On-demand growth of semiconductor heterostructures guided by physics-informed machine learning

Developing tailored heterostructures on demand is essential to meet the growing needs of semiconductor devices. However, traditional methods remain constrained by simulation-based design and iterative trial-and-error optimization. Here, we introduce SemiEpi, a self-driving platform designed for molecular beam epitaxy (MBE) that enables multi-step semiconductor heterostructure growth through in situ Reflection High Energy Electron Diffraction monitoring and on-the-fly feedback control. By integrating MBE reactors, physics-informed machine learning (ML) models, and parameter initialization, SemiEpi designs heterostructures, identifies optimal initial conditions, and proposes experiments for material growth. As a demonstration, we optimized high-density InAs quantum dot growth with a target emission wavelength of 1240 nm, achieving a density of 5 × 10¹⁰ cm^-2, a 1.6-fold increase in photoluminescence intensity and a reduced full width at half maximum of 29.13 meV through feedback control of growth temperatures. We further demonstrate SemiEpi's versatility across different MBE reactors, highlighting its potential to address challenges in multi-step heterostructure growth, enable hardware-independent frameworks, and enhance process repeatability and stability.