AI-Driven R&D Cost Reduction in Semiconductor Design: A Silvaco Case Study of Advanced Si:Sb-Bi/Al₂O₃ Devices

This work reports a computational optimization study of heavily doped (10⁻³) Si:Sb–Bi/Al₂O₃ heterostructures using AI-enhanced Silvaco TCAD simulations to address key material design trade-offs. A central challenge in this system arises from the inverse relationship between beneficial lattice strain and electronic transport performance. Baseline simulations indicate that although full bismuth incorporation (x = 1.0) produces a maximum compressive strain of 0.70%, it also causes a severe degradation in carrier transport, reducing the effective electron mobility to 80 cm² V⁻¹ s⁻¹ and increasing the interface trap density (D_it) to 2.5 × 10¹² cm⁻² eV⁻¹ as a result of pronounced ionized impurity scattering and defect clustering. To overcome this limitation without extensive experimental trial-and-error, a multivariable machine-learning optimization framework was embedded directly into the TCAD simulation flow. The AI-driven analysis identified an optimal stoichiometric “sweet spot” at a bismuth fraction of approximately x ≈ 0.35. This configuration achieved a balanced trade-off, restoring the carrier mobility to 112 cm² V⁻¹ s⁻¹ (a 40% improvement over the fully doped case), while maintaining a high active carrier concentration of 9.1 × 10¹⁹ cm⁻³ and suppressing inter-face defect densities to a manageable level of 5.8 × 10¹¹ cm⁻² eV⁻¹. From a tech-no-economic standpoint, the proposed simulation-to-fabrication methodology significantly improves development efficiency. The AI-guided workflow effectively replaces approximately thirteen iterative experimental fabrication cycles, leading to an estimated 66% reduction in R&D time-to-market and an overall development cost reduction of about 75%, thereby demonstrating the strong potential of this approach for cost-effective and competitive semiconductor device design.