This paper presents an efficient and accurate artificial intelligence (AI) model, to calculate the emerging monatomic phase change material antimony, by leveraging the cutting-edge AI technologies like active learning (AL) and deep learning (DL). The required dataset size of the AI model training is reduced by about one order of magnitude, by using the state-of-the-art AL technology, leading to efficient AI model training. It is shown that the potential energy surface, phonon dispersion, coordination number, radial distribution function, angular distribution function, and phase transition processes calculated by using the proposed AI model could agree well with those calculated by using DFT and DFT-based MD, indicating decent accuracy of the proposed AI model. It is well known that the simulation time of the proposed AI model scales as O(n), which is much more favorable than the O(n³) scaling scenario of the widely-used DFT and DFT-based MD, paving a way to future-generation fully-atomistic device simulations free of finite-size effects.

本文提出了一种高效、准确的人工智能 (AI) 模型，利用主动学习 (AL) 和深度学习 (DL) 等前沿人工智能技术来计算新兴的单原子相变材料锑。AI模型训练所需的数据集大小通过最先进的AL技术减少了大约一个数量级，从而实现了高效的AI模型训练。结果表明，使用所提出的 AI 模型计算的势能面、声子色散、配位数、径向分布函数、角分布函数和相变过程与使用 DFT 和基于 DFT 的 MD 计算的结果非常吻合，表明所提出的 AI 模型的准确度不错。众所周知，所提出的 AI 模型的模拟时间为O ( n )，这比广泛使用的 DFT 和基于 DFT 的 MD的O ( n ³ ) 缩放场景更有利，为未来无有限尺寸效应的全原子器件模拟铺平了道路。