Explore Imec's innovative 3D stacking of HBM on GPUs, optimizing thermal management for enhanced AI performance and efficiency.

IEEE Spectrum's platform is a central hub for technology enthusiasts and professionals, offering insights into  technologies, engineering innovations, and scientific discoveries. Through articles, reports, and interviews, IEEE Spectrum offers insights into emerging technologies, research breakthroughs, and industry trends across various domains. Readers can stay updated with the latest advancements in technology and explore the impact of technology on society and the environment.

IEEE Spectrum

Imec researchers explored stacking high-bandwidth memory (HBM) directly on top of GPUs instead of placing them side-by-side in current 2.5D packages. Initial thermal simulations showed this 3D stacking approach doubled GPU operating temperature to 140°C, making it inoperable. Through system technology co-optimization, including removing redundant silicon layers, reducing GPU clock speed by 50% (enabled by 4x bandwidth increase), merging HBM stacks, and adding bottom-side cooling, the team reduced temperatures back to near-normal 70°C levels. While technically feasible, the optimal configuration between HBM-on-GPU versus GPU-on-HBM remains under investigation.

The Ultimate 3D Integration Would Cook Future GPUs