The new analogue in-memory chip performs computation inside memory, lowering power consumption and latency for AI and data-center workloads.