Statistics of electrical breakdown field in HfO2 and SiO2 films from

millimeter to nanometer length scales

Cédric Sire, Serge Blonkowski, Michael J. Gordon, and Thierry Baron

Citation: Appl. Phys. Lett. 91, 242905 (2007); doi: 10.1063/1.2822420

View online: http://dx.doi.org/10.1063/1.2822420
View Table of Contents: http://apl.aip.org/resource/1/APPLAB/v91/i24
Published by the AIP Publishing LLC.

Additional information on Appl. Phys. Lett.

Journal Homepage: http://apl.aip.org/
Journal Information: http://apl.aip.org/about/about_the_journal
Top downloads: http://apl.aip.org/features/most_downloaded
Information for Authors: http://apl.aip.org/authors

Downloaded 20 Aug 2013 to 131.170.6.51. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://apl.aip.org/about/rights_and_permissions
 APPLIED PHYSICS LETTERS 91, 242905 共2007兲

Statistics of electrical breakdown field in HfO2 and SiO2 films from

millimeter to nanometer length scales
Cédric Sirea兲 and Serge Blonkowski
ST Microelectronics, 850 rue Jean Monnet, 38926 Crolles Cedex, France
Michael J. Gordonb兲 and Thierry Baron
Laboratoire des Technologies de la Microélectronique (CNRS-LTM), 17 Avenue des Martyrs, 38054
Grenoble, France
共Received 12 September 2007; accepted 15 November 2007; published online 14 December 2007兲
The statistics of electrical breakdown field 共Ebd兲 of HfO2 and SiO2 thin films has been evaluated
over multiple length scales using macroscopic testing of standardized metal-oxide-semiconductor
共TiN / SiO2 / Si兲 and metal-insulator-metal 共TiN / HfO2 / TiN兲 capacitors 共10−2 mm2 – 10 ␮m2 area兲 on
a full 200 mm wafer along with conductive-atomic-force microscopy. It is shown that Ebd follows
the same Weibull distribution when the data are scaled using the testing area. This overall scaling
suggests that the defect density is ⬃1015 cm−2 and Ebd is ⬃40 MV/ cm for nanometer-length scales;
as such, breakdown in these materials is most likely initiated by bond breaking rather than punctual
defects. © 2007 American Institute of Physics. 关DOI: 10.1063/1.2822420兴

Dielectric materials with higher permittivity than SiO2 defects, it is common to measure the statistical distribution
are being investigated as gate insulators in metal-oxide- of Ebd for very small 共nanoscale兲 surface areas. For this,
semiconductor 共MOS兲 devices1 and for metal-insulator-metal conductive-atomic-force microscopy 共C-AFM兲 electrical
共MIM兲 capacitors used in dynamic random access memory2 measurements have been successfully used as a local electri-
and rf applications.3 In the case of MOS devices, a thicker cal probe for thin SiO2.6–8 In particular, the local breakdown
dielectric layer 共2 – 3 nm兲 can be used to prevent electron properties of Si/ SiO2 at nanometer scales have been the
tunneling, but the overall capacitance must be kept the same subject of several extensive studies.9–12 Recently, time-
as 1 – 1.5 nm thick SiO2. Due to these demanding require- dependent dielectric breakdown 共TDDB兲 measurements on
ments, new high-K oxide materials such as HfO2, ZrO2, SiO2 / 4H-SiC with C-AFM showed good agreement with
and HfSiO are being explored as alternatives.4 Decreasing measurements performed on “standardized” SiO2 / Si capa-
device dimensions also require more detailed understanding citors.13 Comparable agreements of the dielectric breakdown
of electrical breakdown at the nanoscale. Unfortunately, kinetics have been observed with Pr2O3 on Si, and it was
breakdown fields in high-K materials have generally been demonstrated that C-AFM was a reliable approach for char-
shown to be lower than in SiO2.5 In this letter, we compare acterizing TDDB events.14
the statistical distributions of breakdown electric field 共Ebd兲 In the present study, macroscopic breakdown tests were
for different dielectrics 共SiO2 and HfO2兲 using macro- carried out on two dielectric films: 3 nm SiO2 共thermally
scopic IV 共current versus voltage兲 tests on standardized ca- grown兲 on Si 共2 ⫻ 1019, n type兲 and 4 nm HfO2 共atomic layer
pacitors 共艌100 ␮m2兲 with nanoscale measurements using deposition兲 on TiN using TiN / SiO2 / Si and TiN / HfO2 / TiN
conductive-atomic-force microscopy 共C-AFM兲 under UHV. standardized capacitors, respectively.15 Electrical measure-
IV curves and Ebd were measured with both methods for ments were made with a four-point probe station and
different oxide thicknesses 共1.2– 6 nm兲 on p- and n-type Si Hewlett-Packard 4192 picoampere meter on 68 different ca-
共MOS structures兲 as well as TiN/oxide/TiN stacks 共MIM ca- pacitors 共areas from 100 to 80 000 ␮m2兲, distributed ran-
pacitors兲. High defect densities 共1015 cm−2兲 and large fields domly over a full 200 mm Si wafer.
共40 MV/ cm兲 suggest that breakdown in high-K materials at C-AFM electrical measurements were performed with an
the nanoscale occurs via bond breaking rather than punctual Omicron AFM/scanning tunneling microscopy system under
defects such as vacancies and interstitials. UHV conditions 共⬍10−9 torr兲 with conductive diamond tips
Electrical breakdown measurements on high-K materials 共B doped兲.16 The AFM tip served as the top electrode for the
typically show much lower fields 共Ebd兲 than for SiO2 of com- “test” capacitor and voltage was applied to the substrate with
parable thickness.5 This observation is usually explained by the tip at virtual ground 共i.e., the tip was connected to an
the presence of a larger local electric field on chemical bonds in-vacuum current preamplifier referenced to ground兲. Elec-
when the dielectric constant is high. However, the statistical trical contact between sample and stage was assured with
mean value of Ebd is seen to increase substantially when the indium solder, and all samples were outgassed at 250 ° C for
surface area for testing decreases. This latter effect can be 3 h at ⬍10−8 torr. IV measurements were carried out on a
explained by the probability of finding a defect which leads 9 ⫻ 9 grid in contact mode 共normal force= 20 nN兲 with the
to breakdown: larger surface area mean higher overall prob- XY scanning of the AFM tip stopped. Reproducibility of the
ability. In order to investigate the inherent density and role of tip-surface electrical contact was investigated with respect to
contact force; above a particular threshold 共20 nN兲, IV traces
a兲
Electronic mail: [email protected]
and Ebd were seen to be contact-force invariant.
b兲
Present address: Deptartment of Chemical Engineering, University of Figure 1 shows a topographic image 共1.5⫻ 1.5 ␮m2兲 of
California-Santa Barbara, Santa Barbara, CA 93106-5080. a 2 nm thick SiO2 layer after a 9 ⫻ 9 grid of breakdown test

0003-6951/2007/91共24兲/242905/3/$23.00 91, 242905-1 © 2007 American Institute of Physics

Downloaded 20 Aug 2013 to 131.170.6.51. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://apl.aip.org/about/rights_and_permissions
242905-2 Sire et al. Appl. Phys. Lett. 91, 242905 共2007兲

FIG. 1. AFM topography image taken after a 9 ⫻ 9 IV grid on a 2 nm thick

SiO2 film thermally grown on a Si substrate.

points was measured. Bright spots in the image correspond

to the location of the IV test; the topographic “bump”
共⬃8 Å兲 is due to trapped negative charges that give rise to an
FIG. 3. 共Color online兲 Cumulative probability of breakdown for 4 nm thick
additional electrostatic force on the AFM tip causing the HfO2 on TiN, comparing macro—共standard capacitors兲 and nanotesting
tip to pull away slightly to maintain constant force feed- 共C-AFM兲. Lines are fits to a Weibull distribution 关Eq. 共1兲兴.
back.17–19 Typical IV traces for the grid experiment are
shown in Fig. 2 for 4 nm HfO2 on TiN, where positive bias
represents electron injection from the AFM tip 共i.e., the
“gate” electrode兲. Five IV traces are superposed to represent
P = 1 − exp − 冋冉 冊 册 Ebd
E0
␤
S
S0
= 1 − exp −
Vbd
V0
冋冉 冊 册 ␤
S
S0
. 共1兲

measurement dispersion. In these evaluations, breakdown Here, S is the capacitor surface area, S0 is a reference surface
voltage was defined as the point when the IV slope became area, and ␤ and E0 共or V0兲 are the Weibull parameters. Tak-
infinite 共arrow on Fig. 2兲. Similar macroscopic IV tests 共not ing the reference surface S0 = 104 ␮m2, one can determine
show兲 were conducted on the same films using standardized V0 and ␤ using other surface areas 共100 and 2500 ␮m2, re-
capacitors in the MOS or MIM configurations discussed spectively兲. Once the Weibull parameters are known from
earlier. macroscopic testing, the corresponding tip-surface contact
Figure 3 shows a comparison of the cumulative break- area for the C-AFM measurement can be determined
down probability versus voltage for the macro- and AFM- 共10± 5 nm2兲. This value is quite reasonable, as estimated
based measurements of 4 nm HfO2 on TiN. Capacitor areas from scanning electron microscopy 共SEM兲 images of the tip
end and AFM topographic images of Si nanocrystals. In the
for macrotesting ranged from 100 to 104 ␮m2. As expected,
latter case, 10 nm structures with spacings of 20– 100 nm
the mean breakdown voltage is much larger 共⬃5 times兲 for
could easily be resolved with the diamond-coated tip used
the AFM test compared to the 104 ␮m2 capacitor. The cumu- for IV testing. In addition, Olbrich et al. measured tip radii
lative probability 共P which represent the likelihood that huge on the order of 10– 20 nm at the apex by SEM.16
currents will flow at a particular applied voltage兲 variation The aforementioned analysis was applied to different di-
with breakdown voltage 共Vbd兲 or field 共Ebd兲 can be fit with a electrics 共SiO2 and HfO2兲 to evaluate whether a single
Weibull distribution, Weibull scaling law can explain breakdown results from the
macro- to nanoscales. Figure 4 shows another Weibull plot
for SiO2 and HfO2 where the data for all surface areas are
compared. In this case, individual Weibull plots 共i.e., Fig. 3兲
are condensed to a “cumulative” probability of breakdown
for each film by linearizing Eq. 共1兲 and taking probability
ratios 共P1 and P2兲 for different surface areas 共S1 and S2兲,

ln关− ln共1 − P1兲兴 − ln关− ln共1 − P2兲兴 = ln 冉冊S1

S2
. 共2兲

When this relation is applied to all the data for a particu-

lar film, a straight line is seen 共Fig. 4兲. The fitted surface area
for SiO2 / Si is seen to be ⬃10 times lower than that for
HfO2 / TiN. This difference is likely due to surface roughness
of the substrate being transferred to the dielectric film during
growth. HfO2 surfaces had a rms roughness of ⬃1 nm, while
the SiO2 was exceptionally smooth 共rms roughness
FIG. 2. 共Color online兲 Five superposed nanometric IV characteristics mea- ⬍0.1 nm兲. Since the tip apex could penetrate into larger
sured by C-AFM on a 4 nm HfO2 / TiN MIM structure. “valleys” on the HfO2 surface 共i.e., the lateral wavelength of
Downloaded 20 Aug 2013 to 131.170.6.51. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://apl.aip.org/about/rights_and_permissions
242905-3 Sire et al. Appl. Phys. Lett. 91, 242905 共2007兲

thickness is 3.9/ K times the physical film thickness, K being

the dielectric constant. In this case, ␤ and the local contact-
ing surface area remain unchanged, but Ebd-EOT = Ebd
⫻ 共K / 3.9兲, with K = 17 for HfO2. In this case, the equivalent
oxide breakdown fields are Ebd-EOT ⬃ 15 MV/ cm at the mac-
roscopic scale and Ebd-EOT ⬃ 60 MV/ cm at the nanometer
scale.
In terms of a charge percolation model,21 our results in-
dicate that nearly any atomic bond is a potential breakdown
path. With such a hypothesis, electrical breakdown in these
films would be initiated by bond breaking rather than by
extrinsic defects such as vacancies or interstitials. Another
recent study5 on SiO2 has shown breakdown fields of
⬃30 MV/ cm which is very near the bond-breaking limit for
Si–O.20 Additionally, the present work demonstrates that
AFM-based statistical testing is a viable, informative, and
potentially easier alternative to macroscopic electrical testing
which may require more processing steps 共such as lithogra-
phy, etching, electrode deposition, etc.兲.
FIG. 4. 共Color online兲 Weibull plot of the breakdown electric field distribu-
tion for 3 nm SiO2 / Si and 4 nm HfO2 / TiN for different surface areas of The authors would like to especially acknowledge J. F.
testing, normalized to the largest area 共800 00 ␮m2兲. Damlencourt and Y. Campidelli for deposition as well as Y.
Morand and S. Descombes for sample integration.
surface roughness was ⬃25 nm for HfO2, close to the tip
1
diameter兲, the contact surface would be larger for HfO2 com- G. D. Wilk, R. M. Wallace, and J. M. Anthony, J. Appl. Phys. 89, 5243
共2001兲.
pared to the smoother SiO2 surface. 2
Y. Aoki, T. Ueda, H. Shirai, T. Sakoh, T. Kitamura, and S. Arai, Tech. Dig.
The linear nature of the Weibull plots in Fig. 4 for both - Int. Electron Devices Meet. 2002, 831.
materials strongly suggests that breakdown occurs via the 3
H. Hu, C. Zhu, Y. F. Lu, B. J. Cho, and W. K. Choi, IEEE Electron Device
same mechanism for testing areas of 10 nm2 all the way up Lett. 23, 514 共2002兲.
to 0.01 mm2 with the Weibull slope ␤ = 44 for Si/ SiO2 and
4
G. Ribes, J. Mitard, M. Denais, S. Bruyere, F. Monsieur, C. Parthasarathy,
␤ = 14.4 for TiN / HfO2. Higher ␤ indicates a lower disper- E. Vincent, and G. Ghibaudo, IEEE Trans. Device Mater. Reliab. 5, 5
共2005兲.
sion of Ebd for SiO2. The lower ␤ value for HfO2 may also 5
J. W. McPherson, J. Kim, A. Shanware, H. Mogul, and J. Rodriguez, IEEE
be explained by the roughness effect. Weibull statistics arise Trans. Electron Devices 50, 1771 共2003兲.
from the assumption that the density of defects 共which gen- 6
A. Olbrich, B. Ebersberger, and C. Boit, Proceedings of the 36th IEEE
erate breakdown兲 is identical for all surface areas tested. Annual International Reliability Physics Reno, 1998 共unpublilshed兲, p.
With respect to the C-AFM breakdown measurements, sur- 163.
7
faces in the 1 – 10 nm2 range were seen to follow the same A. Olbrich, B. Ebersberger, and C. Boit, Appl. Phys. Lett. 73, 3114
共1998兲.
Weibull trend. As such, it would be safe to assume that the 8
M. Porti, M. Nafria, X. Aymerich, A. Olbrich, and B. Ebersberger, Appl.
density of defects must be at least an order of magnitude Phys. Lett. 78, 4181 共2001兲.
greater than the tip-surface area during the test. This being 9
M. Porti, M. Nafría, X. Aymerich, A. Olbrich, and B. Ebersberger, J. Appl.
said, 10–100 defects in a surface area of 1 – 10 nm2 leads to Phys. 91, 2071 共2002兲.
10
⬃1015 defects/ cm2—on the same order as the surface atomic M. Porti, R. Rodrigez, M. Nafria, X. Aymerich, A. Olbrich, and B.
density. We can likewise estimate the volume density of de- Ebersberger, J. Non-Cryst. Solids 280, 138 共2001兲.
11
M. Porti, R. Rodrigez, M. Nafria, X. Aymerich, A. Olbrich, and B.
fects using the tip-surface contact area and film thickness. Ebersberger, Microelectron. Eng. 59, 265 共2001兲.
For this case, 100 defects in a volume of 3 nm3 correspond 12
A. Ando, K. Miki, and K. Sakamoto, Proceedings of the IWGI Tokyo,
to 5 ⫻ 1022 defects cm−3, which is nearly equivalent to the 2001 共unpublished兲, p. 124.
13
atomic density. The mean Ebd measured via C-AFM should P. Fiorenza and V. Raineri, Appl. Phys. Lett. 88, 212112 共2006兲.
14
not be far from the “intrinsic” value. For HfO2 共SiO2兲, Ebd P. Fiorenza, R. Lo Nigro, V. Raineri, and S. Salvatore Lombardo, Appl.
Phys. Lett. 87, 231913 共2005兲.
approaches 13 MV/ cm 共40 MV/ cm兲 at the nanometer scale, 15
F. Mondon and S. Blonkowski, Microelectron. Reliab. 43, 1259 共2003兲.
whereas standard device measurements typically show 16
A. Olbrich, B. Ebersberger, and C. Boit, J. Vac. Sci. Technol. B 17, 1570
⬃4 MV/ cm 共23 MV/ cm兲. Ebd is largest for the smallest sur- 共1999兲.
17
face, with breakdown in HfO2 at significantly lower fields M. Porti, M. Nafria, M. C. Blum, and X. Aymerich, Appl. Phys. Lett. 81,
than SiO2; the latter is in agreement with the value given by 3615 共2002兲.
18
A. Neugroschel, L. Wang, and G. Bersuker, J. Appl. Phys. 96, 3388
McPherson 共⬃30 MV/ cm is needed to break bonds in
共2004兲.
SiO2兲.20 A plot similar to that of Fig. 4 can be drawn by 19
M. J. Gordon and T. Baron, Phys. Rev. B 72, 165420 共2005兲.
replacing Ebd with an equivalent oxide breakdown field using 20
J. W. McPherson, J. Appl. Phys. 95, 8101 共2004兲.
Ebd-EOT 共defined as Vbd/EOT兲, where EOT the effective oxide 21
J. H. Stathis, J. Appl. Phys. 86, 5757 共1999兲.

Downloaded 20 Aug 2013 to 131.170.6.51. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://apl.aip.org/about/rights_and_permissions