FeNN-DMA: A RISC-V SoC for SNN acceleration

--Spiking Neural Networks (SNNs) are a promising, energy-efficient alternative to standard Artificial Neural Networks (ANNs) and are particularly well-suited to spatio-temporal tasks such as keyword spotting and video classification. However, SNNs have a much lower arithmetic intensity than ANNs and are therefore not well-matched to standard accelerators like GPUs and TPUs. Field Programmable Gate Arrays (FPGAs) are designed for such memory-bound workloads and here we develop a novel, fully-programmable RISC-V-based system-on-chip (FeNN-DMA), tailored to simulating SNNs on modern UltraScale+ FPGAs. We show that FeNN-DMA has comparable resource usage and energy requirements to state-of-the-art fixed-function SNN accelerators, yet it is capable of simulating much larger and more complex models. Using this functionality, we demonstrate state-of-the-art classification accuracy on the Spiking Heidelberg Digits and Neuromorphic MNIST tasks. RTIFICIAL Neural Networks (ANNs) have demonstrated super-human performance in areas ranging from image classification to language modelling. However, training current ANNs, and even simply performing inference with them, come at a high energy cost, meaning they face significant limitations in their practical adoption. The human brain provides a tantalising existence proof that a far more efficient form of neural network is possible, as it runs on only 20 W and is far more powerful and flexible than any current ANN. Some of these properties are encapsulated in a biologically-inspired type of ANN known as Spiking Neural Networks (SNNs), in which individual neurons are stateful, dynamical systems and communicate with each other using spatio-temporally sparse events known as spikes. The main energy savings in SNNs come from this event-based communication because, by removing the continuous exchange of activations, the costly matrix multiplication of weights and activations at the heart of ANN computation is replaced by simply adding the weights associated with spiking neurons. This is particularly effective when spikes are rare events.