A 65nm Compute-In-Memory 7T SRAM Macro Supporting 4-bit Multiply and Accumulate Operation by Employing Charge Sharing

被引:10
作者
Kushwaha, Dinesh [1 ]
Sharma, Aditya [1 ]
Gupta, Neha [1 ]
Raj, Ritik [1 ]
Joshi, Ashish [2 ]
Mishra, Jwalant [3 ]
Kohli, Rajat [3 ]
Miryala, Sandeep [4 ]
Joshi, Rajiv [5 ]
Dasgupta, Sudeb [1 ]
Bulusu, Anand [1 ]
机构
[1] Indian Inst Technol, Dept Elect & Commun, Roorkee, India
[2] Intel Technol India Pvt Ltd, Bangalore, Karnataka, India
[3] NXP Semicond, Bangalore, Karnataka, India
[4] Brookhaven Natl Lab, Upton, NY 11973 USA
[5] Thomas J Watson Res Ctr, Yorktown Hts, NY 10598 USA
来源
2022 IEEE INTERNATIONAL SYMPOSIUM ON CIRCUITS AND SYSTEMS (ISCAS 22) | 2022年
关键词
Analog; bitline; binary; compute-In-Memory (CIM); energy; multiply and accumulate (MAC); PROCESSOR; 6T-SRAM;
D O I
10.1109/ISCAS48785.2022.9937908
中图分类号
TM [电工技术]; TN [电子技术、通信技术];
学科分类号
0808 ; 0809 ;
摘要
In this work, we propose an energy-efficient 64x64 compute-in-memory (CIM) SRAM macro using a 7T bit-cell in 65nm CMOS UMC PDK. It supports 4-bit inputs, 4-bit weights & 4-bit outputs and performs 4-bit MAC operations. It also supports multiple row activations performing 1024 4bx4b multiply and accumulate (MAC) operations in one clock cycle. Inputs are realized by the number of pulses on the read wordline (RWL), which discharges read bitline (RBL) according to bitwise multiplication of weights & inputs. Outputs of 4 columns storing 4-bit weights are then combined via charge sharing to perform a binary-weighted average representing MAC operation, further quantized by a flash analog to digital converter (ADC) giving 4-bit output. The proposed CIM macro achieves an energy efficiency of 28.9 TOPS/W and throughput of 212.9 GOPS operating at supply voltage 1 V with a 2 GHz clock frequency.
引用
收藏
页码:1556 / 1560
页数:5
相关论文
共 28 条
[1]   IMAC: In-Memory Multi-Bit Multiplication and ACcumulation in 6T SRAM Array [J].
Ali, Mustafa ;
Jaiswal, Akhilesh ;
Kodge, Sangamesh ;
Agrawal, Amogh ;
Chakraborty, Indranil ;
Roy, Kaushik .
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS I-REGULAR PAPERS, 2020, 67 (08) :2521-2531
[2]   CONV-SRAM: An Energy-Efficient SRAM With In-Memory Dot-Product Computation for Low-Power Convolutional Neural Networks [J].
Biswas, Avishek ;
Chandrakasan, Anantha P. .
IEEE JOURNAL OF SOLID-STATE CIRCUITS, 2019, 54 (01) :217-230
[3]  
Biswas A, 2018, ISSCC DIG TECH PAP I, P488, DOI 10.1109/ISSCC.2018.8310397
[4]   A Low-Power Convolutional Neural Network Face Recognition Processor and a CIS Integrated With Always-on Face Detector [J].
Bong, Kyeongryeol ;
Choi, Sungpill ;
Kim, Changhyeon ;
Han, Donghyeon ;
Yoo, Hoi-Jun .
IEEE JOURNAL OF SOLID-STATE CIRCUITS, 2018, 53 (01) :115-123
[5]  
Chen YH, 2016, ISSCC DIG TECH PAP I, V59, P262, DOI 10.1109/ISSCC.2016.7418007
[6]   CAP-RAM: A Charge-Domain In-Memory Computing 6T-SRAM for Accurate and Precision-Programmable CNN Inference [J].
Chen, Zhiyu ;
Yu, Zhanghao ;
Jin, Qing ;
He, Yan ;
Wang, Jingyu ;
Lin, Sheng ;
Li, Dai ;
Wang, Yanzhi ;
Yang, Kaiyuan .
IEEE JOURNAL OF SOLID-STATE CIRCUITS, 2021, 56 (06) :1924-1935
[7]  
Dong Q, 2020, ISSCC DIG TECH PAP I, P242, DOI [10.1109/ISSCC19947.2020.9062985, 10.1109/isscc19947.2020.9062985]
[8]  
Gonugondla SK, 2018, ISSCC DIG TECH PAP I, P490, DOI 10.1109/ISSCC.2018.8310398
[9]  
Hinton G., 2012, 2016 ACM IEEE 43 ANN, V29, P82
[10]  
Hubara I, 2018, J MACH LEARN RES, V18