Sub-pJ per operation scalable computing: The PULP experience

2016 IEEE SOI-3D-Subthreshold Microelectronics Technology Unified Conference (S3S) Pub Date : 2016-10-01 DOI:10.1109/S3S.2016.7804389

D. Rossi

{"title":"Sub-pJ per operation scalable computing: The PULP experience","authors":"D. Rossi","doi":"10.1109/S3S.2016.7804389","DOIUrl":null,"url":null,"abstract":"Ultra-low power operation and extreme energy efficiency are strong requirements for a number of high-growth Internet of-Things (IoT) applications requiring near-sensor processing. A promising approach to achieve major energy efficiency improvements is near-threshold computing. However, frequency degradation due to aggressive voltage scaling may not be acceptable for performance-constrained applications. The PULP platform leverages multi-core parallelism with explicitly-managed shared L1 memory to overcome performance degradation at low voltage, while maintaining the flexibility and programmability typical of instruction processors. PULP supports OpenMP, OpenCL, and OpenVX parallel programming with hardware support for energy efficient synchronization. Multiple silicon implementations of PULP have been taped out and achieve hundreds of GOPS/W on video, audio, inertial sensor data processing and classification, within power envelopes of a few milliwatts. PULP hardware and software are open-source, with the goal of supporting and boosting an innovation ecosystem focusing on ULP computing for the IoT.","PeriodicalId":145660,"journal":{"name":"2016 IEEE SOI-3D-Subthreshold Microelectronics Technology Unified Conference (S3S)","volume":"31 1","pages":"0"},"PeriodicalIF":0.0000,"publicationDate":"2016-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"4","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"2016 IEEE SOI-3D-Subthreshold Microelectronics Technology Unified Conference (S3S)","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1109/S3S.2016.7804389","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}

引用次数: 4

Abstract

Ultra-low power operation and extreme energy efficiency are strong requirements for a number of high-growth Internet of-Things (IoT) applications requiring near-sensor processing. A promising approach to achieve major energy efficiency improvements is near-threshold computing. However, frequency degradation due to aggressive voltage scaling may not be acceptable for performance-constrained applications. The PULP platform leverages multi-core parallelism with explicitly-managed shared L1 memory to overcome performance degradation at low voltage, while maintaining the flexibility and programmability typical of instruction processors. PULP supports OpenMP, OpenCL, and OpenVX parallel programming with hardware support for energy efficient synchronization. Multiple silicon implementations of PULP have been taped out and achieve hundreds of GOPS/W on video, audio, inertial sensor data processing and classification, within power envelopes of a few milliwatts. PULP hardware and software are open-source, with the goal of supporting and boosting an innovation ecosystem focusing on ULP computing for the IoT.

查看原文

微信好友朋友圈 QQ好友复制链接

本刊更多论文

Sub-pJ / per操作可扩展计算:PULP体验

超低功耗运行和极高的能源效率是许多需要近传感器处理的高增长物联网(IoT)应用的强烈要求。近阈值计算是实现重大能源效率改进的一种有希望的方法。然而，由于积极的电压缩放导致的频率退化对于性能受限的应用可能是不可接受的。PULP平台利用多核并行性和显式管理的共享L1内存来克服低电压下的性能下降，同时保持指令处理器的灵活性和可编程性。PULP支持OpenMP、OpenCL和OpenVX并行编程，硬件支持节能同步。PULP的多个硅实现已经被录下来，在几毫瓦的功率范围内，在视频、音频、惯性传感器数据处理和分类上实现了数百GOPS/W。PULP硬件和软件都是开源的，其目标是支持和推动一个专注于物联网ULP计算的创新生态系统。

本文章由计算机程序翻译，如有差异，请以英文原文为准。

求助全文

约1分钟内获得全文去求助

来源期刊

2016 IEEE SOI-3D-Subthreshold Microelectronics Technology Unified Conference (S3S)

自引率

0.00%

发文量