Migrating computation to memory was proposed a long time ago as a way to overcome the memory bandwidth and latency bottleneck, as well as increase the computation parallelism. While the concept had been applied to several ...