Although I understand the push to bigger and bigger infrastructure for AI, I really don’t understand why Microsoft, Google, Amazon,… persist in using energy inefficient silicon transistors.
At this stage, it seems obvious that there is a need to scale-up spintronics MRAM manufacturing (Avalanche technolog, Everspin,…) because it is a key step in enabling spintronics technology which is a beyond CMOS technology that could enable much more energy efficient digital logic computing , while having the great advantage to also be more amenable to AI.
Beyond CMOS technology like spintronics and Non-Volatile-Memory MRAM (SOT-MRAM, VCMA MRAM,…) would help enable things like Intel MESO concept, and I guess current AI infrastructure could actually be used to fastrack MRAM R&D (to find the right combination of materials,…).