EXTENDED ABSTRACT: Compared to traditional Si materials, compound semiconductor materials such as GaN, SiC, and GaAs exhibit significant advantages in terms of power density, resistance to high temperatures and high pressures, and high-frequency response. Currently, compound semiconductor technology is advancing rapidly, with new structures emerging continuously and epitaxy processes involving complex multi-ffeld coupling. Given the challenges of insufffcient AI empowerment and low research and development (R&D) efffciency in the development and application of new structures, the integration of semiconductor material science with artiffcial intelligence becomes particularly crucial. This paper will introduce examples of R&D and engineering applications of compound semiconductor materials within the AI4CS (Artiffcial Intelligence For Compound Semiconductor) research and production system. Based on large-scale data fusion with consistent relational representations and data enhancement techniques that blend virtual and real data, this paper achieves the construction and efffcient management of a multi-source, heterogeneous database for compound semiconductor materials. It conducts in-depth exploration in the research ffeld of data-driven knowledge mining of compound semiconductor material structure-process-performance relationships. For high-dimensional and sparse sensor data in the material processing process, deep neural networks (DNNs) and recurrent neural networks (RNNs) are employed to construct intelligent prediction models for compound epitaxy process parameters. These models enable precise predictions of multiple key parameters such as concentration and epitaxy rate for critical compound semiconductor materials like GaN HEMT and GaAs PHEMT. Through the establishment of the AI4CS research and production system, it is expected to signiffcantly enhance R&D and production efffciency, providing new avenues for the innovation and development of compound semiconductor materials.

Keywords: compound semiconductor materials; epitaxial structures; AI4CS