Compute in‐Memory with Non‐Volatile Elements for Neural Networks: A Review from a Co‐Design Perspective