Alessio Carpegna;Alessandro Savino;Stefano Di Carlo
{"title":"Spiker+: A Framework for the Generation of Efficient Spiking Neural Networks FPGA Accelerators for Inference at the Edge","authors":"Alessio Carpegna;Alessandro Savino;Stefano Di Carlo","doi":"10.1109/TETC.2024.3511676","DOIUrl":null,"url":null,"abstract":"Including Artificial Neural Networks in embedded systems at the edge allows applications to exploit Artificial Intelligence capabilities directly within devices operating at the network periphery, containing sensitive data within the boundaries of the edge device. This facilitates real-time decision-making, reduces latency and power consumption, and enhances privacy and security. Spiking Neural Networks (SNNs) offer a promising computing paradigm in these environments. However, deploying efficient SNNs in resource-constrained edge devices requires highly parallel and reconfigurable hardware implementations. We introduce Spiker+, a comprehensive framework for generating efficient, low-power, and low-area SNN accelerators on Field Programmable Gate Arrays for inference at the edge. Spiker+ presents a configurable multi-layer SNN hardware architecture, a library of highly efficient neuron architectures, and a design framework to enable easy, Python-based customization of accelerators. Spiker+ is tested on three benchmark datasets: MNIST, Spiking Heidelberg Dataset (SHD), and AudioMNIST. On MNIST, it outperforms state-of-the-art SNN accelerators in terms of resource allocation, with a requirement of 7,612 logic cells and 18 Block RAMS (BRAMs), and power consumption, draining only 180 mW, with comparable latency (780 <inline-formula><tex-math>$\\mu$</tex-math></inline-formula>s/img) and accuracy (97%). On SHD and AudioMNIST, Spiker+ requires 18,268 and 10,124 logic cells, respectively, requiring 51 and 16 BRAMs, consuming 430 mW and 290 mW, with an accuracy of 75% and 95%. These results underscore the significance of Spiker+ in the hardware-accelerated SNN landscape, making it an excellent solution for deploying configurable and tunable SNN architectures in resource and power-constrained edge applications.","PeriodicalId":13156,"journal":{"name":"IEEE Transactions on Emerging Topics in Computing","volume":"13 3","pages":"784-798"},"PeriodicalIF":5.4000,"publicationDate":"2024-12-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10794606","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"IEEE Transactions on Emerging Topics in Computing","FirstCategoryId":"94","ListUrlMain":"https://ieeexplore.ieee.org/document/10794606/","RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"COMPUTER SCIENCE, INFORMATION SYSTEMS","Score":null,"Total":0}
引用次数: 0
Abstract
Including Artificial Neural Networks in embedded systems at the edge allows applications to exploit Artificial Intelligence capabilities directly within devices operating at the network periphery, containing sensitive data within the boundaries of the edge device. This facilitates real-time decision-making, reduces latency and power consumption, and enhances privacy and security. Spiking Neural Networks (SNNs) offer a promising computing paradigm in these environments. However, deploying efficient SNNs in resource-constrained edge devices requires highly parallel and reconfigurable hardware implementations. We introduce Spiker+, a comprehensive framework for generating efficient, low-power, and low-area SNN accelerators on Field Programmable Gate Arrays for inference at the edge. Spiker+ presents a configurable multi-layer SNN hardware architecture, a library of highly efficient neuron architectures, and a design framework to enable easy, Python-based customization of accelerators. Spiker+ is tested on three benchmark datasets: MNIST, Spiking Heidelberg Dataset (SHD), and AudioMNIST. On MNIST, it outperforms state-of-the-art SNN accelerators in terms of resource allocation, with a requirement of 7,612 logic cells and 18 Block RAMS (BRAMs), and power consumption, draining only 180 mW, with comparable latency (780 $\mu$s/img) and accuracy (97%). On SHD and AudioMNIST, Spiker+ requires 18,268 and 10,124 logic cells, respectively, requiring 51 and 16 BRAMs, consuming 430 mW and 290 mW, with an accuracy of 75% and 95%. These results underscore the significance of Spiker+ in the hardware-accelerated SNN landscape, making it an excellent solution for deploying configurable and tunable SNN architectures in resource and power-constrained edge applications.
期刊介绍:
IEEE Transactions on Emerging Topics in Computing publishes papers on emerging aspects of computer science, computing technology, and computing applications not currently covered by other IEEE Computer Society Transactions. Some examples of emerging topics in computing include: IT for Green, Synthetic and organic computing structures and systems, Advanced analytics, Social/occupational computing, Location-based/client computer systems, Morphic computer design, Electronic game systems, & Health-care IT.