Research on the Development of a Multi-Stress Coupled Thermal Resistance Evaluation Fixture for AI Computing Power Chips and the In-Situ Testing Method with Thickness Constraints

With the rapid development of artificial intelligence technology, the heat dissipation of AI computing power chips becomes increasingly challenging due to rising power consumption and heat generation. Thermal interface materials (TIMs), essential for chip heat dissipation, suffer dynamic thermal resistance failure in multi-stress coupling environments, a core issue limiting chip reliability. Existing testing methods fall short in precisely evaluating the thermal resistance changes of TIMs under such conditions.This study proposes a brand-new evaluation method - a rigid thickness-limiting constraint module for heterogeneous interfaces and an in-situ collaborative loading mechanism for multi-stress fields. This study also creatively develops a thickness-rigid constraint in-situ thermal resistance test fixture. The fixture locks amorphous TIMs' thickness (±0.05 mm error) and enables synchronous multi-stress loading, overcoming traditional testing drawbacks. At the same time, it realizes the synchronous loading of multiple stresses including temperature, humidity, temperature cycling, and mechanical stress, breaking through the problems that the testing thickness in the traditional steady-state method is not constant and the thermal resistance change rate cannot be tracked in-situ. In experiments, thermal gels are dispensed into the fixture, and initial thermal resistance is measured. Parallel samples with varying thicknesses undergo combined mechanical shock and reliability tests. Results show that under 1.0 mm thickness, thermal resistance change rates of different gels after 1000 hours of high-temperature testing range from 0% to 18%, unachievable by conventional methods. Through the combined reliability tests, high-temperature stress and temperature cycling stress are identified as the key stresses affecting the reliability of thermal gels, and an explanation of the degradation mechanism is provided.This method has passed systematic verification, providing a reliability evaluation benchmark for the thermal management design of AI computing power chips and promising new breakthroughs in the field of heat dissipation of electronic devices.