Platinum nanoparticles grown by atomic layer deposition for charge storage memory applications

Article abstract of DOI:10.1149/1.3365031

This paper explores platinum nanoparticle formation during the early stages of growth by atomic layer deposition. Particle size and distribution can be controlled by altering growth parameters. The particles show excellent temperature stability up to 900°

Full text of DOI:10.1149/1.3365031

Journal of The Electrochemical Society, 157 ͑6͒ H589-H592 ͑2010͒
H589
0013-4651/2010/157͑6͒/H589/4/$28.00 © The Electrochemical Society
Platinum Nanoparticles Grown by Atomic Layer Deposition
for Charge Storage Memory Applications
z
*
Steven Novak, Bongmook Lee, Xiangyu Yang, and Veena Misra
Department of Electrical and Computer Engineering, North Carolina State University, Raleigh,
North Carolina 27695, USA
This paper explores platinum nanoparticle formation during the early stages of growth by atomic layer deposition. Particle size and
distribution can be controlled by altering growth parameters. The particles show excellent temperature stability up to 900°C as
examined by transmission electron microscopy and in situ heating. Capacitance–voltage and charge retention measurements
demonstrate the memory effect in metal-oxide-semiconductor capacitors with embedded nanoparticles. The size, density, charge
storage, and temperature stability of the platinum nanoparticles make them attractive for use as charge storage layers for non-
volatile memory devices.
© 2010 The Electrochemical Society. ͓DOI: 10.1149/1.3365031͔ All rights reserved.
Manuscript submitted November 13, 2009; revised manuscript received February 22, 2010. Published April 19, 2010.
As charge storage memory densities increase and device dimen-
sions decrease, there is a need to replace floating polysilicon gates
with discrete charge storage materials. Memory technologies utiliz-
ing discrete charge storage layers such as silicon nitride, metal nano-
particle, or semiconducting nanoparticles are expected to lessen the
impact of localized oxide defects, lateral coupling of charge storage
layers between adjacent devices, and stress-induced leakage
current.1,2 Nanoparticle-based memories have reduced lateral con-
ductions over nitride based memories due to the coulomb blockade
effect, which is present for nanoscale charged particles.3 Addition-
ally, metal nanocrystals have higher density of states than semicon-
ducting nanoparticles, allowing higher charge storage densities and
faster read/write times.4 High work function metals, such as Au, Pt,
Ni, and Ru, are being looked at for nonvolatile memory
applications5,6 partly because they create additional depth to the po-
tential well formed in between the blocking and inter-
polydielectrics, which allows longer retention times in memory de-
vices. Pt and Ru, in particular, show low diffusivities and high
thermal stabilities in conventional complementary metal oxide semi-
conductor processing, making them particularly attractive for charge
storage memory technologies.
Achieving long-range uniformity and high densities of nanocrys-
tal growth remains one of the largest hurdles to large-scale manu-
facturing of nanocrystal memories. Atomic layer deposition ͑ALD͒
processes for thin films have the advantage of monolayer growth
precision over a large area and the ability to deposit conformally
over high aspect ratio features. This uniformity and control over
small dimensions of ALD are also attractive for nanoparticle forma-
tion for memory applications. Growth of Ru nanocrystals by ALD
has been shown previously, but high cycle counts and low densities
make this process undesirable for commercial memory
applications.7 A hybrid chemical vapor deposition ͑CVD͒/ALD pro-
cess has also been used to grow Ru nanocrystals; however, the
growth window is very small due to the CVD aspect of crystal
growth, and uniformity and densities on SiO2 are undesirable for
memory applications.8 The recent development of a Pt thin-film
ALD process has opened the possibility of adapting it for use in
nanocrystal growth.9 In this study, we have investigated the initial
stages of Pt ALD to form nanoparticles that are attractive for non-
volatile charge storage memories. We have explored the role of the
starting surface, the ALD deposition conditions, and the top capping
layer on the size, density, and stability of Pt nanoparticles.
O2 ͑SiO2͒ or deposited by ALD of trimethylaluminum and H2O
͑Al2O3͒. Pt was deposited by ALD using ͑methylcyclopentadienyl͒
trimethyl-platinum ͑MeCpPtMe3͒ and ultrahigh purity O2. To exam-
ine Pt growth, the platinum ALD process was performed on Al2O3
and SiO2 dielectrics with cycles ranging from 5 to 50 cycles in steps
of five cycles at 270°C. Unless otherwise stated, the Pt ALD was
done with a 0 s exposure time for the Pt precursor. With 0 s expo-
sure, the stop valve between the reaction chamber and the vacuum
pump of the ALD tool was continuously opened and pumped. For
the ALD samples that received exposure, the stop valve was closed
before the precursor was introduced into the chamber, and the pre-
cursor was allowed to react with the substrate for a longer period of
time.
Physical characterization was done by atomic force microscopy
͑AFM͒ and transmission electron microscopy ͑TEM͒, and chemical
analysis was done by X-ray photoelectron spectroscopy ͑XPS͒.
Capacitors were fabricated by thermal oxidation of Si ͑6 nm͒ on
RCA-cleaned p-type silicon wafers. Pt nanoparticles were grown on
the thermal SiO2 using 35 cycles of the Pt ALD process. Thirty five
cycles was chosen as the cycle count where densities and sizes were
relatively high, but there was still a window before the films were
continuous. HfO2 was deposited by ALD ͑ϳ14.4 nm͒ to act as the
inter-polyoxide. The gate was formed by depositing 90 nm of W in
an ultrahigh vacuum radio-frequency sputtering system and pat-
terned into 100 ϫ 100 ␮m capacitors by photolithography and wet
etching. Samples were annealed for 5 min in 5% H2/N2 at 450°C to
reduce charges in the dielectric. Capacitance vs voltage ͑C-V͒ was
measured with an HP4284 LCR, current vs voltage measurements
were made with an HP4155b semiconductor parameter analyzer, and
the physical parameters were extracted by least-squares curve-fitting
using the NCSU CVC program.10
Results and Discussion
The ALD process for Pt as well as other noble metals is depen-
dent on the presence of either oxygen or hydrogen as a reactant for
the precursor.9,11 The Pt ALD process used in this study consists of
a pulse of O2 reacting with the surface layer of Pt to create PtOx.
Then, a pulse of MeCpPtMe3 reacts with the PtOx layer, and during
the subsequent purge, O atoms are removed, leaving only Pt. On the
next cycle, the O2 pulse removes any unreacted precursor and oxi-
dizes the Pt surface, preparing it for the next cycle. Previous reports
of this mechanism assume a Pt surface to start with, and conformal
growth.9,12 The nucleation of the Pt ALD process, however, is very
dependent on the starting surface conditions. The presence of oxy-
gen at the deposition surface determines how Pt nucleates in the
ALD process.
Experimental
Both SiO2 and Al2O3 dielectrics were grown on ͑100͒-oriented
p-type silicon wafers. Dielectrics were either grown thermally in dry
In our study, the nucleation and initial growth of Pt were inves-
tigated on both SiO2 and Al2O3 as these are potential tunnel dielec-
trics for memory devices. AFM images ͑Fig. 1͒ show that for both
dielectrics there is an apparent delay in the nucleation of the Pt film
*
Electrochemical Society Student Member.
z E-mail: vmisra@ncsu.edu
Downloaded on 2015-01-06 to IP 130.194.20.173 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).

H590
Journal of The Electrochemical Society, 157 ͑6͒ H589-H592 ͑2010͒
Figure 3. TEM image of Pt nanoparticles grown from 30 cycles Pt ALD on
Al2O3. With continuous pumping of the ALD precursor, small particle sizes
are easily obtained but with a larger spread in size distribution. ͑b͒ By al-
lowing the Pt precursor more time to saturate the surface of the dielectric, a
larger but more uniform size distribution is obtained. ͑c͒ Further increase in
precursor exposure leads to minor increases in the particle size. However, the
change in size caused by increasing the exposure time from 10 to 20 s is
much less than the change seen by increasing the exposure time from 0 to
10 s.
Figure 1. ͑Color online͒ AFM image of Pt ALD growth on ͑a͒ SiO2 and ͑b͒
Al2O3 dielectrics. Particles grow both laterally and vertically with increased
ALD cycles. All images are 1 ϫ 1 ␮m scans with 5 nm Z-scale
during the initial cycles. After 20 ALD cycles, the presence of Pt
islands can be detected by AFM. The particles are assumed to be
spherical, with their radius equal to the Z dimension in AFM. Be-
cause the size and spacing of the particles are smaller than typical
AFM tip widths ͑7–10 nm͒, the expected distortion in the XY plane
is observed. With increased ALD cycles, the particle sizes continue
to increase on both dielectrics. As can be seen, the Pt islands on
SiO2 initially grow larger than the islands on Al2O3; however, the Pt
islands on Al2O3 nucleate with a slightly higher density than SiO2.
Although the mechanism causing this difference is still being inves-
tigated, it has been reported that the density of hydroxyl bonds on
the surface of Al2O3 is about 3 times denser than on SiO2
surfaces.13,14 Therefore, one would expect that the nucleation den-
sity of the Pt ALD process would be greater on an Al2O3 surface. At
40 cycles, the particle densities calculated from the AFM images for
Pt on Al2O3 and SiO2 are 3.45 ϫ 1011 and 2.5 ϫ 1011 cm−2, re-
spectively. Indeed, the higher density on Al2O3 is inline with expec-
tations based on the OH densities of the two dielectrics. In addition,
the Al2O3 films in this work were grown by ALD of trimethylalu-
minum and H2O and, thus, are expected to contain a high density of
OH groups.
Root-mean-square ͑rms͒ roughness and particle height as ob-
tained from AFM ͑Fig. 2͒ indicate that the increase in particle den-
sity on Al2O3 dielectrics is accompanied by a decrease in particle
size and sample roughness as compared to the same cycle number
on SiO2. Therefore, the particles on Al2O3 are smaller but have a
higher density than the particles on SiO2. At the 50 cycle mark, the
Pt on Al2O3 already shows a smoother continuous film, which can
be seen both in the AFM images ͑Fig. 1͒ and by a reduction in
variance in the particle height, while the 50 cycles on SiO2 show a
semidiscrete rougher film. The rms roughness data ͑Fig. 2a͒ also
show that between 40 and 50 cycles of Pt on Al2O3, there is a drop
in rms values, further indicating a move from discrete particles to a
continuous layer of Pt. The smoother nature of the Pt film on Al2O3
is a direct consequence of the higher nucleation density. The par-
ticles merge sooner after nucleation, compared to ALD Pt on SiO2,
due to their closer proximity and higher density, and require less
time to grow vertically before they merge, resulting in a smoother
film.
The impact of ALD process parameters on the discontinuous
nature of the films and the particle size was further verified by TEM
imaging ͑Fig. 3a͒. With continuous pumping of the precursor ͑no
exposure time͒, particle sizes agree with particle height data ob-
tained from AFM images. Increasing the precursor exposure to 10 s
results in an increase in the particle size due to the larger reaction
time ͑Fig. 3b͒. This also indicates that the ALD process with no
exposure grows Pt films below the saturated growth regime. Along
with the larger particle sizes in the film with exposure, there is also
a tighter distribution in the sizes. Further increase in the precursor
exposure to 20 s ͑Fig. 3c͒ results in a similar particle distribution as
the 10 s precursor exposure; a slightly larger particle size is ob-
tained, but the size distributions are very similar. Independent of the
precursor exposure, the particle size can be controlled by adjusting
the ALD cycle count so that the increase in particle size obtained
with an increase in exposure can be countered with reduction in
ALD cycles while still maintaining tight particle distributions ͑Fig.
4͒.
To ensure that the deposited nanoparticles were indeed metallic,
XPS was also performed. A Pt 4f peak located at 71.3 eV15 with 20
cycles of Pt ALD on Al2O3 is clearly observed ͑Fig. 5͒, and the
intensity of the Pt peak increases with cycle count, indicating an
increase in Pt concentration and, hence, in the particle size. The
second peak of the Pt 4f doublet is located at 74.23 eV and overlaps
with the Al 2p peak at 74.3eV.15 The increase in the peak height
around 74.3 eV is attributed to the increase in the Pt signal, because
the growth of any film should shorten the X-ray penetration depth
into the Al2O3 film and dampen the XPS signal. Similar XPS results
Figure 2. ͑Color online͒ ͑a͒ RMS rough-
ness data and ͑b͒ particle height vs ALD
cycle count. The lower rms roughness and
particle size can be attributed to a higher
nucleation density.
Downloaded on 2015-01-06 to IP 130.194.20.173 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).

Journal of The Electrochemical Society, 157 ͑6͒ H589-H592 ͑2010͒
H591
Figure 6. ͑Color online͒ 200 Cycles Pt ALD on ͑a͒ SiO2, ͑b͒ Al2O3, and ͑C͒
H-terminated Si. All AFM scans are 1 ϫ 1 ␮m with 5 nm Z-scale. Pt on
SiO2 shows a folded pattern resembling a brain, while Al2O3 shows a
smoother film. Relatively small disbursed particles are found on
H-terminated Si, indicating lack of Pt nucleation.
on various surfaces ͑Fig. 6͒ show the effect of particle nucleations
on continuous films. Although both films are continuous on SiO2
and Al2O3, the difference in initial nucleation leads to a different
morphology of Pt on SiO2 compared to Al2O3. The less densely
nucleated Pt particles on the SiO2 surface grow until they merge,
forming a “brain” pattern with deep channels between the folds. The
Pt films, more densely nucleated on the Al2O3 surface, merge sooner
due to their closer proximity and higher density, resulting in a
smoother, more uniform film with smaller grains. An HF last silicon
sample was also included in the ALD runs ͑Fig. 6c͒. The HF treat-
ment leaves Si–H bonds on the surface of the substrate, which pre-
vents reactions and absorption of the precursor at the samples sur-
face. This inhibits the Pt ALD process from nucleating initially.
Repeated exposure to O2 during ALD cycling eventually creates
disbursed Si–OH bonds at the surface, which act as spots to nucleate
Pt grains. However, even with 200 cycles, continuous Pt deposition
was not achieved on HF last silicon. The large size of the Pt particles
on Si ͑Fig. 6c͒ indicates that a few Pt particles nucleate in the early
cycles and grow with further ALD cycles. With repeated ALD cy-
cling, newer grains start to nucleate, creating relatively smaller par-
ticles. To obtain continuous films on HF last Si, much larger number
of cycles is necessary. Continuous films have been obtained on HF
last Si, but the growth mechanism is still under study. Similar is-
landlike growth behavior to what we observed on H-terminated Si
has been shown for HfO2 dielectric growth on H-terminated Si.17
A cross-sectional schematic of the fabricated test structure is
shown in Fig. 6a. Capacitor–voltage data ͑Fig. 7b͒ show a large
hysteresis effect in metal-oxide-semiconductor ͑MOS͒ capacitors
embedded with Pt nanoparticles. Pt charging is measured by calcu-
lating the flatband ͑FB͒ voltage on the forward and reverse sweeps
of the hysteresis measurement. At low voltages, there is no signifi-
cant charging in the Pt layer; however, after about 8 V of applied
bias, a large hysteresis is observed. The high voltages coincide with
the onset of Fowler–Nordheim tunneling as seen from current–
voltage measurements made on the same test structure ͑not shown͒,
which is desirable for low power memory devices. The charge den-
sity was calculated from the change in FB voltage as shown in Eq.
1
Figure 4. TEM image of 10 cycles of ALD Pt with 10 s precursor exposure.
The increase in particle size caused by the precursor exposure is counter-
acted with the lower cycle count. The result is small, individual particles that
are dense and uniformly distributed across the sample.
confirmed the presence of Pt on SiO2 ͑not shown͒. These data sug-
gest that the Pt being deposited is metallic, a key requirement for
memory storage.
At higher cycles of Pt ALD, the films become continuous as seen
in ALD Pt film literature.9,16 AFM images of 200 cycles of Pt ALD
Figure 5. ͑Color online͒ XPS data of Pt ALD on Al2O3. A detectable Pt
signal16 is present after 20 cycles of ALD matching up with the appearance
of particles in AFM. The apparent rise in the Al 2p peak comes from the
doublet peak of Pt. Similar XPS behavior is seen for ALD on SiO2.
Figure 7. ͑Color online͒ ͑a͒ CV hysteresis
data show charging of Pt nanoparticles.
The positive FB shift with positive voltage
indicates electron trapping in the nanopar-
ticle layer. Little to no charging is seen for
applied voltages less than 8 V, but a large
charging effect is seen from FB shifts at
higher voltages. ͑b͒ Retention data were
measured after programming the capacitor
at 15 V for 5 s, and the particles show
good retention after 104 s with less than
8% charge loss.
Downloaded on 2015-01-06 to IP 130.194.20.173 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).

H592
Journal of The Electrochemical Society, 157 ͑6͒ H589-H592 ͑2010͒
acteristics after this change is currently under study and will be
reported in a later publication.
Conclusions
In summary, we have demonstrated a process to grow Pt nano-
particles by a low temperature ALD. A process window that utilizes
ALD cycle numbers and precursor exposure time to control the par-
ticle size was demonstrated. The initial particle density depends
heavily on the starting surface. This leaves the possibility of tuning
particle density with surface treatments and the ability to change
particle size with ALD cycles, both relatively independent of each
other. Charging effects are seen for Pt nanoparticles embedded in
MOS capacitor structures with high charge densities and good re-
tention characteristics with low charge loss over time. The Pt par-
ticles are thermally stable and can withstand source–drain activation
anneals of less than 1000°C. However, to allow better device de-
sign, further investigation into the electrical behavior and dielectric
stability at anneal temperatures at or above 1000°C is needed.
Therefore, the high work function, high densities, and good thermal
stability of ALD-based Pt nanoparticles are an attractive route for
future nonvolatile memory devices and other emerging devices.
Figure 8. TEM images of Pt particles embedded in Al2O3 annealed in situ.
͑a͒ As-deposited particles show high density. ͑b͒ Almost no change in the
particle sizes and structure is seen after annealing up to 900°C for 15 s. ͑c͒
After annealing at 1000°C for 5 s, there is a dramatic change in the nano-
particle size and distribution as particles diffuse and coalesce.
CG-NP · ⌬VFB
⌬N =
͓1͔
q
where CG-NP is the capacitance between the gate and Pt nanopar-
ticles. Peak densities of about 2 ϫ 1013 cm−3 are obtained from
hysteresis measurements. The FB shift is positive at positive volt-
ages, with respect to the control sample. This indicates the storage of
electrons with positive bias, as would be expected for a metal stor-
age layer.
Acknowledgments
We thank Protochips for providing transmission electron micro-
scope grids and equipment. This work was supported by the Na-
tional Science Foundation ͑NSF ECCS 0802157͒.
Retention characteristics of the memory capacitors were deter-
mined by measuring the FB voltage as a function of time after
charging the Pt particles with a 5 s, 15 V voltage pulse ͑Fig. 7c͒. The
programming pulse gave an initial FB shift of about 3.5 V. After
104 s, the FB shift decreased by about 300 mV, which is less than
8% charge loss. These retention characteristics indicate that the high
voltages do not significantly degrade the dielectric and, hence, make
the use of Pt particle attractive for memory devices.
North Carolina State University assisted in meeting the publication costs
of this article.
References
1. D. Adams, P. Farfell, M. Jacunski, D. Williams, J. Jakubczak, M. Knoll, and J.
Murry, in Nonvolatile Memory Technology Review, IEEE, pp. 96–99 ͑1993͒.
2. H. I. Hanafi, S. Tiwari, and I. Khan, IEEE Trans. Electron Devices, 43, 1553
͑1996͒.
3. I. Kim, S. Han, K. Han, J. Lee, and H. Shin, IEEE Electron Device Lett., 20, 630
͑1999͒.
Temperature stability of the Pt nanoparticles is important for
their use in charge storage devices. Floating gate metal-oxide-
semiconductor field-effect transistors need to withstand tempera-
tures around 900–1050°C for the source–drain activation anneal.
The stability of nanoparticles under temperature is a critical function
of the substrate and the capping layer. In this study, the annealing of
ALD Pt nanoparticles ͑20 cycles with 10 s exposure͒ completely
embedded within Al2O3 was performed in situ on a custom trans-
mission electron microscope grid provided by Protochips.18 This
grid allows for very rapid changes in temperature within the trans-
mission electron microscope. Negligible change was observed in the
size and distribution after annealing at 900°C for 15 s ͑Fig. 8͒,
indicating that the Pt particles can withstand temperatures needed to
activate the source–drain region. However, after further annealing at
1000°C for 5 s, a dramatic change was seen in the particles. The
larger size and lower density of particles after 1000°C annealing
indicate that Pt nanoparticles underwent ͑i͒ diffusion through the
dielectric and ͑ii͒ coalescence to create larger particles, which are
clearly observable in TEM ͑Fig. 8c͒. The effect on electrical char-
4. Z. Liu, C. Lee, V. Narayanan, G. Pei, and E. C. Kan, IEEE Trans. Electron Devices,
49, 9 ͑2002͒.
5. D. Panda, A. Dhar, and S. K. Ray, Semicond. Sci. Technol., 24, 115020 ͑2009͒.
6. P. K. Singh, G. Bisht, R. Hofmann, K. Singh, S. Mahapatra, in IEEE International
Memory Workshop, IEEE, pp. 1–4 ͑2009͒.
7. M. Zhang, W. Chen, S. Ding, X. Wang, D. W. Zhang, and L. Wang, J. Vac. Sci.
Technol. A, 25, 775 ͑2007͒.
8. D. B. Farmer and R. G. Gordon, J. Appl. Phys., 101, 124503 ͑2007͒.
9. T. Aaltonen, M. Ritala, T. Sajavaara, J. Keinonen, and M. Leskela, Chem. Mater.,
15, 1924 ͑2003͒.
10. J. R. Hauser and K. Ahmed, in Proceedings of the AIP Conference, AIP, pp. 235–
239 ͑1998͒.
11. K. J. Park, Ph.D. Thesis, North Carolina State University, Raleigh, NC ͑2005͒.
12. T. Aaltonen, A. Rahtu, M. Ritala, and M. Leskela, Electrochem. Solid-State Lett.,
6, C130 ͑2003͒.
13. L. T. Zhuravlev, Langmuir, 3, 316 ͑1987͒.
14. J. H. Cantrell, J. Appl. Phys., 96, 3775 ͑2004͒.
15. NIST X-ray Photoelectron Spectroscopy Database, Version 3.5, National Institute
of Standards and Technology, Gaithersburg ͑2003͒ ͑http://srdata.nist.gov/xps/͒.
16. Y. Zhu, K. A. Dunn, and A. E. Kaloyeros, J. Mater. Res., 22, ͑2007͒.
17. R. L. Puurunen and W. Vandervorst, J. Appl. Phys., 96, 7686 ͑2004͒.
18. L. F. Allard, W. C. Bigelow, M. Jose-Yacaman, D. P. Nackashi, J. Damiano, and S.
E. Mick, Microsc. Res. Tech., 72, 208 ͑2009͒.
Downloaded on 2015-01-06 to IP 130.194.20.173 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).