Liquid Metal Embedded Elastomer Composites for Thermal Management of Next Generation High Performance Computing

Due to the rapid increase in power density of semiconductor devices, particularly for AI training applications, the industry is quickly adopting advanced cooling solutions such as liquid cooling with intricate microchannel designs to more efficiently dissipate heat from these power-hungry devices. Power densities are projected to rise from current values of ~1 W/mm² to 2–4 W/mm² within the next 2–3 years [1]. The stringent thermal requirements on the technology roadmap, combined with large silicon die sizes (4–8× reticle size) and significant warpage driven by coefficient of thermal expansion (CTE) mismatch between substrates, silicon, and other package components, create both challenges and opportunities for novel functional materials be used for interfaces. In this talk, I will present recent efforts to translate advances from the flexible electronics community into semiconductor packaging by adopting liquid metal embedded elastomer (LMEE) [2,3] technology to address next-generation thermal interface material (TIM) challenges. LMEEs exhibit a unique combination of thermal and mechanical properties: high thermal performance enabled by dispersed liquid metal microdroplets, along with the stretchability and adhesion characteristics of soft elastomers. We will review the latest performance and reliability data for these material architectures under emulated environments representative of next-generation AI data centers. Addressing TIM limitations at these power densities presents a significant opportunity to reduce data center power consumption and enable more sustainable growth of high-performance computing infrastructure.