Beyond Wisdom in III-V Optoelectronics: Physics-Enhanced Neural Networks Unveil the Role of Cubic Boron Arsenide.

Abstract

Cubic boron arsenide (BAs), an emerging III-V semiconductor renowned for its exceptional transport properties, is gaining recognition as a promising material for next-generation electronics and optoelectronics. Despite advances exploiting BAs as an individual semiconductor, its broader role within the III-V semiconductor family remains underappreciated. Here we propose the BAs-III-V material platform as active layers to advance beyond conventional wisdom in III-V optoelectronics since the crystal and band structure of BAs endow the platform with high bandgap tunability. To accelerate the generation of materials for target substrates and bandgaps (TGTMatGen), we extracted structural information from synthetic BAs crystals to explore lattice-matched BAs-III-V and developed physics-enhanced deep neural networks (PhysenNet) to access bandgaps. PhysenNet excels in bandgap prediction, achieving a mean absolute error of 0.0251 eV between predicted and experimental values. Guided by TGTMatGen, we designed two previously unattainable devices-a 1.55 µm surface-emitting laser on GaAs and a 3 µm photodetector on InP-both demonstrating commendable performance. We believe our findings contribute not only to the advancement of optoelectronics but also to the broader semiconductor landscape through interdisciplinary innovation.