826
IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 58, NO. 3, MARCH 2011
Integration of Self-Assembled Redox Molecules
in Flash Memory Devices
Jonathan Shaw, Yu-Wu Zhong, Kevin J. Hughes, Tuo-Hung Hou, Hassan Raza, Member, IEEE, Shantanu Rajwade,
Julie Bellfy, James R. Engstrom, Héctor D. Abruña, and Edwin Chihchuan Kan, Senior Member, IEEE
Abstract—Self-assembled monolayers (SAMs) of either ferrocenecarboxylic acid or 5-(4-Carboxyphenyl)-10,15,20-triphenylporphyrin-Co(II) (CoP) with a high-κ dielectric were integrated
into the Flash memory gate stack. The molecular reduction–
oxidation (redox) states are used as charge storage nodes to reduce charging energy and memory window variations. Through
the program/erase operations over tunneling barriers, the device
structure also provides a unique capability to measure the redox
energy without strong orbital hybridization of metal electrodes
in direct contact. Asymmetric charge injection behavior was observed, which can be attributed to the Fermi-level pinning between
the molecules and the high-κ dielectric. With increasing redox
molecule density in the SAM, the memory window exhibits a
saturation trend. Three programmable molecular orbital states,
i.e., CoP0 , CoP1− , and CoP2− , can be experimentally observed
through a charge-based nonvolatile memory structure at room
temperature. The electrostatics is determined by the alignment
between the highest occupied or the lowest unoccupied molecular
orbital (HOMO or LUMO, respectively) energy levels and the
charge neutrality level of the surrounding dielectric. Engineering
the HOMO–LUMO gap with different redox molecules can potentially realize a multibit memory cell with less variation.
Index Terms—Coulomb blockade effect, high-κ dielectric, nonvolatile memory devices, reduction–oxidation (redox)-active molecules, self-assembled monolayer (SAM).
I. I NTRODUCTION
F
LASH memory device structures with discrete charge
storage such as nanocrystals (NCs) and dielectric traps are
potential candidates for continuous memory scaling by maintaining a coupling ratio and reducing crosstalk in conventional
floating-gate devices. A metal NC embedded in a high-κ dielectric [1] has been proposed to improve low-voltage program/
erase (P/E) efficiency by the 3-D field enhancement effect [2].
The metal work function also provides an additional offset to
Manuscript received May 28, 2010; revised November 18, 2010; accepted
November 25, 2010. Date of publication December 30, 2010; date of current
version February 24, 2011. This work was supported by the National Science
Foundation through the Cornell Center of Materials Research Interdisciplinary
Research Group on Controlling Electrons at the Interface. The review of this
paper was arranged by Editor Y.-H. Shih.
J. Shaw, S. Rajwade, H. D. Abruña, and E. C. Kan are with the School of
Electrical and Computer Engineering, Cornell University, Ithaca, NY 14853
USA (e-mail: jts57@cornell.edu).
Y.-W. Zhong is with the Laboratory of Photochemistry, Institute of Chemistry, Chinese Academy of Sciences, Beijing 100864, China.
K. J. Hughes and J. R. Engstrom are with the School of Chemical and
Biomolecular Engineering, Cornell University, Ithaca, NY 14853 USA.
T.-H. Hou is with the School of Electrical and Computer Engineering,
National Chiao Tung University, Hsinchu 300, Taiwan.
H. Raza is with the College of Engineering, University of Iowa, Iowa City,
IA 52242 USA.
J. Bellfy is with Villanova University, Villanova, PA 19085 USA.
Digital Object Identifier 10.1109/TED.2010.2097266
the Si band edges to prolong retention. However, nonuniformity
in NC size and density raises concerns of the parametric yield
for aggressive scaling [3], [4]. Charge storage in dielectric traps
is also vulnerable to trap density and energy variations.
In comparison, a combination of the top-down lithography
and the bottom-up molecule self-assembly processes can offer
a uniform charge density and possible stable multilevel storage
in a single memory cell [5]. The monodisperse nature of the
molecular orbitals (MOs) can potentially reduce cell variations,
whereas the distinct energy levels may enable stepwise charging for precise control of each memory state. By attaching
reduction–oxidation (redox) molecules as a floating gate in the
complementary metal–oxide–semiconductor (CMOS) structure, one also creates a high-impedance nonamperometric electrode for probing the molecular redox states, in contrast to the
low-impedance connections used in cyclic voltammetry (CyV)
[6]–[9]. Previous studies have shown capacitance and conductance peaks associated with the charging and the discharging of
the carriers stored at the MOs [7] and CyV hysteresis owing to
the redox of molecules attached to the SiO2 surface [6], [8], [9].
Compared with the CyV measurements, where the insulating
barrier was only deposited on one side [6], [9], the new structure
provides insulating barriers above and below the molecules,
which further lessen the possibility of orbital hybridization
from the gate. A strong coupling between the electrode and
the molecule may lead to molecule-independent switching behavior and modification of the intrinsic spacing of the energy
levels [10], which makes it difficult to study the moleculespecific behaviors. Furthermore, the lack of an insulating capping layer in CyV compromises carrier retention. An additional
shortcoming with the CyV method is the slow timescale due
to its reliance on ionic movement. The operation speed can be
enhanced significantly with an all-electron conduction mechanism in the proposed structure. Furthermore, the specifically
tailored redox molecules can be self-assembled as a monolayer
in current CMOS technology through the chosen functional end
group, where the discrete levels of the MOs can be preserved as
discrete memory states.
In this paper, we have integrated various molecules with a
high-κ dielectric in an otherwise standard Flash memory gate
stack. We report the charge retention and P/E characteristics.
We further confirm that porphyrin molecules have the thermal
budget that can withstand postmetal gate annealing [11]—an
important requirement for CMOS compatibility. Compared
with the previous studies [6]–[9], we were able to confirm
an interaction with multiple states of redox-active molecules,
instead of interface traps.
0018-9383/$26.00 © 2010 IEEE
SHAW et al.: INTEGRATION OF SAMs IN FLASH MEMORY DEVICES
Fig. 1. Schematics of the MOS capacitor structure with (a) FcCOOH (S1) and
(b) CoP (S2) molecules integrated as storage nodes.
827
Fig. 2. Molecular structure of FcCOOH, CoP, and the respective dummy
molecules used to control surface coverage density of the redox-active molecules. The approximate distance from the attachment site to the central transition metal tm and the thickness as-deposited on the surface tM are illustrated
for each redox-active molecule.
II. D EVICE FABRICATION
The metal–oxide–semiconductor (MOS) capacitor structure,
similar to that used in Flash memory devices [12], [13],
with redox molecules embedded in the control dielectric, was
fabricated. Two types of redox molecules were integrated:
ferrocenecarboxylic acid (FcCOOH) and 5-(4-Carboxyphenyl)10,15,20-triphenyl-porphyrin-Co(II) (CoP). The schematics of
the devices with FcCOOH (S1) and CoP (S2) molecules are
illustrated in Fig. 1. Both molecules are stable in air, with
FcCOOH and CoP exhibiting two and three stable redox states,
respectively.
We followed a self-assembled monolayer (SAM) formation
process similar to previous literature [9], utilizing the carboxyl
functional group on the OH-terminated surface. For S1, after
2.2- or 3.0-nm dry thermal oxide growths on a p-Si(100)
substrate, the wafer was immersed in a solution of dimethyl
formamide (DMF) with 1-mM FcCOOH. The solution was kept
at 80 ◦ C for 120 min under an argon environment during the
self-assembly process. For S2, the wafer was first placed in a
1% hydrofluoric acid (HF) solution to remove the native oxide,
followed with deionized (DI) water rinse/N2 blow dry, and then
immediately loaded into the atomic layer deposition (ALD)
chamber. After 3 nm of Al2 O3 , the wafer was immersed in a
1-mM CoP solution of DMF at room temperature for 24 h under
an argon environment. It is worthwhile to note that native SiO2
forms readily in air. A native oxide of ∼1 nm was present before
Al2 O3 growth on a separate test sample, which went through
an identical HF etch process, measured by ellipsometry. Prior
to the growth of the control dielectric, the native SiO2 was
measured to be ∼1.4 nm on our test sample with no molecules
present. The native SiO2 possibly formed during DI water rinse/
N2 dry after etching by HF and the initial cycles of thermal
Al2 O3 deposition. The surface density has been estimated to
be around 1 × 1014 and 0.45 × 1014 cm−2 for FcCOOH and
CoP SAMs, respectively [6], assuming a full surface coverage.
Furthermore, the thickness values of FcCOOH and CoP SAMs
measured by ellipsometry were in the range of 0.9–1.2 and
1.5–1.6 nm, respectively, which is in good agreement with a
previous estimate where the thickness is estimated to be 0.8 and
1.5 nm by impedance spectroscopy [8]. After the self-assembly
process, DMF was used to rinse the wafer to remove any
residual molecule that is not covalently bonded to the surface.
The SAM was covered by 30 nm of ALD Al2 O3 grown by the
reaction of trimethylaluminum and H2 O at 110 ◦ C. The Cr/Al
control gate was then deposited and patterned, followed by a
H2 /N2 forming gas anneal at various conditions.
We report the molecular structure of FcCOOH and CoP
molecules in Fig. 2. tm is the approximate distance from the
attachment site to the redox-active transition metal, whereas
tM is the thickness of the molecule. The dummy molecules,
exhibiting no redox states, were added in the control splits for
various surface density coverage of the redox-active molecules.
Dummy molecules were selected based on structural similarity
to the active molecules to ensure solubility of the dummy molecule in the deposition solution and to promote effective mixing
of the dummy and active molecules on the surface. The dummy
molecule for FcCOOH is benzoic acid (Bz), where 1 : 100
mixing is derived from 0.01 mM of FcCOOH and 1 mM of Bz.
Assuming the 1 : 100 mixing ratio in the bulk solution leads to
a 1 : 100 mixing of FcCOOH:Bz on the attachment surface, the
density corresponds to an intermolecule spacing of ∼5.8 nm for
828
IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 58, NO. 3, MARCH 2011
Fig. 3. XPS data of the (a) FcCOOH SAM deposited on thermally grown
SiO2 and (b) CoP SAM deposited on Al2 O3 ALD. For XPS studies, no dummy
molecules were used for preparing the SAM solution. The SAM solution
concentration of FcCOOH and CoP is 1 mM.
FcCOOH. We estimate the minimum intermolecule distance of
a 0.1-eV Coulomb blockade as
∆E =
1 q1 q2
≤ 0.1 eV,
4πεr ε0 r
r ≥ 5.75 nm
(1)
where ∆E is the energy shift due to repulsion force from the
neighbor molecules, q1 and q2 are the charges stored in the
redox-active molecules, r is the intermolecule distance, and εr
and ε0 are the relative and vacuum permittivities. Here, we
assume the εr of Bz to be 2.5. CoP was mixed with 5-(4Carboxyphenyl)-10,15,20-triphenyl-21,23H-porphyrin (H2 P),
which acts as the dummy molecule, at a 1:20 ratio.
III. R ESULTS AND D ISCUSSION
Fig. 3 shows the X-ray photoelectron spectroscopy (XPS)
measurement on the samples of as-deposited FcCOOH and
CoP SAMs. The 2p1/2 and 2p3/2 electron configurations corresponding to the orbital binding energy peaks of iron and
cobalt can be clearly observed for the FcCOOH and CoP SAMs,
respectively. Since ferrocene is an organometallic compound
that contains two cyclopentadienyl rings bound to a central iron,
the observation of peaks at binding energy levels characteristic
of the Fe 2p1/2 and 2p3/2 levels is evidence of ferrocene on
the tunneling oxide. Similarly, in the case of the CoP SAM,
the characteristic cobalt binding energy peaks indicate the
existence of CoP deposited on the Al2 O3 surface.
The XPS features may be also used to estimate the absolute densities of the redox-active molecules [14]. To quantify
the coverage density of the redox molecules chemisorbed to
the oxide surfaces, we use X-ray photoelectron spectra from the
evaporated Au (not shown) on a silicon substrate as a reference
standard. The integrated intensity under the [Au(4f)] feature
from the standard is proportional to σAu NAu λAu T (EAu ),
where σAu , NAu = 5.88 × 1022 atoms/cm3 [15], λAu =
1.78 nm [16], and T (EAu ) are the photoelectron cross section, atomic density, inelastic mean free path, and analyzer
transmission, respectively, which is inversely proportional to
the photoelectron kinetic energy. If we assume that the redox molecules form a finite thickness monolayer on the surface dSAM with the atomic density NSAM , and dSAM ≪
inelastic mean free path of SAMλSAM , the area under the
Fe(2p) and Co(2p) features will then be proportional to
Fig. 4. CyV of 2-mM FcCOOH in acetonitrile containing 0.1-M TBAPF6 as
a supporting electrolyte. Ag/AgCl was used as the reference electrode.
σFe NSAM dSAM T (EFe )/ cos θ and σCo NSAM dSAM T (ECo )/
cos θ, where σAu /σFe ∼ 1.04 and σAu /σCo ∼ 0.89 [17] are
for Fe(2p) and Co(2p), respectively, and θ = 55◦ is the photoelectron takeoff angle from the surface normal. With these
assumptions, the surface density NSAM dSAM (in molecules per
square centimeter) of FcCOOH and CoP can then be calculated
as 1.58 × 1014 and 3.6 × 1013 molecules/cm2 , respectively, for
SAMs undiluted with dummy molecules, which agree well with
previous estimation by CyV [6]. Considering the assumptions
that were made with the estimation method and the background
noise, the absolute accuracy is approximately ±35%.
Furthermore, the contact angles measured before and after
FcCOOH attachment were 10◦ and 80◦ , respectively, which
indicates that the film went from a hydrophilic oxide to a
more hydrophobic organic surface. For CoP, the contact angles
were 31◦ and 55◦ before and after the self-assembly process,
respectively. Our results from XPS, ellipsometry, and contact
angle measurements suggest that a monolayer of FcCOOH and
CoP molecules was formed on the SiO2 and Al2 O3 surfaces,
respectively. Furthermore, CyV was performed with 2 mM of
FcCOOH in acetonitrile containing 0.1 M of tetrabutylammonium hexafluophosphate (TBAPF6 ) as a supporting electrolyte,
as illustrated in Fig. 4. The current peaks are associated with the
oxidation (negative current) and the reduction (positive current)
of the molecules. One reversible redox state was observed at
+0.71 V versus the Ag/AgCl reference electrode.
Apart from this, from the resonant peak in the ultraviolet/
visible light (UV/Vis) absorption spectrum shown in Fig. 5, the
HOMO–LUMO energy gap can be estimated by the de Broglie
equation E = hc/λ, where c is the speed of light in a vacuum,
and E is the energy gap that corresponds to the resonant
wavelength λ. The spectroscopic HOMO–LUMO energy gap
was estimated to be ∼2.8 and ∼2.65 eV for FcCOOH and CoP
molecules, respectively, which matches well with the ab initio
calculations by the density functional theory (DFT) [18], [19].
The HOMO–LUMO energy levels are estimated by the DFT
to be about −4.51/−1.72 eV for the FcCOOH molecule and
−4.8/−2.2 eV for the CoP molecule [18], [19].
Consider the electrical characterization of the FeCOOH sample, for which the high-frequency capacitance–voltage (HFCV)
measurements in Fig. 6(a) were taken by applying stress
SHAW et al.: INTEGRATION OF SAMs IN FLASH MEMORY DEVICES
Fig. 5. UV/Vis absorption spectrum of (a) FcCOOH in acetonitrile and
(b) CoP in DMF. Deduced from the maximum absorption peak of the lowest
energy band at 440 nm, the spectroscopic HOMO–LUMO energy gaps are estimated to be ∼2.8 and ∼2.65 eV for FcCOOH and CoP molecules, respectively.
Fig. 6. (a) HFCV measurements and (b) ∆VFB as a function of the programming voltage at 10 K and room temperature for capacitors with FcCOOH
molecules. The mixture between FcCOOH and Bz has a 1 : 100 ratio. The
device was annealed at 400 ◦ C for 30 min. Carrier injection tests were
carried out by applying stress voltages for 5 s before each CV sweep. The
programming voltage Vp for electron and hole charging is applied from ±1
to ±7 V in a 2-V increment. Subsequently, the gate voltage is swept from
accumulation to inversion for electron charging and inversion to accumulation
for hole charging. The size of the MOS capacitor is 150 × 150 µm2 .
voltages on the control gate for 5 s to ensure that the program
operation reaches the steady state, followed by sweeping the
voltage from positive to negative direction for studying electron
injection and from negative to positive direction for hole injection from the silicon substrate. All subsequent HFCV measurements performed at 10 K were under a light source to promote
minority carrier generation. For this sample, 2.2 nm of thermal
oxide was grown as the tunneling oxide with a forming gas
anneal performed at 400 ◦ C for 30 min. The surface coverage
ratio is assumed to be identical to the bulk solution ratio of
1 : 100 between FcCOOH and Bz (as a dummy molecule). First,
we notice that, at small programming voltages, hole injection
showed a large memory window, whereas electron injection
is negligible, which indicates that hole injection is preferred
when a FcCOOH SAM is embedded in Al2 O3 . Also worth
noticing is the CV stretch-out at higher programming voltages,
which can be attributed to charge leakage back to the silicon
substrate. Fig. 6(b) shows the amount of flatband shift ∆VFB
as a function of the programming voltage extracted from the
HFCV measurements. The participation of both electrons and
holes is evident from the linear increase in ∆VFB by positive
and negative writing conditions, respectively.
The injected charges have three possible storage sites:
1) dielectric traps of Al2 O3 ; 2) traps near the Al2 O3 /FcCOOH
interface; and 3) FcCOOH itself to reduce/oxidize the molecule.
Since the sample without any molecules embedded in the gate
829
stack has shown negligible ∆VFB for programming voltages of
−10 to +20 V, charges are not likely to be stored in dielectric
traps of Al2 O3 . A linear increase in ∆VFB with respect to the
programming voltage is an indication of charges being stored
at the interface states, whereas a staircase behavior has been regarded as a signature of charge storage in the redox molecules,
which has discrete energy levels corresponding to the various
MOs [5]. Saturation at high positive biases can be a result of
Frenkel–Poole (F–P) leakage through the control dielectric or
the Coulomb blockade effect [5]. The charge neutrality level
(CNL) of Al2 O3 deposited by ALD is around −5.2 eV [20].
The CNL can be regarded as a local Fermi level of the interface
states or metal-induced gap states [21], which are dangling
bonds that disperse across the band gap of the dielectric. It is
evident that the energy-level alignment between the MOs and
the surrounding dielectric’s CNL is crucial for determining the
memory properties [5], [20], [21]. Coulomb staircase behavior
at negative gate biases was not observed for several possible
reasons: First, holes have a preference to relax to the CNL,
which is slightly above the HOMO energy level. Second, upon
interface formation, the energy level may have been shifted due
to fractional charge transfer at the interface. Further evidence
comes from the fact that the control sample with only Bz molecules also results in hole storage with a slightly smaller memory
window than samples with FcCOOH molecules. Knowing that
Bz molecules do not exhibit any redox states, injected holes are
most likely stored in the interfacial traps created by the dangling
bonds. This indicates that both FcCOOH and Bz molecules
generate traps at the dielectric interface as one would expect,
which are the preferential sites for hole storage.
Moreover, the preference for hole storage can be further
explained by the Fermi-level pinning theory [5], [13], [22].
Fig. 7 illustrates the band diagram with an FcCOOH molecule
as the storage node. Electron injection is forbidden at low
programming voltages because electrons must have an energy
greater than or equal to ELUMO − Ec + ∆Ech,e = 2.35 eV +
∆Ech,e , where Ec is the silicon conduction band, and ∆Ech,e
is the single-electron charging energy. Upon injection, electrons would preferentially relax to the interface traps near the
CNL with lower energy. From the energy alignment, the energy required for hole injection is Ev − EHOMO + ∆Ech,h =
−0.66 eV + ∆Ech,h , where Ev is the silicon valence band, and
∆Ech,h is the single-hole charging energy. Thus, hole storage
is energetically favorable than electron storage.
Apart from this, Fig. 8 illustrates the change in memory
window for different mixture ratios between FcCOOH and Bz
at negative gate biases. For this sample, the forming gas anneal
was performed at 300 ◦ C for 90 min with 3 nm of thermal
SiO2 as the tunneling oxide. The amount of ∆VFB increases
proportionally with the increase in the density of FcCOOH
molecules in the SAM, which suggests that increasing the number of FcCOOH molecules increases the number of sites for
hole storage. This also confirms our previous argument, derived
from the energy band diagram, that holes have preference to be
stored at interfacial trap sites created by FcCOOH molecules.
To pinpoint the cause of ∆VFB saturation at positive
gate biases and the charge storage mechanism, we decreased
the density of FcCOOH molecules and performed HFCV
830
IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 58, NO. 3, MARCH 2011
Fig. 9. (a) HFCV and (b) ∆VFB as a function of the programming voltage at
10 K and room temperature for FcCOOH molecules. The two mixture ratios between FcCOOH and Bz molecules are 1 : 200 and 1 : 100 with a tunneling oxide
of 3-nm thermal SiO2 . A 60-min forming gas anneal was performed at 200 ◦ C.
Measurements taken on the sample with no molecules are also shown for
comparison. HFCV measurements were taken at 100 kHz. The programming
voltage Vp for electron and hole charging is applied from +5 to +30 V
and −5 to −18 V, respectively. Subsequently, the gate voltage is swept from
accumulation to inversion for electron charging and inversion to accumulation
for hole charging. The size of the MOS capacitor is 200 × 200 µm2 .
Fig. 7. Energy band diagram representation of the capacitor structure with
FcCOOH molecules. The CNL of Al2 O3 is shown, and the FcCOOH molecule
is assumed to be in its neutral state.
jection and CV stretch-out are no longer present, possibly due
to interface dipole formation, which shifted the band alignment.
At 10 K, the amount of ∆VFB increases linearly in proportion to
the positive gate biases, whereas a saturation behavior was observed at room temperature. In addition, the saturation voltage
is approximately the same for the two different mixture ratios.
Both the temperature and concentration dependences suggest
that the saturation behavior originates from F–P conduction
through the control dielectric at room temperature [5], which
is a nonideal situation in our present process integration.
∆VFB can be described by the following relation:
∆VFB ≈
Fig. 8. ∆VFB as a function of the programming voltage for devices with
FcCOOH molecules mixed with Bz molecules at mixture ratios of 1 : 100, 1:10,
and FcCOOH molecules only in the deposition solution. A 90-min forming gas
anneal was performed at 300 ◦ C. Measurements taken on the sample with no
molecules are also shown for comparison.
measurements at cryogenic temperatures. A new batch of samples (with 3 nm of SiO2 as the tunneling oxide) was annealed
at 200 ◦ C for 60 min, and the mixture ratios between FcCOOH
and Bz molecules were 1 : 100 and 1 : 200 (e.g., 0.005 mM of
FcCOOH to 1 mM of Bz), which translates to the intermolecular distance for FcCOOH of 5.8 and 8.2 nm, respectively,
assuming that the surface coverage ratio is the same as the bulk
solution for preparing the SAM. The concentration of FcCOOH
molecules is reduced to moderate the Coulomb repulsion force,
which could affect the charging potential of neighboring molecules. A lower annealing temperature was chosen to avoid
molecule degradation.
In Fig. 9, HFCV and ∆VFB as a function of the programming
voltage are shown. In Fig. 9(a), we notice that asymmetric in-
Q
q×n×N
≈
Ccont
Ccont
(2)
where Q, q, n, N , and Ccont are the total stored charge density
in the molecules, the elemental charge, the number of charges
per molecule, the number density of the molecule, and the
capacitance of the control oxide, respectively. The ∆VFB shift
is approximately 1.3 V in Fig. 9, and from (2), we obtain a trap
density of 2 × 1012 cm−2 . Once all the traps are occupied, any
additional electron would need to overcome the repulsion force
from the electrons stored in the lowest energy state of interface
traps to enter the MO. As we increase the gate bias further, the
electrons would gain enough energy against the repulsion force
but tends to leak out of the control dielectric by F–P conduction.
There is one more interesting observation. Modeling the
molecule as a thin dielectric layer, we can obtain the total
voltage drop from the SiO2 /Si interface to the central redoxactive atom Vch as
Vch =
εox
εM tm
VG
εox
εox
εM tM + εcont tcont
ttunn +
ttunn +
(3)
where ttunn , tcont , and tm are the tunnel oxide thickness
(SiO2 ), the physical thickness of the Al2 O3 control dielectric,
and the approximate distance from the attachment site to
the redox-active transition metal, respectively, as illustrated
in Fig. 2. tm is estimated to be 0.5tM . εM and tM are the
dielectric constant and the thickness of the molecule, respectively, assuming that the silicon channel is under inversion or
SHAW et al.: INTEGRATION OF SAMs IN FLASH MEMORY DEVICES
831
Fig. 10. (a) HFCV and (b) ∆VFB as a function of the programming voltage
at 10 K and room temperature for the device with CoP molecules. The mixture
ratio between CoP and H2 P is 1 : 20. A 30-min forming gas anneal was
performed at 400 ◦ C. Measurements taken on the sample with no molecules
are also shown for comparison. HFCV measurements were taken at 100 kHz.
The programming voltage Vp for electron and hole charging is applied from 0
to +42 V and 0 to −18 V, respectively. Subsequently, the gate voltage is swept
from accumulation to inversion for electron charging and inversion to accumulation for hole charging. The size of the MOS capacitor is 200 × 200 µm2 .
accumulation. εM and tM were determined in previous reports
through impedance characterization to be 2.4 and 0.8 nm,
respectively [8]. εcont = 9 and εox = 3.9 are the dielectric
constants of the control and tunneling oxides, respectively. In
Figs. 6(b) and 9(b), the total blockade voltage can be calculated
using (3) to be 2.1 eV in both cases, assuming that the energy
required to initiate charge injection is approximately the same
as eVch , with a shift in the absolute value for different annealing
conditions. The shift indicates that the slightly different band
alignment is due to the interfacial dipoles and the fixed charge at
the molecule/Al2 O3 interface, but the HOMO–LUMO energy
gap is identical in both cases. Depending on the annealing
condition, the density of interface traps will vary as a function
of energy, which leads to the difference in flatband voltage.
However, the consistency in the blockade voltage suggests
that the MO energy spacing is preserved within the dielectric
environment.
If the molecule density is high enough to enable free lateral transport (continuous floating gate), ∆Ech will not be
significant, and the discrete energy levels of the molecule
would not be observable. The dilute FcCOOH/Bz samples
have a total blockade energy of 2.1 eV, which indicates that
the charging energy remains negligible after diluting the concentration of FcCOOH molecules. Therefore, it is likely that
the carriers were injected into the interfacial traps since significant trap states are involved at the FcCOOH/Al2 O3 or
the Bz/Al2 O3 interface. The constant plateau regions suggest
that the HOMO–LUMO energy gap acts as the energy barrier
that carriers must overcome to initiate electron/hole injection.
However, carriers quickly relax to surrounding interface states
to favor the lower CNL energy.
Next, we consider the electrical characterization for the CoP
redox-active molecules. Fig. 10(a) and (b) shows the HFCV and
memory window as a function of the programming voltage,
respectively. Gate injection prohibits hole injection beyond
programming voltages below −18 V. On the contrary, electron
injection shows two distinct levels of the Coulomb staircase,
which is in good agreement with the previous measurements using CyV [6]–[9]. Assuming that full coverage density of 4.5 ×
1013 cm−2 and 1:20 dilution of the deposition solution lead to
Fig. 11. Energy band diagram at thermal equilibrium states (a) before and
(b) after a CoP molecule is in contact with the surrounding dielectric. The CNL
and density of states are determined by the oxide’s atomic configuration, which
are sensitive to composition stoichiometries and deposition surface. The figure
is not drawn to scale. The electron-filled interface states are indicated as the
shaded regions. The HOMO–LUMO energy gap is estimated from the UV/Vis
absorption spectra to be ∼2.65 eV.
equal dilution of the SAM, the 1:20 mixture ratio with H2 P
translates to a CoP number density of about 2.2 × 1012 cm−2
and a flatband voltage shift of about 1.32 V for the singleelectron injection. This value matches well with the amount of
flatband voltage shift observed in Fig. 10, which is approximately 1.4 V at room temperature and 1.6 V at 10 K. At room
temperature, the large leakage between the carrier storage
sites and the gate, possibly through hopping, prevents efficient
charging of CoP molecules. Contrarily, measurements that were
taken at 10 K also show Coulomb staircase behavior, but
carrier storage begins at lower programming voltages due to
the elimination of F–P leakage. It is worthwhile to note that the
initial flatband voltage is ∼ +3.5 V, which differs from an ideal
p-type MOS stack, which exhibits a flatband voltage of
∼ −0.9 V. The difference in flatband voltage is from the large
amount of fixed charge created at the molecule/Al2 O3 interface
due to the nonideal growth surface.
Fig. 11(a) and (b) illustrates the energy band diagrams before
and after depositing Al2 O3 as the control dielectric on top of
the molecules, respectively. The HOMO–LUMO energy levels
for neutral CoP0 were estimated by DFT calculation [19],
whereas the CNL of ALD Al2 O3 is −5.2 eV. According to the
HOMO–LUMO energy levels of CoP0 and CNL, electrons have
the energy preference to relax to the CNL, and the Coulomb
staircase would not be observed before the control dielectric
is deposited. On the other hand, the interface dipole formation
at the CoP/Al2 O3 interface can lead to a different thermal
equilibrium state, likely a monocation (CoP1+ ), established
by hole transferring from the interface states into CoP0 . In
addition, electron injection to the LUMO energy level is evident
from the Coulomb staircase. Therefore, we expect the CNL to
be in close proximity with the LUMO so that injected carriers
832
IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 58, NO. 3, MARCH 2011
will not have the tendency to relax to the interfacial trap states.
Electron injection into the CoP1− states to become a dianion
(CoP2− ) is not favorable until the LUMO energy is one ∆Ech,e
below the silicon conduction band edge, which corresponds
to the 5-eV blockade voltage at room temperature. Sequential
electron injection can be stable in the orbital states, as long
as the leakage back to the substrate is not severe. The singleelectron charging energy can be calculated by
∆Ech,e =
q2
C
(4)
where C is the self-capacitance of the storage node. However,
the charging energy can only be estimated using a detailed
calculation considering the complex geometry of the redox
molecule embedded in the dielectric, which is not performed
here. Instead, we will calculate the charging energy by first
determining the self-capacitance by the following relation:
Fig. 12. (a) Programming and (b) retention measurements of MOS capacitors
with FcCOOH and CoP molecules. ∆VFB was extracted from HFCV measurements. Devices were stressed for 5 s at +10 V before the retention measurement
and programming tests were applied at +10 V. The sample with FcCOOH
molecules has a tunneling oxide of 3-nm thermally grown SiO2 , and a 60-min
forming gas anneal was performed at 200 ◦ C. The mixture ratio between
FcCOOH and Bz is 1 : 200. The sample with CoP molecules has a heterogeneous
tunnel oxide composed of 1.4-nm native SiO2 and 3-nm plasma ALD Al2 O3 .
IV. C ONCLUSION
Cself = CCoP × ACoP
(5)
where Cself , CCoP , and ACoP are the self-capacitance, the
capacitance per unit area, and the areal coverage of each CoP
molecule, respectively. The capacitance per unit area can be
extracted from impedance spectroscopy, which is 1.5 µF/cm2
for the device with CoP molecules, estimated from previous
literature [8]. ACoP is estimated from the amount of ∆VFB ,
which separates each reduced states.
In Fig. 10, the amount of ∆VFB between the plateau regions
is ∼1.5 V at 10 K. In (2), we calculate the number density N
to be 2.5 × 1012 cm−2 , which translates to the unit cell area
Acell of 4 × 10−13 cm2 . Assuming a close-packed SAM and
the mixture ratio in a liquid solution to be the same as the asdeposited molecules, the areal coverage of each CoP molecule
ACoP is 2 × 10−14 cm2 , which can be calculated by the relation
Acell /ACoP ∼ 20, considering the mixture ratio of 1:20 with
H2 P. In (5), the self-capacitance of the CoP molecule can be
readily calculated to be 3 × 10−20 F.
In (4), the single-electron charging energy ∆Ech,e can be
calculated to be ∼5.6 eV. The εM , tm , and tM values for the
CoP SAM are 2.5, 0.8 nm, and 1.6 nm, respectively. The ttunn
of the heterogeneous tunneling oxide is 2.7 nm. With these
molecular parameters, the first plateau region in Fig. 10 has a
blockade voltage of ∼5.3 eV at 10 K by (3), which is in good
agreement with the single-electron charging energy of the CoP
molecule (5.6 eV).
Finally, Fig. 12 shows the programming and retention characteristics for FcCOOH and CoP molecules. The retention
time tR /programming time tP ratio of the device with CoP
molecules is clearly improved compared with the device with
FcCOOH molecules as the storage node, possibly due to the
large capture cross section of the redox molecules in comparison with the interfacial trap states in the device with FcCOOH
molecules. Further improvement on program efficiency, voltage
operation, and retention is possible by adjusting the oxide
thickness ratio, tunnel oxide thickness, and surface coverage
density of the molecules; integrating different redox molecules
with various orbital energy levels; and improving the dielectric
quality.
We have successfully integrated a monolayer of redox molecules in a Flash memory device structure using a solution-based
self-assembly technique and demonstrated three programmable
MO states of the CoP molecule, including CoP0 , CoP1− , and
CoP2− , at room temperature, which may help realize step
charging into a multibit memory cell. For the device with
FcCOOH molecules, the memory window increases proportionally with the density of the redox molecules, and the band
offset of the HOMO–LUMO energy levels of FcCOOH molecules and the CNL of Al2 O3 determines the preferred carrier
storage sites. With the abundant choices of redox molecules
and their inherent monodispersion in size and energy levels,
our proposed approach can be readily integrated into a MOSbased nonvolatile memory cell and pave the wave for realizing
multilevel molecular memory devices. Furthermore, mixture of
porphyrins with different transitional metals or integration of
multistate molecules [23] may offer additional MO states while
maintaining a small bit error rate.
R EFERENCES
[1] T.-H. Hou, C. Lee, V. Narayanan, U. Ganguly, and E. C. Kan, “Design
optimization of metal nanocrystal memory—Part II: Gate-stack engineering,” IEEE Trans. Electron Devices, vol. 53, no. 12, pp. 3103–3109,
Dec. 2006.
[2] C. Lee, U. Ganguly, V. Narayanan, T.-H. Hou, and E. C. Kan, “Asymmetric electric field enhancement in nanocrystal memories,” IEEE Electron
Device Lett., vol. 26, no. 12, pp. 879–881, Dec. 2005.
[3] L. Perniola, B. De Salvo, G. Ghibaudo, A. F. Para, G. Pananakakis,
V. Vidal, T. Baron, and S. A. Lombardo, “Modeling of the programming
window distribution in multi-nanocrystal memory,” IEEE Trans. Nanotechnol., vol. 2, no. 4, pp. 277–284, Dec. 2003.
[4] J. T. Shaw, T.-H. Hou, H. Raza, and E. C. Kan, “Statistical metrology
of metal nanocrystal memories with 3-D finite-element analysis,” IEEE
Trans. Electron Devices, vol. 56, no. 8, pp. 1729–1735, Aug. 2009.
[5] T.-H. Hou, U. Ganguly, and E. C. Kan, “Programmable molecular orbital
states of C60 from integrated circuits,” Appl. Phys. Lett., vol. 89, no. 25,
pp. 253 113-1–253 113-3, Dec. 2006.
[6] Q. Li, S. Surthi, G. Mathur, G. Gowda, T. A. Sorenson, R. C. Tenent,
W. G. Kuhr, S.-I. Tamaru, J. S. Lindsey, Z. Liu, D. F. Bocian, and
V. Misra, “Electrical characterization of redox-active molecular monolayers on SiO2 for memory applications,” Appl. Phys. Lett., vol. 83, no. 1,
pp. 198–200, Jun. 2003.
[7] Q. Li, G. Mathur, M. Homsi, S. Surthi, V. Misra, V. Malinovskii,
K. H. Schweikart, L. H. Yu, J. S. Lindsey, Z. M. Liu, R. B. Dabke,
A. Yasseri, D. F. Bocian, and W. G. Kuhr, “Capacitance and conductance
SHAW et al.: INTEGRATION OF SAMs IN FLASH MEMORY DEVICES
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
characterization of ferrocene-containing self-assembled monolayers on
silicon surfaces for memory applications,” Appl. Phys. Lett., vol. 81, no. 8,
pp. 1494–1496, Aug. 2002.
Q. Li, “Approach towards hybrid silicon/molecular electronics for memory applications,” in Electrical and Computer Engineering. Raleigh,
NC: North Carolina State Univ., 2004, p. 152.
G. Mathur, S. Gowda, Q. Li, S. Surthi, Q. Zhao, and V. Misra, “Properties
of functionalized redox-active monolayers on thin silicon dioxide—A
study of the dependence of retention time on oxide thickness,” IEEE
Trans. Nanotechnol, vol. 4, no. 2, pp. 278–283, Mar. 2005.
S. Kubatkin, A. Danilov, M. Hjort, J. Cornil, J.-L. Bredas,
N. Stuhr-Hansen, P. Hedegard, and T. Bjornholm, “Single-electron
transistor of a single organic molecule with access to several redox
states,” Nature, vol. 425, no. 6959, pp. 698–701, Oct. 2003.
Z. Liu, A. A. Yasseri, J. S. Lindsey, and D. F. Bocian, “Molecular memories that survive silicon device processing and real-world operation,”
Science, vol. 302, no. 5650, pp. 1543–1545, Nov. 2003.
C. Lee, J. Meteer, V. Narayanan, and E. C. Kan, “Self-assembly of metal
nanocrystal on ultra-thin oxide for nonvolatile memory applications,”
J. Electron. Mater., vol. 34, no. 1, pp. 1–11, Jan. 2005.
T.-H. Hou, U. Ganguly, and E. C. Kan, “Fermi-level pinning in nanocrystal memories,” IEEE Electron Device Lett., vol. 28, no. 2, pp. 103–106,
Feb. 2007.
A. S. Killampalli, P. F. Ma, and J. R. Engstrom, “The reaction of
tetrakis(dimethylamido)-titanium with self-assembled alkyltrichlorosilane monolayers possessing –OH, −NH2 , and −CH3 terminal groups,”
J. Am. Chem. Soc., vol. 127, no. 17, pp. 6300–6310, May 2005.
N. W. Ashcroft and N. D. Mermin, Solid State Physics. Fort Worth, TX:
Harcourt Brace College, 1976.
C. J. Powell and A. Jablonski, “Consistency of calculated and measured
electron inelastic mean free paths,” J. Vac. Sci. Technol. A., vol. 17, no. 4,
pp. 1122–1126, Jul. 1999.
J. H. Scofield, “Hartree–Slater subshell photoionization cross-sections at
1254 and 1487 eV,” J. Electron Spectrosc., vol. 8, no. 2, pp. 129–137,
1976.
M. G. Grigorov, J. Weber, N. Vulliermet, H. Chermette, and
J. M. J. Tronchet, “Numerical evaluation of the internal orbitally resolved
chemical hardness tensor: Second order chemical reactivity through
thermal density functional theory,” J. Chem. Phys., vol. 108, no. 21,
pp. 8790–8798, Jun. 1998.
M.-S. Liao and S. Scheiner, “Electronic structure and bonding in metal
porphyrins, metal = Fe, Co, Ni, Cu, Zn,” J. Chem. Phys., vol. 117, no. 1,
pp. 205–219, Jul. 2002.
S. B. Samavedam, L. B. La, P. J. Tobin, B. White, C. Hobbs,
L. R. C. Fonseca, A. A. Demkov, J. Schaeffer, E. Luckowski, A. Martinez,
M. Raymond, D. Triyoso, D. Roan, V. Dhandapani, R. Garcia,
S. G. H. Anderson, K. Moore, H. H. Tseng, C. Capasso, O. Adetutu,
D. C. Gilmer, W. J. Taylor, R. Hegde, and J. Grant, “Fermi level pinning
with sub-monolayer MeOx and metal gates,” in IEDM Tech. Dig., 2003,
pp. 13.1.1–13.1.4.
J. Tersoff, “Theory of semiconductor heterojunctions: The role of quantum dipoles,” Phys. Rev. B, vol. 30, no. 8, pp. 4874–4877, Oct. 1984.
S. Lenfant, D. Guerin, F. Tran Van, C. Chevrot, S. Palacin, J. P. Bourgoin,
O. Bouloussa, F. Rondelez, and D. Vuillaume, “Electron transport through
rectifying self-assembled monolayer diodes on silicon: Fermi-level pinning at the molecule–metal interface,” J. Phys. Chem., vol. 110, no. 28,
pp. 13 947–13 958, Jul. 2006.
D. Gryko, J. Z. Li, J. R. Diers, K. M. Roth, D. F. Bocian, W. G. Kuhr, and
J. S. Lindsey, “Studies related to the design and synthesis of a molecular
octal counter,” J. Mater. Chem., vol. 11, no. 4, pp. 1162–1180, Feb. 2001.
833
Yu-Wu Zhong received the Ph.D. degree from the Chinese Academy of
Sciences, Beijing, China, in 2004, under the supervision of Prof. G.-Q. Lin.
He was a Postdoctoral Associate with Prof. E. Nakamura at the University of
Tokyo, Tokyo, Japan, from 2004 to 2006 and with Prof. H. D. Abruña at Cornell
University, Ithaca, NY, from 2006 to 2009. He is currently a Professor with the
Laboratory of Photochemistry, Institute of Chemistry, Chinese Academy of Sciences. His research interests include the synthesis of functional organometallic
materials and their application for molecular electronics and photovoltaics.
Kevin J. Hughes received the B.S. degree in chemical engineering from
Carnegie Mellon University, Pittsburgh, PA, in 2003. He is currently working
toward the Ph.D. degree with the School of Chemical and Biomolecular Engineering, Cornell University, Ithaca, NY. His thesis research involves the growth
of inorganic materials on thin organic layers using atomic layer deposition.
He spent two years as a Research Engineer with Lubrizol Corporation.
Tuo-Hung Hou received the B.S. and M.S. degrees from National Chiao
Tung University, Hsinchu, Taiwan, in 1996 and 1998, respectively, and the
Ph.D. degree from Cornell University, Ithaca, NY, in 2008, all in electrical
engineering.
From 1998 to 2000, he served as a Second Lieutenant in the Taiwanese
Army. In April 2000, he joined the Advanced Module Technology Division,
Taiwan Semiconductor Manufacturing Company (TSMC), where he worked
in the area of deep-submicrometer front-end process development. From 2001
to 2003, he was also a TSMC Assignee with the International Semiconductor
Manufacturing Technology, Austin, TX, engaged in high-κ gate dielectric
research for two years. He is currently an Assistant Professor with the School
of Electrical and Computer Engineering, National Chiao Tung University. His
current research interests include novel devices by nanocrystal and nanotube
integration.
Hassan Raza (M’07) received the B.S. degree (with honors) from the University of Engineering and Technology, Lahore, Pakistan, in 2001 and the M.S. and
Ph.D. degrees from Purdue University, West Lafayette, IN, in 2002 and 2007,
respectively, all in electrical engineering.
In May 2007, he joined the Center for Nanoscale Systems, Cornell University, Ithaca, NY, as a Postdoctoral Associate. He is currently an Assistant
Professor with the College of Engineering, University of Iowa, Iowa City.
His research interests include the experimental, theoretical, and computational
aspects of quantum transport for novel logic and memory devices in various
material systems, e.g., graphene, carbon nanotubes, molecules, and quantum
dots.
Dr. Raza was a recipient of the Magoon Award for Excellence in Teaching
from Purdue University in 2004 and the Presidential Faculty Fellowship of the
University of Iowa in 2010.
Shantanu Rajwade received the B.Tech. degree in electrical engineering and
the M.Tech. degree in microelectronics from the Indian Institute of Technology,
Mumbai, India, in 2007. He is currently working toward the Ph.D. degree in
electrical and computer engineering with Cornell University, Ithaca, NY.
His research interests include device physics, nanoscale carrier transport,
low-power logic and memory devices, and their applications to nonvolatile
computing in future embedded system-on-a-chip platforms.
Jonathan Shaw received the B.S. degree in electrical engineering from the
University of California, San Diego, La Jolla, in 2006. He is currently working
toward the Ph.D. degree with the School of Electrical and Computer Engineering, Cornell University, Ithaca, NY.
His current research interests include nanocrystal and molecular-based memory devices.
Julie Bellfy is currently working toward the B.S. degree in chemical engineering with Villanova University, Villanova, PA.
Her current research interests include metal–oxide–semiconductor capacitors, thin films, and monolayers.
834
James R. Engstrom received the B.S. degree in chemical engineering from
the University of Minnesota, Minneapolis, in 1981 and the Ph.D. degree in
chemical engineering from the California Institute of Technology, Pasadena,
in 1987.
From 1998 to 2001, he was with Symyx Technologies, as the Director and,
eventually, the Vice President of High-Throughput Screening and Electronic
Materials, where he spearheaded the development of new technologies for highthroughput screening. In 2001, he joined Cornell University, Ithaca, NY. Since
2002, he has been a member of the Graduate Field of Chemistry and Chemical
Biology, Cornell University. He is currently the B. P. Amoco/H. Laurance Fuller
Professor with the School of Chemical and Biomolecular Engineering, Cornell
University. He is the author of more than 70 peer-reviewed publications and
has presented his work at more than 180 national and international meetings,
as well as to private industry. He is the holder of ten patents. He is widely
recognized for his work concerning supersonic molecular beam scattering of
thin-film precursors from semiconductor surfaces and fundamental studies of
thin-film deposition, making use of precisely controlled beams of molecular
and/or atomic species, with applications in silicon-based microelectronics. His
current research focuses on two areas, namely, inorganic–organic interfaces
and organic thin-film electronics. Concerning the former, he is actively engaged in research that seeks to build robust interfaces between organic thin
films/monolayers and inorganic (dielectric or metallic) thin-film overlayers.
This topic has applications in the areas of molecular electronics and barrier
layers for microelectronics interconnect and packaging. Concerning the latter,
his group is currently aggressively pursuing fundamental studies of smallmolecule organic thin films, such as pentacene. In this work, in addition to
using molecular-beam-based techniques, his group is also making use of realtime in situ X-ray scattering that gives unprecedented information concerning
the dynamics of growth.
Dr. Engstrom is a Fellow of the American Vacuum Society. He was a
recipient of numerous awards, including the National Science Foundation Presidential Young Investigator Award and two College of Engineering Teaching
Awards.
Héctor D. Abruña received the Ph.D. degree, with R. W. Murray and
T. J. Meyer, from the University of North Carolina at Chapel Hill in 1980.
He was a Postdoctoral Research Associate with A. J. Bard at the University
of Texas at Austin. After a brief stay at the University of Puerto Rico, San Juan,
Puerto Rico, in 1983, he joined Cornell University, Ithaca, NY, where he is
currently an Emile M. Chamot Professor of Chemistry.
Prof. Abruña is a Fellow of the American Association for the Advancement
of Science. He was a recipient of the Presidential Young Investigator Award,
the Alfred P. Sloan Foundation Research Fellowship, the John S. Guggenheim
Fellowship, the Tajima Prize of the International Society of Electrochemistry,
the J. W. Fulbright Senior Research Fellowship, and the Iberdrola Fellowship.
IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 58, NO. 3, MARCH 2011
Edwin Chihchuan Kan (S’86–M’91–SM’05) received the B.S. degree from
National Taiwan University, Taipei, Taiwan, in 1984 and the M.S. and Ph.D.
degrees from the University of Illinois at Urbana-Champaign in 1988 and 1992,
respectively, all in electrical engineering.
In January 1992, he joined Dawn Technologies as a Principal ComputerAided Design (CAD) Engineer, developing advanced electronic and optical
device simulators and technology CAD frameworks. From 1994 to 1997, he
was with Stanford University, Palo Alto, CA, as a Research Associate. From
1997 to 2002, he was an Assistant Professor with the School of Electrical and
Computer Engineering, Cornell University, Ithaca, NY, where he is currently a
Professor. He spent the summers of 2000 and 2001 at IBM Microelectronics,
Yorktown Heights and Fishkill, NY, in the Faculty Partner Program. In 2004
and 2005, he was a Visiting Researcher with Intel Research, Santa Clara, CA,
and a Visiting Professor with Stanford University during his sabbatical leave.
His main research areas include complementary metal–oxide–semiconductor
(CMOS) technology, semiconductor device physics, Flash memory, CMOS
sensors, ultralow-power radio link, composite CAD, and numerical methods
for partial and ordinary differential equations.
Dr. Kan was a recipient of the Presidential Early Career Award for Scientists
and Engineer in October 2000 from the White House, as well as several
teaching awards from Cornell Engineering College for his CMOS and microelectromechanical system courses.