Skydiver: A Spiking Neural Network Accelerator Exploiting Spatio-Temporal Workload Balance
This work addresses energy efficiency issues in SNN hardware accelerators for applications like image processing, though it appears incremental as it builds on existing SNN methods with specific optimizations.
The paper tackled the problem of unpredictable and unbalanced workloads in Spiking Neural Network (SNN) accelerators, which degrade energy efficiency, by proposing Skydiver, an FPGA-based accelerator that exploits spatio-temporal workload balance, resulting in improved throughput by 1.4X and 1.2X on image segmentation and MNIST classification tasks, achieving 22.6 KFPS throughput and 42.4 uJ/Image prediction energy with 98.5% accuracy.
Spiking Neural Networks (SNNs) are developed as a promising alternative to Artificial Neural networks (ANNs) due to their more realistic brain-inspired computing models. SNNs have sparse neuron firing over time, i.e., spatio-temporal sparsity; thus, they are useful to enable energy-efficient hardware inference. However, exploiting spatio-temporal sparsity of SNNs in hardware leads to unpredictable and unbalanced workloads, degrading the energy efficiency. In this work, we propose an FPGA-based convolutional SNN accelerator called Skydiver that exploits spatio-temporal workload balance. We propose the Approximate Proportional Relation Construction (APRC) method that can predict the relative workload channel-wisely and a Channel-Balanced Workload Schedule (CBWS) method to increase the hardware workload balance ratio to over 90%. Skydiver was implemented on a Xilinx XC7Z045 FPGA and verified on image segmentation and MNIST classification tasks. Results show improved throughput by 1.4X and 1.2X for the two tasks. Skydiver achieved 22.6 KFPS throughput, and 42.4 uJ/Image prediction energy on the classification task with 98.5% accuracy.