Shumeng Li, Tianqi Xu, Fukun Su, Xian Tang, Yupeng Chen
{"title":"具有单周期串行输入机制的65nm 110GOPS 8T-SRAM内存中计算宏","authors":"Shumeng Li, Tianqi Xu, Fukun Su, Xian Tang, Yupeng Chen","doi":"10.1109/ICCS56666.2022.9936324","DOIUrl":null,"url":null,"abstract":"This paper presents a high-speed 8T-SRAM computing-in-memory (CIM) macro, which adopts a novel single cycle serial input (SCSI) mechanism and a matching weighted capacitor register circuit, achieving excellent input linearity and calculation speed without input DACs. The register capacitor arrays can be reused by the output SAR ADCs as their four most significant bits (MSBs) DAC capacitors, further improving the area efficiency. The 4kb 8T-SRAM macro supports 4-bit input, 4-bit weight, and 6-bit output, it executes 16 columns of $4b\\times 4b$ MAC in parallel, achieving peak throughput of 110.1 GOPS and energy efficiency of 31.4 TOPS/W in 65nm CMOS process.","PeriodicalId":293477,"journal":{"name":"2022 IEEE 4th International Conference on Circuits and Systems (ICCS)","volume":"24 1","pages":"0"},"PeriodicalIF":0.0000,"publicationDate":"2022-09-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"A 65nm 110GOPS 8T-SRAM Computing-in-Memory Macro with Single Cycle Serial Input Mechanism\",\"authors\":\"Shumeng Li, Tianqi Xu, Fukun Su, Xian Tang, Yupeng Chen\",\"doi\":\"10.1109/ICCS56666.2022.9936324\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"This paper presents a high-speed 8T-SRAM computing-in-memory (CIM) macro, which adopts a novel single cycle serial input (SCSI) mechanism and a matching weighted capacitor register circuit, achieving excellent input linearity and calculation speed without input DACs. The register capacitor arrays can be reused by the output SAR ADCs as their four most significant bits (MSBs) DAC capacitors, further improving the area efficiency. The 4kb 8T-SRAM macro supports 4-bit input, 4-bit weight, and 6-bit output, it executes 16 columns of $4b\\\\times 4b$ MAC in parallel, achieving peak throughput of 110.1 GOPS and energy efficiency of 31.4 TOPS/W in 65nm CMOS process.\",\"PeriodicalId\":293477,\"journal\":{\"name\":\"2022 IEEE 4th International Conference on Circuits and Systems (ICCS)\",\"volume\":\"24 1\",\"pages\":\"0\"},\"PeriodicalIF\":0.0000,\"publicationDate\":\"2022-09-23\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"2022 IEEE 4th International Conference on Circuits and Systems (ICCS)\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://doi.org/10.1109/ICCS56666.2022.9936324\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"\",\"JCRName\":\"\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"2022 IEEE 4th International Conference on Circuits and Systems (ICCS)","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1109/ICCS56666.2022.9936324","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
A 65nm 110GOPS 8T-SRAM Computing-in-Memory Macro with Single Cycle Serial Input Mechanism
This paper presents a high-speed 8T-SRAM computing-in-memory (CIM) macro, which adopts a novel single cycle serial input (SCSI) mechanism and a matching weighted capacitor register circuit, achieving excellent input linearity and calculation speed without input DACs. The register capacitor arrays can be reused by the output SAR ADCs as their four most significant bits (MSBs) DAC capacitors, further improving the area efficiency. The 4kb 8T-SRAM macro supports 4-bit input, 4-bit weight, and 6-bit output, it executes 16 columns of $4b\times 4b$ MAC in parallel, achieving peak throughput of 110.1 GOPS and energy efficiency of 31.4 TOPS/W in 65nm CMOS process.