2D Thermal Contour Modeling of 14 nm SOI FinFET Using Machine Learning for Efficient Thermal Profile Prediction

Recent advancements in the semiconductor industry have emphasized the need to meet demanding criteria for time, power, and performance. Novel devices such as FinFETs and nanosheets have emerged to tackle short-channel effects, albeit with diminished thermal conductivity. Notably, SOI FinFETs introduces thermal bottlenecks, exacerbating concerns regarding self-heating. While Finite Element Method (FEM) modelling aids in comprehending thermal properties and predicting temperatures, its time-intensive nature poses challenges. The growing interest in FEM supplemented Machine Learning (FSML) presents a viable and cost-effective alternative. This study focuses on lever-aging machine learning to predict SOI FinFET thermal profiles, aiming to reduce computational costs. Datasets is generated using a 14 nm FinFET on the Synopsys TCAD FEM simulator. CNN and VAE models based ML model developed for the same. The obtained results demonstrate promising accuracy and efficiency, paving the way for optimized device designs and thermal management strategies. Evaluation metrics, including MSE, accuracy, loss, IoU, and SSIM, underscore the effectiveness of the proposed methodology