Track 19: Machine Learning and Intelligent Design of Semiconductor Materials

English: In the post-Moore era, the demand for novel semiconductors such as wide-bandgap materials, 2D materials, and perovskites is surging rapidly for chips, optoelectronics, and power devices. Traditional R&D relies heavily on trial-and-error experiments and high-throughput DFT calculations. However, the vast chemical composition and crystal structure spaces lead to high computational costs and years of development cycles, making it difficult to rapidly screen target materials. Machine learning has thus emerged as a core approach to breaking efficiency bottlenecks, driving a paradigm shift in R&D from empirical trial-and-error to data-driven intelligence. This forum aims to build predictive models based on material databases, first-principles data, and experimental characterization data to quickly and accurately predict key electrical and optical properties such as bandgap, mobility, and defect stability; inversely generate novel semiconductor compositions and structures tailored to device specifications; and construct an autonomous "computation-prediction-synthesis-characterization" closed loop. It focuses on mainstream models like graph neural networks, physics-informed neural networks, Bayesian optimization, and generative models, targeting applications in 2D semiconductors, gallium nitride, perovskite photovoltaics, and heteroepitaxial interfaces, thereby compressing new material development cycles while balancing performance, stability, and fabricability.

中文: 后摩尔时代芯片、光电、功率器件对宽禁带、二维、钙钛矿等新型半导体需求激增。传统研发依赖试错实验与DFT高通量计算，化学组分与晶体结构空间庞大，计算成本高、周期长达数年，难以快速定向筛选目标材料，机器学习成为突破效率瓶颈的核心手段，推动研发从经验试错转向数据驱动的智能范式。本论坛以材料数据库、第一性原理数据、实验表征数据为基础，构建预测模型，快速精准预判带隙、迁移率、缺陷稳定性等关键电学光学性能；反向生成满足器件指标的全新半导体组分结构；搭建“计算-预测-合成-表征”自主闭环，大幅压缩新材料研发周期，兼顾性能、稳定性与制备可行性。聚焦图神经网络、物理信息神经网络、贝叶斯优化与生成模型等主流模型，应用涵盖二维半导体、氮化镓、钙钛矿光伏、异质外延界面等方向，攻克数据稀缺、模型可解释性差等痛点。