A. Agarwal, S. Hsu, M. Anders, S. Mathew, Gregory K. Chen, Himanshu Kaul, Sudhir K. Satpathy, R. Krishnamurthy
{"title":"A 350mV–900mV 2.1GHz 0.011mm2 regular expression matching accelerator with aging-tolerant low-VMIN circuits in 14nm tri-gate CMOS","authors":"A. Agarwal, S. Hsu, M. Anders, S. Mathew, Gregory K. Chen, Himanshu Kaul, Sudhir K. Satpathy, R. Krishnamurthy","doi":"10.1109/VLSIC.2016.7573514","DOIUrl":null,"url":null,"abstract":"A regular expression matching on-die accelerator, consisting of a hybrid deterministic finite automata (HDFA) and Bloom filter (BLM), with measured 2.1GHz operation, is fabricated in 14nm tri-gate CMOS and occupies 0.011mm2. HDFA integrates probability-based truncation, unique transition-state pairs isolation, parallel common transitions detection, and NFA empty transitions. BLM implements a fused 2-hash NOR match bit-line, 1bit read circuits, and sparse H3 hash. These techniques and aging-tolerant read/write register file circuits (120mV/180mV improved VMIN) achieve 350mV-900mV wide dynamic voltage range with peak throughput of 15.2Gbps-17.1Gbps consuming 3.7mW-4.5mW measured at 750mV, 25°C and maximum energy-efficiency of 17.5Tbps/W-12.5Tbps/W measured at near-threshold 350mV.","PeriodicalId":6512,"journal":{"name":"2016 IEEE Symposium on VLSI Circuits (VLSI-Circuits)","volume":"95 1","pages":"1-2"},"PeriodicalIF":0.0000,"publicationDate":"2016-06-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"2016 IEEE Symposium on VLSI Circuits (VLSI-Circuits)","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1109/VLSIC.2016.7573514","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 0
Abstract
A regular expression matching on-die accelerator, consisting of a hybrid deterministic finite automata (HDFA) and Bloom filter (BLM), with measured 2.1GHz operation, is fabricated in 14nm tri-gate CMOS and occupies 0.011mm2. HDFA integrates probability-based truncation, unique transition-state pairs isolation, parallel common transitions detection, and NFA empty transitions. BLM implements a fused 2-hash NOR match bit-line, 1bit read circuits, and sparse H3 hash. These techniques and aging-tolerant read/write register file circuits (120mV/180mV improved VMIN) achieve 350mV-900mV wide dynamic voltage range with peak throughput of 15.2Gbps-17.1Gbps consuming 3.7mW-4.5mW measured at 750mV, 25°C and maximum energy-efficiency of 17.5Tbps/W-12.5Tbps/W measured at near-threshold 350mV.