VQ4SNN: Vector Quantization for Memory-Efficient FPGA Spiking Neural Networks

The article presents VQ4SNN, a novel architecture for Spiking Neural Networks (SNNs) that employs Vector Quantization (VQ) to optimize memory usage on FPGAs. By implementing a two-level memory structure with compact pointers and a shared codebook for quantized weight vectors, VQ4SNN achieves a 52-61% reduction in Block RAM (BRAM) usage compared to existing uncompressed FPGA SNNs, while maintaining inference accuracy and avoiding increased logic utilization. This development is significant for practitioners as it enables more efficient deployment of SNNs on resource-constrained edge devices, facilitating advancements in energy-efficient AI applications.