With the artificial intelligence (AI) market witnessing a soaring demand for high-bandwidth memory (HBM), NVIDIA is spearheading an initiative to integrate the supply chain. Its bespoke high-bandwidth memory solution, NVHBM, ingeniously incorporates memory controllers into the substrate at the foundation of the HBM stack. This innovation not only amplifies bandwidth by 30% and cuts power consumption by 15% but also liberates 25% of the space on XPU compute dies. Such groundbreaking technology has spurred a rare partnership between longstanding competitors Samsung Electronics and TSMC. The duo is now joining forces to craft base chips utilizing a 3nm process, aiming to cater to the burgeoning needs of future AI accelerators.
