AI/ML Based 100X Faster Thermal Analysis Methodology With 95% Accuracy

The new generation SOCs requires a thermal solution to maintain temperatures within operating limits. Design and optimization of suitable thermal solution and upfront thermal analysis are key to design thermal friendly design based on all critical workloads in very fast manner. There are two common methods of predicting SOC thermal performance: computational simulation and experimental measurement. These two approaches include complicated operations and experimental setup. Thus, it is quite difficult to build computational simulations that fully capture the complex logical relationships between the properties of a material, geometry, power and their related factors, and some of these relationships may even be unknown. Therefore, there is an urgent need to develop intelligent and high-performance prediction models that can correctly predict the SOC thermal solution at a low temporal and computational cost. In this work, a new methodology for optimization of SOC Thermal Performance process is developed, using Regression-Based Reduced-Order Modelling Techniques. This methodology can be applied to any type of platform configurations to reduce computation effort, results reduction in overall design time and cost. Also, this approach (Implemented through software applications-tool) can be deployed internally as well as customer experience thermal tool, allows customers to predict the optimum thermal solutions tailored to their project requirements. This tool is developed using blending different regression models for better prediction of SOC thermal performance, it comprises different types of heat sink models (extruded, folded fin, heat pipe embedded, and vapor chamber) allows user to select right tradeoff between performance and cost. In this work, a new methodology to estimate thermal performance analysis is demonstrated for complex inhouse SOC with about 100X faster and with about 95% accurate reference to industry solutions available. In this analysis all data are normalized to ensure sensitive design details are protected