DCGAN inference on a microcontroller: 12.6M parameters, 512KB SRAM, 26-second generation, pure C [P]

In three linesDCGAN with 12.6M parameters runs on RISC-V CH32H417 microcontroller (512KB SRAM). Generates 64×64 cat faces in 26 seconds using pure C inference engine with int8 per-channel quantization. Weights streamed from SD card via double buffering. Z vector seeded with 200 bytes quantum random data (ANU QRNG). No existing frameworks (TFLite, CMSIS NN) — built from scratch.

Summary generated by Claude — human-verified