Tiny TPU

A toy Tensor Processing Unit (TPU) is a custom chip designed to accelerate machine learning inference and training. It solves the problem of slow ML workloads by performing only math operations, unlike GPUs which also handle rendering. The project targets developers and engineers interested in hardware design and ML fundamentals, particularly those seeking a well-documented open-source ML accelerator. It is positioned as a re-invention of the TPU, built from a novice perspective with a hacky-first approach, and delivered as an open-source hardware design.