AI semiconductor hybrid, intelligent transformer/ computation inside memory, has 4X increased LLM inference speed using 2.2X less power

 https://www.eurekalert.org/news-releases/1102359

"as AI model grows/ processed sentences become longer, computational load/ memory requirements surge, leading to speed reductions/ high energy consumption... overcome using sequential memory-based Mamba structure processing information over time, increasing efficiency... further improved with Transformer–Mamba Hybrid Model including PIMBA processing inside memory, combining advantages of both Transformer and Mamba"