1. Introduction

SiC has a wide band gap, high thermal conductivity, high breakdown field strength and chemical inertness making it of interest in applications involving high-temperature and high power electronic devices [1–4]. 3C-SiC, 4H-SiC and 6H-SiC are well studied with band gaps of 2.36, 3.26 and 3.02 eV respectively, where 4H-SiC is the most industrially common [5–7].

Prior to the widespread deployment for high-power electronics, SiC was used as a semiconductor for light-emitting diode (LED) technology with yellow [8] and blue [1] emission characteristics. Its indirect band gap resulted in very low quantum efficiencies ($\eta \cong 10^{-5}$) which prevented its continued use in LED applications after the development of nitride semiconductors. Despite this, there are defect-related radiative emission pathways in SiC that can be further studied and possibly further developed for future LED use. This is motivated by the earth abundance and low cost of SiC.

Stability and reliability in power electronic applications are negatively impacted by defects such as dislocations and Shockley stacking faults, as well as their growth under electrical operation. These defect centers can behave as radiative recombination sites, where degradation of SiC devices can be analyzed through a series of spectral characterizations. Study of luminescence, due to these recombination centers, therefore can aid in investigations on degradation such as to clarify failure mechanisms including switching capabilities and related I-V characteristics [9,10].

To date, studies to understand SiC degradation include induced defect formation through current-based stress testing [11–15] and growth mechanisms [16–19]. Various experiments using processes such as particle irradiation and laser excitation [19–21], combined with theoretical modelling [22–26] have resulted in an understanding of the underlying mechanisms. This understanding includes defect-assisted recombination, quantum well models and stacking fault formation and movement.

Clear visualizations of stacking fault expansion during continuous electrical operation in SiC have been recorded in the literature [27]. It is understood that long term operation of devices causes recombination-induced stacking fault expansion [22,28]. The deterioration of I-V characteristics is a side-effect of these defects.

Due to the nature of the 4H-SiC’s indirect band gap, it is commonly understood that bulk band-to-band emission at 390 nm is possible with the assistance of particles or quasiparticles such as phonons [23].

Experimental studies using photoluminescence intensity mappings have also characterized other emission ranges in SiC. Various spectral peaks at or about 420 nm, 455 nm, 480 nm and 500 nm are recorded in the literature with evidence pointing to 1SSF, 4SSF, 3SSF and 2SSF as contributing factors, respectively [16,19,29]. Here, for example, 1SSF refers to the single Shockley stacking fault, and the higher-numbered Shockley stacking faults are written in an analogous manner. These stacking faults act as quantum wells, where electron confinement in a lower conduction band level aids carriers in recombination. For example, a localized energy level of the 1SSF lowers the conduction band energy of a 4H-SiC crystal by roughly 0.3 eV resulting in the 420 nm emission [22,29].

Surface structure variations such as neutral oxygen vacancies and carbon related surface defects have also been reported to contribute to the blue-green (460–530 nm) emissions [17,30].

Porous SiC and SiC subjected to microindentation has also shown improved blue-green luminescence [30–33].

4H-SiC has a larger band gap, higher temperature range and higher carrier mobility than their polytype counterparts which make it favourable to examine carrier and recombination related properties. In this work, a series of spectra and IV measurements taken from commercial 4H-SiC bipolar junction transistors are presented and discussed. A set of spectra and IV curves were taken during a series of grinding and polishing conditions. Samples were ground at varying surface roughness to inspect their effects on the recombination pathways. Emission at 420 nm under forward biased operation, which is commonly attributed to 1SSFs [16], is observed to depend on both surface polish and stress test operation.

2. Experimental details

2.1 SiC sample preparation

A 4H-SiC bipolar junction transistor (GA10JT12-263, GeneSiC) was mounted in epoxy resin. The die schematic diagram can be seen in Fig. 1. The package was ground using a Struers Tegramine 25 with 200 grit SiC sandpaper to expose an optically translucent surface of the die at a side facet of the die. Contacts to the base-collector junction were then soldered to wires to provide electrical contact throughout the experiments.

 

Fig. 1. Structure of a bipolar junction transistor through a) a cross-sectional view and b) perspective projection which indicate the ground facet of the sample

Download Full Size | PDF

The surface preparation of the sample can be categorized into three stages. In Stage 1, the sample was ground with 200 grit SiC sandpaper and then cleaned with distilled water as seen in Fig. 2.

 

Fig. 2. 4H-SiC GeneSiC GA10JT12-263 SJT die after a) exposing the surface and b) forward biasing

Download Full Size | PDF

Further grinding was performed in Stage 2 with 500 grit SiC sandpaper. The surface finish was improved and a noticeable portion of the sample, in which near-surface damage is expected, was removed. This near surface damage is associated with the rough surface finish from Stage 1. Finally in Stage 3, 1200 grit grinding and polish was performed on the sample where the number of pits and artifacts have again substantially decreased. Nikon microscope images of these surfaces were taken before optical and electrical characterization of each stage and can be seen in Fig. 3.

 

Fig. 3. Sample microstructure 1) after Stage 1, 200 grit grinding 2) after Stage 2, 500 grit and 3) after Stage 3 1200 grit

Download Full Size | PDF

Measurements of the electroluminescence spectrum as well as I-V behaviour was characterized under base-collector forward bias for each of Stage 1, Stage 2 and Stage 3. Stage 2a measurements refer to measurements made on the sample immediately following the grind step of Stage 2. Following the Stage 2a measurement, the sample was operated under 0.2A forward bias for two hours. Stage 2b measurements refer to measurements made on the sample immediately after this forward bias treatment. The same labelling method was used in Stage 3. A colour emission comparison of an unprocessed sample to the experimental sample after Stage 3b can be found in Fig. 4 where the sample was connected in series with the unprocessed sample under 0.2 A forward bias operation.

 

Fig. 4. Comparison of a) an unprocessed sample to b) the experimental sample at 0.2 A after Stage 3b

Download Full Size | PDF

2.2 Characterization and measurements

The optical emission of the sample from the cut edge was analyzed using a 1800 Line VIS Holographic Monochromator. The detector was an S12053 Si Avalanche Photodiode (Hamamatsu Photonics), under 159V reverse bias at room temperature. Measurements were taken at 0.2 A through the forward biased base-collector junction.

Measurements were taken and the intensities were normalized with respect to the 390 nm peak. The electroluminescence in Stage 1, over an operating period of 2 hours, shows the initial conditions of the transistor luminescence with peaks at 390 nm, 445 nm and 480-510 nm as seen in Fig. 5. After the Stage 2a grinding using 500 grit, a noticeable color change was observed during forward bias characterization of the sample. The spectrum was taken twice as seen in Fig. 6 where measurements can be identified as a) immediately after grinding and b) after 2 hours of operation. During Stage 2a, the luminescence at 445 nm slightly increased while the broad blue-green emission peak, centered around 500 nm, slightly decreased. An emergence of a 420 nm peak can be seen in Stage 2b, suggesting either nucleation or expansion of 1SSF [22].

 

Fig. 5. Stage 1 electroluminescence spectra of the sample at initial conditions

Download Full Size | PDF

 

Fig. 6. Stage 2 electroluminescence spectra after a) grinding and b) 2 hours of operation

Download Full Size | PDF

In Stage 3, Fig. 7, shows that the peaks have shifted more drastically in comparison to Stage 2. More notably, the broad blue-green emission peak has decreased and broadened while the 420 nm peak has become more distinguished. Further characterization of the transistor, after another 2 hour operating period in Stage 3b, shows that the 420 nm peak, with a 445 nm tail, has begun to overtake the primary luminescence spectrum. The collective electroluminescence and I-V characteristics can be seen in Fig. 8, and Fig. 9 respectively. Fig. 8 shows the evolution of the spectrum over the entire set of experiments. From the I-V characteristics, the ideality factor, n, increased from 1.25 to 1.46 to 2.01, and finally 2.80 from Stages 1, 2, 3a and b respectively which explains increase in forward voltage at given current during the measurements. Ideality factor determination was based on exponential regression.

 

Fig. 7. Stage 3 electroluminescence spectra after a) grinding and b) 2 hours of operation

Download Full Size | PDF

 

Fig. 8. Collective spectrum of the sample

Download Full Size | PDF

 

Fig. 9. IV Characteristics

Download Full Size | PDF

3. Discussion

In Fig. 5 the initially observed peaks can be seen at 390, 445, along with a broad peak around 500 nm. Bulk 4H-SiC has a band gap emission wavelength of 380 nm (3.26 eV). The slight shift in emission wavelength of 4H-SiC towards 390 nm is due to phonon assisted band-to-band recombination [23]. The 445 nm peak was also observed from previous works [33]. Blue emission at 445 nm was observed from carbon doped porous silicon with formation of $\beta$-SiC. This emission wavelength of our sample may be due to a similar quantum confinement effect or surface states [34]. Meanwhile, the broadband blue-green emission from the spectrum is potentially attributed to a mixture of radiative recombinations due to the presence of carbon related surface defects and partial dislocations [17,30].

In Stage 2, the 420 nm centered peak was first observed, in Fig. 6(a), and is known to be caused by 1SSFs [22]. The lower conduction band energy of 1SSFs with respect to the bulk 4H-SiC band gap ($\sim$0.3eV) would provide a more favourable recombination pathway. While the 420nm peak starts to dominate the overall recombination during the experiment, a slight change of the emission colour can also be observed. Further comparison of Stage 2 to Stage 1 shows a $\sim$2 nm red-shift can also be seen at the 390 nm peak at Stage 2a while no shifting in Stage 2b is seen.

The grinding process may have induced some shear stress to the sample and literature has attributed the origin of stacking fault formation and expansion to result from a combination of dislocation motions [14,35].

Numerous references have noted the contributions of forward bias effects on stacking fault expansion [10–14,36,37]. It is noteworthy that, despite after several hours of forward bias operation in Stage 1, the 420 nm peak did not appear. This would suggest that the grinding of Stage 1 may have induced other lattice damage thereby preventing electrically induced formation of 1SSFs which are known to arise from the glide of partial dislocations.

Stage 3 is consistent with changes observed previously, as seen in Fig. 7. The 500 nm broad peak has further reduced while the 420 nm peak has grown. After 2 hours of operation as seen in stage 3b, the 420 nm peak has increased much more dramatically. This could possibly be attributed the smoother surface with reduced near-surface damage that enhances dislocation glide near the surface, thereby expanding 1SSFs. In tandem, the 500 nm broad peak decreased, showing a slight tail from the 445 nm emission. A small red-shift, at the 390 nm peak, is also observed when comparing stages which amounts to $\sim$0.5 nm between Stage 3a and 3b. While the shifting seen in Stage 2 occurred only after surface changes, the additional red-shift in Stage 3 indicates that the cause may not be due to surface roughness. The change of emission color can also be directly observed from the human eye as seen in Fig. 4; the colour is distinct from the faulted sample with a purple hue after Stage 3b compared to the unprocessed sample’s original blue hue.

4. Conclusion

As a promising third generation semiconductor, SiC is earth-abundant and non-toxic may also be suitable for applications ranging from bio-compatible markets to future micro-LED displays. It is important to analyse the effects of surface states and stacking faults on SiC device performance.

Comparing Stages 1-3, based on Fig. 8, the relative intensity of the broad peak around 500nm emission has been gradually reduced with smoother surfaces. The change of surface roughness can directly be observed to improve from Fig. 3. This can be interpreted as the reduction in radiative recombination due to carbon related surface defects as a result of surface area reduction.

The formation and expansion of stacking faults under the forward bias operation is a commonly known degradation in 4H-SiC bipolar devices [9,22,28]. Partial dislocation glide that enables stacking fault expansion is favoured in low defect density material. Aggressive grinding treatments known to create lattice damage would therefore suppress partial dislocation glide compared to finer polish steps. Our data is consistent with this conjecture, the growth of the 420 nm peak with each stage in the experimental work can be understood as accompanying the degradation of the I-V characteristics of the base-collector p-n junction. The presented electroluminescence spectra and I-V characteristics in Figs. 8 and 9 also further indicate that a higher power consumption is also expected due to the increasing ideality factor which matches known literature.