkzidkalclk46(~ -Ea 2mk~ k22 with

m
s 2 = - m(m + 2)k22+ m + 2 v2 (A. 10)
98(k 1 + k 4 - k 2 - k 3 ) (A.1)
it follows :
using the &function for the momenta, denoting k3 as k - k z, first 2 (k~z m + 2 )4 1 1
integrating over kz and choosing the z-axis in the direction of k we I(kl)=(2~ ) O T - r e + l E a ~ (m+1)2 (m+2)
obtain :
1 +kk2 [ m \3/2 1 v
I(kx)=2~dk ~ dk2k~ , ~ dx 9 [--| --fdx(v2--x2) 3/2 (A.11)
0 ~ 2 -kk2
\m+2] m+2;
which finally takes the form
I. . . . . . 2[m+Xk2[ m ~3/2{k~ m + 2 ~2
or tgl)=tqTr) / m - ~ ) ~ ) IT-m+lgoJ
dk ~
I(kO =2~I T I dk2 k2~ k~ m+2
o

kz k2
Equation (A.12) yields directly (25) with
k2 k2 32n 2 m+l 2 m 312
2 4]
Now integrate over the azimuthal and polar angles of k and choose in which the only remaining unknown parameter is now IM[ 2, which
the z-axis in the direction of kl to find: can in principle be evaluated from (17).
m +co +oo
I(kl)=(2n)2~ j ~ dk ~ dk2k z References
1. S.M. Sze : Physics of Semiconductor Devices, 2nd ed. (Wiley, New
York 1981)
2. P.A. Wolff: Phys. Rev. 95, 1415 (1954)
3. G.A. Baraff: Phys. Rev. 128, 2507 (1962)
kk 1 k~ _Eo_k~). 4. W. Shockley: Solid State Electron. 2, 35 (1961)
m]-2-\ +1) (A.4)
5. E.O. Kane: Phys. Rev. 159, 624 (1967)
6. R.C. Curby, D.K. Ferry: Phys. Lett. 35A, 64 (1971)
In view of the Heaviside function k must take values between 7. P.A. Borodovkii, S.D. Luehinin, V.M. Osadehii, R.M. Pogreb:
k_+ =(k 1 +mk2+_u)/(m+ 1) where Soy. Phys. Semicond. 13, 272 (1979)
8. L.V. Keldysh: Soy. Phys. JETP 21, 1135 (1965)
u 2 = -- m(m + 2)k2z + 2mklk z + m2k 2 - 2m(m + 1)E o . (A.5) 9. H. Shichijo, K. Hess: Phys. Rev. B23, 4197 (1981)
10. E.O. Kane: J. Phys. Chem. Solids 1, 249 (1957)
Therefore 11. R.G. van Welzenis: Appl. Phys. A26, 157 (1981)
12. D.J. Robbins: Phys. Stat. Sol. (b) 97, 9-50 and 387-406 (1980);
I l,k 1,~= ~(2~z)2 +?
-Jo d k 2 k 2 0 [ - m(m + 2)k~ + 2mk~k 2 + m2k~ 98, 11-36 (1980)
13. A. Beattie, P.T. Landsberg: Proc. Roy. Soc. (London) 249, t6
- 2m(m + 1)Eg] (1958)
ktr
14. A.S. Volkov, A.A. Gutkin, S.E. Kumenov: Soy. Phys. Semicond.
9 ~ d k [ - (m + 1)k / + 2k(k 1 + ink2) + k~(m- 1) - 2mEg- 2mkZ~] 7, 64 (1973)
kL