6 July 2018 in IEEE Electron Device Letters
We perform a system-level simulation of hardware spiking neural network (SNN) consisting of silicon-based synaptic transistors and integrate-and-fire (I&F) neuron circuits. Using electrical models of the synaptic device and I&F neuron circuit, a three-layer fully connected SNN in hardware is presented for MNIST pattern recognition by means of ex situ training. Right-justified rate coding is employed as an information encoding method, and negative weight values are implemented by a pair of the synaptic transistors (specifically, excitatory and inhibitory synapses). Furthermore, the variability effect occurring in the devices and circuits is demonstrated. This result indicates that the system has tolerance to the variations and how precisely the variations need to be controlled for hardware SNN applications.
26 October 2022 in ACM Journal on Emerging Technologies in Computing Systems
Spiking neural networks (SNNs) are brain-inspired event-driven models of computation with promising ultra-low energy dissipation. Rich network dynamics emergent in recurrent spiking neural networks (R-SNNs) can form temporally based memory, offering great potential in processing complex spatiotemporal data. However, recurrence in network connectivity produces tightly coupled data dependency in both space and time, rendering hardware acceleration of R-SNNs challenging. We present the first work to exploit spatiotemporal parallelisms to accelerate the R-SNN-based inference on systolic arrays using an architecture called SaARSP. We decouple the processing of feedforward synaptic connections from that of recurrent connections to allow for the exploitation of parallelisms across multiple time points. We propose a novel time window size optimization (TWSO) technique, to further explore the temporal granularity of the proposed decoupling in terms of optimal time window size and reconfiguration of the systolic array considering layer-dependent connectivity to boost performance. Stationary dataflow and time window size are jointly optimized to trade off between weight data reuse and movements of partial sums, the two bottlenecks in latency and energy dissipation of the accelerator. The proposed systolic-array architecture offers a unifying solution to an acceleration of both feedforward and recurrent SNNs, and delivers 4,000X EDP improvement on average for different R-SNN benchmarks over a conventional baseline.