Loboda_Chung Univ Md WBG Workshop Paper July 2014

Emerging Challenges in Evaluation of Silicon Carbide
Materials-Device Interactions
M.J. Loboda*
, G. Chung*
*
Dow Corning Corporation, Midland, MI 48686; mark.loboda@dowcorning.com; gil.chung@dowcorning.com;
ABSTRACT
The last five years have produced significant advances in technology for silicon carbide power devices. Substrates
with diameter of 100 mm are now widely used in production and industry now has started the transition to 150 mm
diameter SiC substrates. Bulk crystal originated defects such as screw and basal dislocation have generally dropped >10x
in concentration to levels in the 103
/cm2
range. Today, micropipes are essentially concern of the past. Now there is more
focus on surface contamination and defects originated in CVD epitaxy, the latter representing the materials-related
defect most commonly impacting device performance and manufacturing yields.
Improvements in substrates and device technology now deliver larger area power devices. Recent news of new
suppliers of SiC MOSFETs is likely to queue increased growth in adoption of SiC power devices in system applications.
As a result there is still focus on how to grow understanding in materials-device interactions that will help improve
manufacturing yields and understand device reliability. The increase in device complexity and area brings new
challenges to develop improved understanding for materials device correlations. This presentation will review the state
of the art in SiC substrates and several emerging issues pertaining to evaluation of materials-device interactions that are
pertinent to current trends in SiC devices. Among the list of critical issues are statistical approaches, the impact of defect
clustering, match of materials characterization strategy to device fabrication and nano-scale surface perturbations. It is
the authors’ objectives to increase awareness of these new issues in order to work with the community to help prioritize
strategies and understand the impact the issues have on device performance and yield.
Keywords: SiC, power electronics, material-device correlations.
Introduction
All power device operation is intimately influenced by the substrate properties. The key substrate properties are
wafer resistivity, wafer thickness, epitaxy layer doping, epitaxy layer thickness, epitaxy layer mobility and epitaxy
defects. Among these metrics, the first four can vary widely in value depending on the location within the substrate.
Defects which perturb the epitaxy surface usually render devices inoperative.
In the case of 4H-SiC substrates, the range of within substrate variations of the key metrics are generally larger than
in silicon substrates. This characteristic of SiC substrates is generally regarded as leading to difficulty to achieve
consistent device operation and fabrication yields. We will discuss current perspectives on surface defects and within
wafer metric variations as it pertains to power device fabrication.
SiC Epitaxy Surface Defects – Power Device Interactions
Surface perturbations in SiC epitaxy films are known killer defects for power devices. A detailed study was
performed in 2013 to check the influence of surface defects on breakdown voltage failures in devices and the ability of
non destructive wafer inspection systems to grade the defect density accurately enough to predict the device breakdown
failures. Many different high quality SiC epitaxy substrates were randomly selected for the test. Laser light scattering
spectrometry (LLS) defect density measurements performed on randomly selected of the wafers in the set. The substrates
were populated with junction Schottky barrier diodes of various areas including 2x2 mm2
and 4x4 mm2
using a
commercially proven process flow. All diodes on the wafers were probed for breakdown failures. Based on the
breakdown voltage yields, or LLS site yields, as a function of die area the characteristic defect density can be estimated
by fitting the LOG(yield) vs. device area data and assuming a Poisson distribution. Figure 1 compares the results.
Invited Paper
Workshop on Defects in Wide Band Gap Semiconductor
September 23rd, 2014 - University of Maryland, College Park

Inspection of Figure 1 shows good agreement between the defect density that was measured electrically and that
from the automated light scattering inspection. Analysis of failures indicates the mean value of defect density is about
1.5/cm2
. Since the defect density values measured by both techniques are comparable it can be concluded that epitaxy
surface defects represent the defect that resulted in the breakdown voltage failures.
As the capability of surface defect inspection improves, it is now possible to resolve smaller features on the surface
of the epitaxy substrate. Recently a defect represented by very small depressions has been detected on SiC epitaxy
substrates. The defect, referred to as a “micropit,” is shown in Figure 2. AFM analysis (Fig. 2c) indicates a typical depth
of <100 nm. KOH molten salt etching has been performed and indicates the pit is tied to a threading screw dislocation
(Fig. 2b). Theoretical simulation and local emission microscope data was reported to show leakage current increase with
micropits [1,2]. Fully processed diodes and MOS capacitors with micropits in the active area, however, show no
difference in reverse breakdown and time zero dielectric breakdown values when compared to devices free of the
defects, respectively. Any impact of these defects on device reliability is not yet established.
Trends in SiC power devices continue to achieve device designs capable of higher voltage and current. As the
device active area is increased, there is likely to be clusters of similar defects in the active region. The clustering will
Figure 3. Correlations between high voltage JBS diode breakdown and leakage current with screw
dislocation density in the anode region. All of the devices in the right figure were measured on the same
1413121110987654321
10
8
6
4
2
0
1413121110987654321
VBR Failure Defect Density
Wafer ID
Auto Inspection Defect Density
Comparison of Characteristic Defect Density (cm-2)
Die not uniformly
distributed on device
wafer
Pre-fabrication LLS
Test on epiwafers.
Assumes die
uniformly distributed
on device wafer
(c)
Registration Marks
TSD
(b)
(a) Figure 2.
Images of
“micropits”
detected on the
surface of 4H-
SiC epitaxy
wafers: (a) laser
interferometer;
(b) KOH etched
epitaxy surface;
(c) atomic force
microscope
analysis.
>106.0-8.0>5.51.0-6.02.0-4.5<2.0<1.0<0.5
5000
4000
3000
2000
1000
0
kV Bin
SDD(cm-2)
Screw Dislocation Density Distribution by Diode Blocking Voltage Bin
2000150010005000
350
300
250
200
150
100
50
0
SDD (/cm2)
leakagecurrentdensity(uA/cm2)
Larg e
Med iu m
Small
die size
Leakage current density(uA/cm2) vs SDD (/cm2) BB9015-14 9kV Bias
VBR@5mA/cm2 J@9 kV
Figure 1. Defect density values for 4H-SiC
wafers fabricated with junction barrier Schottky
diodes. The left side is determined from electrical
failures, the right side from automated laser light
scattering epitaxy surface defect measurements.

make it difficult to interpret the impact of substrate crystalline or epitaxy defects on device operation. The presence of
any epitaxy surface defect in the active device area will inhibit proper device operation. But what is the impact of
clusters of crystalline defects on device operation when surface defects are absent? Figure 3 shows analyses of high
voltage JBS diodes, which are a favorable platform to explore impact of defect clusters. All of the diodes analyzed in
Figure 3 were free of epitaxy surface defects. Once probed for reverse bias leakage, the anodes were removed and the
screw dislocation defect density was measured in the anode region by molten salt etching and x-ray topography. Figure 3
shows that in these devices were the screw dislocation density was between 0-5000/cm2
, it was not the primary factor
which correlated to the device breakdown or leakage currents. The data in Fig. 3 is strong evidence that in the absence of
epitaxy surface defects in the active region of a power device, the identification defects limiting operation of SiC devices
is a complex task and will require large sample sets and significant analysis to resolve.
Impact of Substrate Metrics on Device Operation
When designing SiC devices for commercial production, what is the correct expectation for the distribution of
device performance within wafer in the absence of defects? Power device OEMs use technology computer aided design
(TCAD) modeling software to work up device designs for their products. Often design engineers are limited to put in
values into the model for the SiC electrical properties that are found in reference books [3] and find their models are not
delivering predictions of manufacturing processes to expectations. It is generally regarded that TCAD models for wide
band gap materials are not as accurate for silicon devices. What’s more the models may not account for the within wafer
(WIW) and wafer to wafer (W2W) variations of the electrical properties that impact the device performance. The
challenge becomes the need to rationalize the device model results with measured variation in device performance. Often
the mismatch of device performance shortfalls is instinctively assigned to the substrate material without evidence.
While the quality of SiC substrates is still not yet as good as silicon substrates, the consistency of commercially
available substrate properties is good enough to be represented by standard probability statistics. Using well established
physics models which link the substrate properties to the device performance, a method has been developed to assess the
impact of WIW variations on SiC device performance. A Monte Carlo analysis program (Oracle ® Crystal Ball) is used
to link the SiC electrical WIW property distributions to variations of device performance that can be expected for
devices fabricated on SiC wafers. Comparison of the device performance probability with measurements performed on
wafers populated with simple diodes shows that the simulations can provide a close match to the measured within wafer
device parameter distributions. This simulation strategy can be used to help understand which of the substrate electrical
metrics most impact device performance, and also predict how much change in the substrate is needed to achieve
measureable improvements in device performance. This simulation strategy can help device designers to better align
their designs with the properties of the available SiC wafer product in order to get predictable product performance.
In order to execute the model, it is necessary to assign to each of the substrate metrics a mathematical distribution
function. Using large datasets obtained from measurements on 100 mm and 150 mm diameter 4H-SiC polished wafers
and epitaxy wafers, a “best fit” distribution function was assigned to the wafer thickness, wafer resistivity, epitaxy film
thickness and epitaxy film thickness. A Normal distribution was a suitable fit for all metrics except the wafer resistivity.
This is because of the presences of the growth facet [4] on the wafer. In the region of the facet the nitrogen incorporation
in the wafer is higher than the rest of the wafer and as a result the facet region shows lower resistivity values than the
rest of the wafer. This creates an asymmetry in the resistivity distribution and a “min-extreme” distribution was
employed.
The calculation of pure Schottky diode performance was derived from physics of power device operation [5,6,7]. A
Microsoft Excel spreadsheet was used to calculate the room temperature device breakdown performance and series
resistance, the two critical parameters for device designs. The Crystal Ball software was configured to use Monte Carlo
analysis to determine the probability distribution of these two metrics based on the variability of the substrate metrics
wafer resistivity, wafer thickness, epitaxy layer doping, epitaxy layer thickness. Four thousand iteration cycles were used
in the Monte Carlo analysis.
The simulation was first employed to predict the probability distribution of breakdown voltage and series resistance of a
100 mm diameter, 4H-SiC epiwafer produced at Dow Corning. The substrate was populated with pure Schottky variable
diodes by sputtering Ni metal on the surface using a shadow mask. The anode area was 0.125 cm2
. The backside of the
wafer was metalized with Ti to achieve good electrical connection during wafer probe measurements of current-voltage
which are used to determine breakdown voltage and series resistance. The criteria used for device breakdown was 1mA.

The results of the model and measurements are shown in Figs. 4-5. Table 1 compares the distribution of results from the
model with the measurements on the diodes made on the wafer.
900850800750700650
95
90
80
70
60
50
40
30
VBR (volts)
Percent
50
90
95
768 849 889
Normal
Measured VBR Distribution
Non-Linear Y-Axis
Wafer AS1352-01
Figure 4. Comparison of the model (left) and measured (right) breakdown voltage probability distribution for
Schottky diode made on a 100 mm diameter 4H-SiC wafer.
0.0160.0140.0120.0100.0080.0060.0040.002
99.9
99
95
90
80
70
60
50
40
30
20
10
5
1
0.1
R_s (ohms)
Percent
50
90
95
0.007 0.012 0.014
Measured Series Resistance
Non-Linear Y-Axis
Wafer AS1352-01
Figure 5. Comparison of the model (left) and measured (right) total series resistance probability distribution for
Schottky diode made on a 100 mm diameter 4H-SiC wafer.
Table 1 – Comparison of the model and measured breakdown and total series resistance distributions.
VBR
Probability
Percentile
Monte Carlo
Model (V)
MeasuredDataon
Wafer (V)
50 772 768
90 847 849
95 869 889
Rs
Probability
Percentile
Monte Carlo
Model
(ohms)
MeasuredData
on Wafer (ohms)
50 .0067 .007
90 .0070 .012
95 .0071 .014
The model effectively predicts the distribution of breakdown voltage measured on diodes. It is also predicting the
lower half of the series resistance distribution. The higher than predicted resistance values are most likely due to a non-
optimal metallization process in the laboratory and lack of backside ohmic contact formation in the diode fabrication.
The Monte Carlo analysis does not account for resistance due to metal contacts to the diode. It is interesting to note the
wide range of VBR that is achieved on the wafer and also predicted by the model.

Next the analysis was repeated for the case of 150 mm 4H SiC epiwafers. Figure 6 shows the inputs to the analysis
derived from commercial 150 mm diameter SiC substrate product and the probability distributions of breakdown voltage
and total series resistance.
Wafer Thk= 350 um
TTV= 5 um
ResistivityAve= 21.4 m-ohm
Resistivity Scale= 0.71
Goal= 1200 V
Design= 1600 V
Epi tf= 13 um
Epi WIW sigma= 3%
Nd= 9.7E15/cm3
Epi Nd sigma= 4.3%
Epi Mobility= 825 cm2/V-sec
Die Size= 1 cm2
Trials= 4000
The parameters
selections for the model inputs in Figure 6 were developed as follows. The value for epitaxy film doping target was
chosen based on experience of what is commonly used for 1200 V designs. Next, epitaxy thickness was determined from
the design voltage and the critical field. The critical electric field for SiC was determined as described in [6] where the
critical field is derived from an empirically developed dependence on film doping. Finally, the design voltage in Table 1
were chosen to be sure the smallest value of breakdown voltage at any possible combination of parameters was >=1200
V. This selection process of the design target thus impacts the epitaxy film thickness, and inherently impacts the total
series resistance. The results show that the minimum breakdown voltage expected from the model is about 75% of the
device breakdown design target, while the maximum voltage expected is about 120% of the target. The software also
highlights the correlations between substrate and device parameters. Epitaxy thickness variations of 3% standard
deviation account for 66% of the VBR variation and epitaxy doping variations of 4.3% standard deviation account for
33% of the VBR variation and 51% of the Rs variation.
This result is interesting when put in perspective of the current capability in state of the art SiC CVD epitaxy today.
The input epitaxy and wafer variations are typical of 150 mm diameter substrates available today. The question to ask is
what gains are there to be obtained if the within wafer epitaxy film thickness and doping are reduced. If the standard
deviations are reduced from the values in Fig. 6 to 1% for both thickness and doping respectively, the mean predicted
value of breakdown voltage stays the same at 1506V, but the range is reduced to 1400-1600V with little effect on Rs.
This implies the design target can be reduced from 1600 V to shift the mean breakdown voltage lower, and reduce the
epitaxy thickness. A design target of 1475V results in a breakdown range of 1200-1360V, and corresponds to reducing
the thickness from 13 um in the base case (1600V design) to 12 um (1400 V design). This thickness change has very
little impact on series resistance. So even if the with wafer standard deviations can be reduced from today’s values for
150 mm wafers of 3 % for film thickness and 4.3 % for doping to less than or equal to 1%, the savings in substrate
manufacturing cost is very small.
Summary
Review of the state of the art for surface defects created during SiC epitaxy on commercial substrate products show
these defects are the most important defect limiting SiC power device manufacturing yields. Current typical industry
epitaxy defect levels 100 mm substrates are about 1.5/cm2
. New automated surface analysis techniques are finding new,
smaller defects on epitaxy surfaces, but the impact of these defects still needs to be determined. Analysis of large, high
voltage diodes shows that when the screw dislocation density in the anode region is below 5000/cm2
these defect density
Figure 6. Monte
Carlo analysis of
breakdown voltage
VBR and total
series resistance,
Rs for a Schottky
diode based on

values do not correlate with the breakdown voltage or the device leakage at the breakdown condition. Analysis of the
interaction of SiC substrate metrics and prediction of the breakdown and series resistance of devices fabricated on the
substrates shows that a wide range of within wafer breakdown voltage is expected even when the within wafer epitaxy
variations are relatively small and controlled.
Putting these observations into the context of studies focused to identify the effects of defects like screw and
threading edge dislocations on device performance is a difficult task. In order to isolate of the impact of individual defect
types without cross term effects, experiments require analysis of very large numbers of devices, including comparisons
between matched devices that are free of, or contain the defect of interest. Considering the recent advances in SiC device
technology and that the industry is poised for growth, it is recommended that the SiC community focus its combined
capability to improve SiC epitaxy equipment and processes in order reduce surface defects and increase both substrate
product yields and device yields. Such gains will bolster volume production of SiC devices including larger current and
voltage products, and these gains can become the statistical foundation to effectively study the next discovered factor to
improve SiC technology.
References
1. T.Ishikawa et. al, Materials Science Forum, Vols. 717-720 (2012) p.371
2. T. Katsuno et. al, Materials Science Forum, Vols. 717-720 (2012) p.375
3. M. Levinshtein, S. Rumyantsev, M. Shur, “Properties of Advanced Semiconductor Materials GaN, AlN, InN, BN,
SiC, SiGe,” Wiley, ©2001
4. I.D. Matukov, et. al, Journal of Crystal Growth 266 (2004), p.313
5. Schoen, et. al, IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 45, NO. 7, (1998) p.1595
6. R. Singh, J. Cooper, IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 49, NO. 4, (2002) p.665,
7. M. Bhatnagar and B. J. Baliga, IEEE TRANSACTIONS ON ELECTRON DEVICES Vol.40 NO. 3 (1993), p.645

Loboda_Chung Univ Md WBG Workshop Paper July 2014

More Related Content

What's hot (8)

Viewers also liked (14)

Similar to Loboda_Chung Univ Md WBG Workshop Paper July 2014 (20)

Loboda_Chung Univ Md WBG Workshop Paper July 2014