Thermal and Mechanical Optimization to Enable Reliable High Performance Liquid-Cooled Compute System for Level 4 Autonomous Driving

In this paper, we report an interesting case study on the thermal-mechanical design and process optimization of a liquid-cooled compute system for autonomous vehicle (AV) applications. Two different implementations for an off-the-shelf AI processor inside the AV compute are presented for comparison purposes. The first implementation imposes a great challenge on the AI processor package reliability because of the interactions between the liquid cooling system and the printed circuit board assembly (PCBA) under severe vehicle operating conditions. We observed the AI processor reaching its thermal trip during high-temperature operational reliability stress. This thermal trip event was characterized as either a time-dependent degradation or an abrupt total failure of two critical thermal interfaces along the processor thermal conduction path. The driving force to cause the AI processor thermal trips was identified to be an external loading on the GPU lid generated by motherboard deformation due to high temperature-induced thermal stress. The actual thermal trip occurrence rate is quantitatively estimated by a compound probability distribution-based stress-strength analysis. The strength was referred to as the processor lid holding strength on the TIM1/Lid and TIM2/Lid interfaces. The stress was referred to as the loading force on the processor lid due to PCBA and system interaction. On the other hand, the second implementation takes proper considerations of the motherboard straining engineering through PCBA process improvement, lid strength incoming quality pre-screen, and vehicle level coolant operation condition optimization. As a result, the second implementation addresses the stress on the processor lid generated by the motherboard deformation at high temperatures, making it meet the reliability requirements for automotive vehicle applications.