|dc.description.abstract||In recent years, wide bandgap AlGaN/GaN-on-Silicon high electron mobility transistors (HEMTs) have gradually replaced traditional silicon devices because of their high carrier mobility, high breakdown voltage, high temperature operation and high carrier concentration. However, the performance of GaN HEMTs is still far from its theoretical limits, mostly because of the unavailability of native GaN substrate, resulting in high defect density in the epilayer, such as threading dislocation density, which is typically above 109 cm-2. Wafer bow and/or cracking due to large lattice and thermal mismatch between the silicon substrate and the GaN epilayer is still a concern that needs to be addressed. Additionally, wafer bow could be even serious when the total thickness of the epilayer increases beyond 5 m, which renders the vertical breakdown voltage of the device. Stress relief buffer layer beneath the GaN buffer is often used to minimize the strain caused by the lattice mismatch between the epilayers. Several stress relief buffer layers have been proposed in the past. Nonetheless, graded AlGaN, superlattice (SL) buffers or even the combination of both is most commonly used in modern GaN-on-Si heterostructures. Superlattice buffers in this regards may provide more degree of freedoms, which not only reduces the density of dislocation in the epitaxial layer, but also has the ability to control the wafer bow precisely, resulting in thicker epilayer to achieve higher vertical breakdown voltage. However, the design of superlattice buffer is not straightforward as it involves a number of design parameters, such as the layer composition, thickness, superlattice periods, and position in the buffer. Therefore, it is still a research field of interest.
The present study aims to explore the design and physics of stress control mechanism of AlGaN/GaN HEMTs on silicon using superlattice stress relief layers. To achieve this, we adopted different superlattice buffer layers. For example, we used different aluminum compositions, thicknesses, and periods. The results are consistent with the mathematical Stoney formula. In addition, when compared to the simulated values obtained by a commercial software package, STREEM, we found that the number of SL periods and composition should also be taken into account when controlling the stress during buffer layer growth. Further, in order to allow the superlattice buffer layer to generate more compressive stress through the above method and to maintain the equivalent aluminum composition of the superlattice buffer layer at 50%, the composition was changed from Al0.24Ga0.76N 11 nm/AlN 5.4 to Al0.3Ga0.7N 14 nm/AlN 5.2 nm. The corresponding sum of curvature (κ) calculated from the Stoney formula increases from -0.58 km-1 to -0.87 km-1. Also, the experimental curvature changes from -0.24 km-1 to -0.37 km-1 in each pair, which means that the superlattice buffer layer generates a higher compressive stress than the original one.
In conclusion, our results demonstrated that the SL composition, thickness, and the number of periods need to be optimized to achieve crack-free thick GaN layers on Si. By applying the model developed by the present study, decreasing the SL period from 30 to 20 periods, and optimizing the composition, the edge cracking range of the epilayer on a 150 mm Si substrate could be successfully reduced from 3-5 mm to 2-3 mm, which increases the useable area of the wafer.||en_US|