Scrutinising SiC with X-ray topography

X-ray topography, already on the cusp of revolutionising the quantification of dislocations in SiC wafers, is now available in a high-throughput form that accelerates progress.

BY CHRISTIAN KRANERT AND CHRISTIAN REIMANN FROM FRAUNHOFER IISB AND SHINTARO KOBAYASHI, YOSHINORO UEJI, KENTA SHIMAMOTO AND KAZUHIKO OMOTE FROM RIGAKU

Silicon carbide is now a mature material that enjoys substantial success in the power electronics sector. Devices made from this semiconductor are currently displacing silicon-based incumbents, especially in the fast-growing market of electric mobility, where SiC is being adopted in both vehicles and charging infrastructure.

Alongside ramping sales of SiC devices, there has been a tremendous improvement in the quality of the material, as well as increases to wafer diameter to 150 mm and 200 mm. However, when it comes to crystalline material quality, SiC is certainly not as perfect as silicon.

One of the weaknesses of SiC is that it contains dislocations. These imperfections are not going to completely vanish from this material in the short term, and their presence matters – they can have a severe impact on the yield, the performance and the reliability of the final devices. Due to this, there’s a need to know the dislocation density of a SiC wafer, a metric that reflects the quality of the material. In fact, such information is more prised than ever, because the high standards within the automotive industry are pushing supply chains towards a complete tracking of all components, from raw material to the final product. Consequently, characterising dislocations in the substrate material provides a valuable piece of information for qualifying suppliers, and for tracking device failure.

Figure 1. Typical TSD density mapping obtained with the Rigaku XRTmicron. Red circles indicate automatically detected locations of screw dislocations.

Until recently, the ‘gold standard’ for quantifying dislocations in SiC substrates involved etching this material in an aggressive alkaline melt heated to roughly 500°C. But this approach is far from ideal. One major weakness is that it is destructive to epi-ready wafers. As every SiC boule may differ from the next, wafer manufacturers analyse at least one wafer from every boule to ensure that they meet the required specifications. Assuming an average yield of 30 to 40 wafers per boule, this etching-based evaluation incurs a yield hit of around 3 percent, a loss that could be eradicated with a non-destructive characterisation technique. Additional drawbacks include those related to the stability of the etching process, the reliability of automated etch pit counting and a lack of standardisation.

Addressing all of these concerns is high-resolution, lab-scale X-ray topography (XRT). This technique builds on synchrotron XRT measurements, which have always been used to identify and detect dislocations – but those measurements are typically performed only locally, and require beam time at synchrotron facilities, so are unsuitable for use in an industrial setting.

These limitations make the Rigaku XRTmicron, which brings high-resolution XRT to the lab, a game changer. This instrument allows engineers to visualise single dislocations, thus making it possible to quantify them. As dislocation images recorded by XRT originate purely from local crystallographic strain, this technique is not derailed by variations in doping concentration between different wafers.

Figure 2. Pushing the limits of accuracy: Three neighbouring wafers from the same crystal were measured by XRT on both sides, and the average TSD density over the wafer area taken. As well as resolving the decreasing trend of dislocation density from seed to dome, the data reveals which face of the wafer is facing which direction. One might even estimate the spacing between the individual wafers.

While these advantages have much promise, those within the SiC industry seek assurance that the results provided by this non-destructive technique are consistent with those that come from etching. Investigating whether this is the case is our team at the Centre of Expertise for X-ray Topography, a research collaboration between Rigaku Corporation and Fraunhofer IISB. We address market needs by drawing on Rigaku’s excellence in building state-of-the-art X-ray tools and Fraunhofer IISB’s competences along the SiC value chain.

We have found that once those working in industry have been assured of the validity of XRT, their interest in this technique rapidly increases, along with demands, particularly regarding measurement times. To address this specific demand, we have introduced the FastBPD approach: it brings high-speed, full-wafer XRT measurements to labs and industry lines. But let’s start this story at the beginning…

From TSDs…
The most common polytype of SiC substrate is the 4H variant, which is usually grown by physical vapour phase technology. The material that results contains various defects with different properties. Historically, 4H-SiC contained large volumetric defects, like polytype inclusions, but now they tend to be eliminated completely from prime grade material. Unfortunately, micropipes and stacking faults are still present, leading to almost certain device failure. However, the propensity of these imperfections has plummeted in recent years – and for R&D purposes, the presence of these extended defects can be spotted easily by the naked eye in X-ray topograms.

This leaves us with three common types of dislocation: threading screw dislocations (TSDs), threading edge dislocations (TEDs) and basal plane dislocations (BPDs). With the conventional approach to identify these dislocations, standard potassium hydroxide etching of the silicon-face of SiC substrate material, etch pits from TSDs and TEDs are not easily distinguished. Since today’s prime grade material has far fewer TSDs than TEDs, it is a challenge to determine the TSD density reliably. To overcome this issue, technicians can turn to C-face etching, or add oxidising agents to the potassium hydroxide melt. But this introduces new challenges. The reality is that the SiC world has lacked a technique to reliably quantify the TSD density in substrates, often resulting in the omittance of TSD densities in industrial wafer specifications.

An absence of data for the TSD density matters, because this class of dislocation can wreak havoc in some type of device. For example, if TSDs are present in the channel region of MOSFETs, this can promote electric breakdown, leading to device failure. Another issue is that depending on epitaxial conditions, growth pits can occur at the locations of the TSDs, creating device processing issues and ultimately causing device failure.

For the detection of TSDs, XRT is peerless. With correct diffraction conditions, the topographic image almost exclusively exposes individual dark spots, each corresponding to a single TSD (see Figure 1). Simply counting these spots provides the density of TSDs. However, those working within industry have had to be convinced of the capability of XRT before they are willing to make the switch to this technique. They have required reassurance of how the results by XRT relate to those realised by etching, and they needed convincing of the reliability and the accuracy of defect detection.

We have undertaken tests to evaluate how XRT compares against five other experimental techniques. Our goal was to verify that counting based on XRT contrast delivers the same information as conventional approaches.

The five other experimental techniques were: examining etch pits after epitaxy, which allows one to distinguish between TSDs and TEDs; inspecting etch pits, following etching in a melt of potassium hydroxide and sodium peroxide; counting hillocks on the C-face of SiC after etching in potassium hydroxide; using grazing incidence synchrotron XRT, as this can identify TSDs; and scrutinising growth pits after epitaxy, an approach that allows engineers to relate material imperfections to weaknesses associated with device processing. All five comparisons affirmed the capability of XRT, with tests yielding identical dislocation distributions, numbers, and positions. Based on this overwhelming agreement, we are in no doubt whatsoever over the validity of XRT measurements for TSD detection.

If XRT is to be used in industry, tools that apply this technique must deliver reliable results. To confirm that this requirement is met, we developed a measurement and analysis routine – this included measurement parameters, guidelines for the required image quality, and a robust but fast analysis algorithm, which required less than 5 minutes to analyse the full topogram of a 150 mm wafer. We have found that our instruments provided a high measurement repeatability, giving values within 3 percent of one another, and an inter-machine reproducibility with a similar error. Drawing on this exceptional degree of accuracy, engineers can even measure differences in TSD density between neighbouring wafers (see Figure 2).

The opportunity to undertake non-destructive measurements awakens a desire to measure more wafers per SiC boule. However, if this is to happen, there needs to be an increase in throughput. To ensure this is possible, we have adapted partial wafer measurements to the XRT approach. Thanks to this refinement, one can measure the average TSD density of a full 150 mm within 30 minutes with an error of less than 10 percent.

This chapter of this particular story culminated with the publication of SEMI M91, an industrial standard describing TSD detection by XRT. This documentation ensured that XRT is now an established tool for dislocation detection and, in the case of TSDs, has overtaken the old de-facto standard by becoming a regular alternative.