Gpu-powered method speeds up brain-like computing on specialized chips
GPU-Accelerated Hypergraph Partitioning and Placement to Map SNNs on Neuromorphic Hardware
Distributed, Parallel, and Cluster Computing
Summary
Spiking neural networks (SNNs) are a type of artificial brain that communicate using brief signals called spikes. These networks run on special chips made of many small cores, and how neurons are assigned to these cores affects how fast and energy-efficient the system is. The authors created a new method that uses graphics processors (GPUs) to quickly arrange neurons and reduce communication delays and energy use. Their approach also makes large networks easier to handle by speeding up key steps in the process. Tests show it can cut communication time and energy by a significant amount, all while running much faster than previous methods.
Spiking neural networks (SNNs)Neuromorphic hardwareHypergraph partitioningPlacement optimizationGPU accelerationInter-core communicationMulticastRecursive bisectionEnergy efficiencyLatency
Authors
Marco Ronzani, Cristina Silvano
Abstract
SNNs running on neuromorphic hardware use spikes to achieve sparse and energy-efficient communication over a mesh of cores. In turn, system performance heavily depends on the assignment of neurons to cores: the mapping. Since hardware features inter-core multicast and intra-core replication of spikes, we model SNNs as hypergraphs to exploit both opportunities for reducing communication traffic. Mapping thus comprises two NP-hard problems: hypergraph partitioning and placement on the lattice of cores. High-quality solutions to both are critical, yet increasingly difficult as networks scale to millions of neurons. Therefore, we propose a GPU-accelerated pipeline for SNN mapping: a multi-level partitioning scheme is devised around hardware constraints, while placement is initialized through recursive bisection, followed by refinement pulling together strongly connected cores through repeated swaps. Model-based experiments show upwards of 16% lower latency and 42% lower energy for spike movements over existing sequential tools, while our parallel mapper is on average 18-280x faster.