Integrated Circuits and Systems >
Linearity Performance of Charge Domain In-Memory Computing: Analysis and Calibration
![]() |
HENG ZHANG received the B.S. and M.S. degrees in microelectromechanical systems from North- western Polytechnical University, Xi’an, China, in 2018 and 2021, respectively. He is currently working toward the Ph.D. degree with the School of Electronic Science and Engineering, Nanjing University, Nanjing, China. His research interests include analog and mixed-signal circuit design for deep learning accelerators, and optical packet switching networks for data centers. |
![]() |
YICHUAN BAI (Graduate Student Member, IEEE) received the B.S. and M.S. degrees from the School of Electronic Science and Engineering, Nanjing University, Nanjing, China, in 2019 and 2022, respectively. He is currently working toward the Ph.D. degree in electronic science and tech- nology. His research interests include machine- learning hardware accelerators and neural network compilers. |
![]() |
JUNJIE SHEN majoring in electronic information engineering with the Nanjing University of Posts and Telecommunications, Nanjing, China. He will soon study for a graduate degree with the School of Electronic Science and Engineering, Nanjing University, Nanjing. His research interests include in-memory com- puting architecture and digital integrated circuits, and signal processing and FPGA implementations |
![]() |
YUAN DU (Senior Member, IEEE) received the B.S. degree from Southeast University, Nanjing, China, in 2009, the M.S. and Ph.D. degrees from the Electrical Engineering Department, University of California, Los Angeles, Los Angeles, CA, USA, in 2012 and 2016, respectively. Since 2019, he has been an Associate Professor with Nanjing University, Nanjing. From 2016 to 2019, he was a leading hardware Architect with Kneron Inc., San Diego, CA. His research interests include the de- signs of machine-learning hardware accelerators, high-speed inter-chip/intra-chip interconnects, and RFICs. He was the re- cipient of the Microsoft Research Asia Young Fellow in 2008, Southeast University Chancellor’s Award in 2009, Broadcom Young Fellow in 2015, and IEEE Circuits and Systems Society Darlington Best Paper Award in 2021. |
![]() |
LI DU (Member, IEEE) received the B.S. de- gree from Southeast University, Nanjing, China, in 2011, and the Ph.D. degree in electrical engineer- ing from the University of California, Los Angeles, Los Angeles, CA, USA, in 2016. From June 2013 to September 2016, he was with Qualcomm Inc., San Diego, CA, designing mixed-signal circuits for cellular communications. From September 2016 to October 2018, he was a Hardware Architect Re- search Scientist with Kneron Inc., San Diego, CA, designing high-performance artificial intelligence hardware accelerator. After that, he joined Xin Yun Tech Inc., Westlake, CA, in charge of high-speed analog circuits design for 100G/400G optical communication. He is currently an Associate Professor with the Depart- ment of Electrical Science and Engineering, Nanjing University. His research includes analog sensing circuit design, in-memory computing design, and high-performance AI processors for edge sensing. |
Received date: 2024-03-04
Revised date: 2024-04-11
Accepted date: 2024-05-10
Online published: 2024-11-27
Supported by
National Key Research and Development Program of China under Grant(2022YFB4400900)
Natural Science Foundation of China under Grant(62371223)
Deep learning has recently gained significant prominence in various real-world applications such as image recognition, natural language processing, and autonomous vehicles. While deep neural networks appear to have different architectures, the main operations within these models are matrix-vector multiplications (MVM). Compute-in-memory (CIM) architectures are promising solutions for accelerating the massive MVM operations by alleviating the frequent data movement issue in traditional processors. Ana log CIM macros leverage current-accumulating or charge-sharing mechanisms to perform multiply-and-add (MAC) computations. Even though they can achieve high throughput and efficiency, the computing accuracy is sacrificed due to the analog nonidealities. To ensure precise MAC calculations, it is crucial to analyze the sources of nonidealities and identify their impacts, along with corresponding solutions. In this paper, comprehensive linearity analysis and dedicated calibration methods for charge domain static-random access memory (SRAM) based in-memory computing circuits are proposed. We analyze nonidealities from three areas based on the mechanism of charge domain computing: charge injection effect, temperature variations, and ADC reference voltage mismatch. By designing a 256 × 256 CIM macro and conducting investigations via post-layout simulation, we conclude that these nonidealities don’t deteriorate the computing linearity, but only cause the scaling and bias drift. To mitigate the scaling and bias drift identified, we propose three calibration methods ranging from the circuit level to the algorithm level, all of which exhibit promising results. The comprehensive analysis and calibration methods can assist in designing CIM macros with more accurate MAC computations, thereby supporting more robust deep learning inference.
HENG ZHANG , YICHUAN BAI , JUNJIE SHEN , YUAN DU , LI DU . Linearity Performance of Charge Domain In-Memory Computing: Analysis and Calibration[J]. Integrated Circuits and Systems, 2024 , 1(1) : 43 -53 . DOI: 10.23919/ICS.2024.3422968
[1] |
|
[2] |
|
[3] |
OpenAI et al., “GPT-4 technical report,” 2023, arXiv:2303.08774.
|
[4] |
|
[5] |
|
[6] |
|
[7] |
|
[8] |
|
[9] |
|
[10] |
|
[11] |
|
[12] |
|
[13] |
|
[14] |
|
[15] |
|
[16] |
|
[17] |
|
[18] |
|
[19] |
|
[20] |
|
[21] |
|
[22] |
|
[23] |
|
[24] |
|
[25] |
|
[26] |
|
[27] |
|
[28] |
|
[29] |
|
[30] |
|
[31] |
|
[32] |
|
[33] |
|
[34] |
|
[35] |
|
[36] |
|
[37] |
|
[38] |
|
[39] |
|
/
〈 |
|
〉 |