This study introduces a novel application of machine learning using indium phosphide heterojunction bipolar transistors as an example. The objective is to predict the device performance and optimize the device structure by utilizing an artificial neural network (ANN) to calculate the device direct current (DC) and frequency characteristics. To this end, we develop a physics-inspired ANN that emphasizes the significance of the first-order partial derivative of the current over voltage. The ANN is trained on a data set generated by technology computer-aided design simulations, covering a range of voltage setups, device geometries, and doping concentrations. The resulting model accurately predicts the DC and frequency characteristics of the device, and obtain key performance indicators such as the DC current amplification factor, cut-off frequency, and maximum oscillation frequency. This approach can significantly speed up the device parameter optimization and provide a potential numerical tool for design technology co-optimization.

本研究以磷化铟异质结双极晶体管为例介绍了机器学习的一种新颖应用。目的是通过利用人工神经网络（ANN）计算器件直流（DC）和频率特性来预测器件性能并优化器件结构。为此，我们开发了一种受物理启发的人工神经网络，强调电流对电压的一阶偏导数的重要性。人工神经网络接受由技术计算机辅助设计模拟生成的数据集的训练，涵盖一系列电压设置、器件几何形状和掺杂浓度。所得模型准确预测了器件的直流和频率特性，并获得了直流电流放大系数、截止频率和最大振荡频率等关键性能指标。这种方法可以显着加快器件参数优化速度，并为设计技术协同优化提供潜在的数值工具。