Machine Learning-Based Modeling and BV Optimization for LDMOS With Multifloating Buried Layers

This article introduces a novel approach that leverages machine learning (ML) to optimize the breakdown voltage (BV) of lateral double-diffused MOSFET (LDMOS) devices featuring multifloating buried layers (MFBLs). Moving away from the traditional, complex physical derivation methods, our research integrates neural networks with genetic algorithms to forge an adaptive optimization framework. Initially, we examine the physical attributes of MFBL LDMOS to identify crucial parameters affecting BV performance and determine their reasonable value ranges. A dataset is then generated through TCAD simulations, and a convolutional neural network (CNN) is applied to develop a predictive model for the BV of MFBL LDMOS. In the subsequent phase, genetic algorithms are employed to adaptively optimize the structural parameters, facilitating the derivation of an optimal device model. The effectiveness and feasibility of our ML-based model in optimizing device parameters are validated through comprehensive TCAD simulations. This method addresses the complexities and constraints inherent in traditional physical derivation techniques, offering a more efficient route to enhance the performance of composite buried layer power devices.