Follow the latest coverage, related explainers and connected technology stories.
A paper from UC Berkeley and FuriosaAI examines the use of High Bandwidth Flash to expand the memory of AI accelerators when running large language models. The simulations indicate a reduction in completion time of up to 87%, with potential improvements in energy consumption and write endurance through cache-aware scheduling.
Production capacity for substrates is no longer sufficient to meet advanced packaging needs; package size, layer count, interconnect precision, and integrated functions make each design increasingly similar to a specialized manufacturing process. This affects materials, equipment, qualification, and supply sources, particularly in artificial intelligence, high-performance computing, integrated optics, and automotive packages.
Researchers from Rensselaer Polytechnic Institute and IBM introduced the REACH architecture for managing long-span error-correcting codes inside an HBM controller, targeting reduced decoding overhead in large language model inference workloads. The approach relies on inner codes for common errors and an outer code to repair parts known to be erased.