Thermal and Mechanical Co-Design of Future Chiplet Systems Using Physics-Informed Machine Learning

Advanced semiconductor packaging is rapidly evolving towards large multi-chiplet architectures to meet the increasing demands of artificial intelligence, high-performance computing, and heterogeneous integration. While this transition enables unprecedented system performance, it also introduces strong thermal and mechanical interactions that can significantly affect package reliability and cooling efficiency.

In large chiplet systems, package warpage and thermo-mechanical stresses can degrade thermal interfaces, leading to increased junction temperatures and long-term reliability concerns. At the same time, the design space has become extremely large due to the combination of different package dimensions, chiplet floorplans, material stacks, power distributions, and cooling solutions. Traditional multi-physics simulations alone are insufficient to explore this design space efficiently.

This PhD project aims to develop a next-generation thermal-mechanical co-design framework that combines physics-based modelling, large-scale simulation, experimental characterization, and Physics-Informed Machine Learning (PIML) to accelerate the design of future chiplet packages. By tackling these challenges, the research will contribute to the next generation of robust, thermally and mechanically optimized chiplet architectures, supporting the growing demands of automotive electronics, AI systems, and edge computing. This is an exciting opportunity for students passionate about thermal engineering, microelectronics, and simulation-driven design to make a tangible impact on future technologies.

The successful candidate will:

• Develop scalable thermal-mechanical co-design methodologies for advanced multi-chiplet systems.
• Investigate the impact of chiplet placement, package scaling, material properties, and power heterogeneity on temperature, stress, and warpage.
• Build compact predictive models for thermal and mechanical performance.
• Develop Physics-Informed Machine Learning (PIML) surrogate models trained using finite-element simulations and experimental data.
• Enable co-optimization of thermal performance and mechanical reliability during package design.
• Validate modelling approaches through dedicated thermal and warpage experiments on advanced chiplet demonstrators.