Approximate stochastic GEMM accelerator on Zynq UltraScale+ — bit-serial 22×22 systolic array using XNOR multiply and stochastic number generators. Zero DSPs, AXI-Stream/Lite interface, PetaLinux driver, and image convolution benchmark (Gaussian blur / Sobel, PSNR/SSIM). Targeting Avnet Ultra96-V2. - View it on GitHub