Tiny TPU

Description:

TinyTPU is a complete, silicon-targeted neural-network inference accelerator in the spirit of Google's TPUv1, scaled to a TinyTapeout Sky130 die: a weight-stationary systolic array with a custom 13- instruction ISA, double-buffered software-managed scratchpads, a bias/requantize/ReLU output pipeline, hardware performance counters, and a UART host interface - plus the full software stack (assembler, numpy-to-ISA compiler, golden model, pyserial driver) and a self-checking cocotb verification environment.