NPU Fundamentals 2

The previous room covered NPUs: chips built to accelerate the matrix multiplication a trained neural network needs, done fast and efficiently, but still following the traditional design where memory and the processor are separate, and data has to travel between them for every operation.

Neuromorphic computing asks a different question: instead of building better hardware for that traditional design, what if the hardware itself worked more like a biological brain -- where memory and computation are not separated at all, and neurons only do work when they actually receive a signal? That is the idea this room explores.