Not an expert here but the last time I got curious about the state of processor-in-memory my understanding was that the infeasibility of PIM is in large part due to mutually exclusive optimization paths in the fab processes used for CPUs verses for DRAM
