Compute in‐Memory with Non‐Volatile Elements for Neural Networks: A Review from a Co‐Design Perspective
Source abstract
Abstract
Abstract Deep learning has become ubiquitous, touching daily lives across the globe. Today, traditional computer architectures are stressed to their limits in efficiently executing the growing complexity of data and models. Compute‐in‐memory (CIM) can potentially play an important role in developing efficient hardware solutions that reduce data movement from compute‐unit to memory, known as the von Neumann bottleneck. At its heart is a cross‐bar architecture with nodal non‐volatile‐memory elements that performs an analog multiply‐and‐accumulate operation, enabling the matrix‐vector‐multiplications repeatedly used in all neural network workloads. The memory materials can significantly influence final system‐level characteristics and chip performance, including speed, power, and classification accuracy. With an over‐arching co‐design viewpoint, this review assesses the use of cross‐bar based CIM for neural networks, connecting the material properties and the associated design constraints and demands to application, architecture, and performance. Both digital and analog memory are considered, assessing the status for training and inference, and providing metrics for the collective set of properties non‐volatile memory materials will need to demonstrate for a successful CIM technology.
Source: crossref · Quality: baseline
Research context
OpenAlex topics: Advanced Memory and Neural Computing · Ferroelectric and Negative Capacitance Devices · Neuroscience and Neural Engineering
SciVX discovery tags are terminology-based. OpenAlex topics are source-provided classifications.
Affiliations
Materials Science Division Argonne National Laboratory Lemont IL 60439 USA
Department of Electrical Engineering Purdue University West Lafayette IN 47907 USA
Department of Electrical Engineering Indian Institute of Technology Madras Chennai Tamil Nadu 600036 India
Pritzker School of Molecular Engineering University of Chicago Chicago IL 60637 USA
Funding
U.S. Department of Energy · Office of Science
References (100+)
Show source-provided references
1. Boole G. · 1854
2. Neumann J. v. · 1958
3. A.Turing Intelligent Machinery 1948 http://weightagnostic.github.io/papers/turing1948.pdf(accessed: December2022).
4. 10.1038/nature1696110.1038/nature16961 ↗
5. Krizhevsky A. · 2012
6. 10.1109/CVPR.2009.520684810.1109/CVPR.2009.5206848 ↗
7. Socher R. · 2013
8. Graves A. · 2013
9. D.Amodei D.Hernandez G.Sastry J.Clark G.Brockman I.Sutskever AI and Compute OpenAI https://www.openai.com/blog/ai-and-compute/(accessed: December2022).
10. 10.1038/323533a010.1038/323533a0 ↗
11. 10.1162/neco.1989.1.4.54110.1162/neco.1989.1.4.541 ↗
12. 10.1162/neco.1997.9.8.173510.1162/neco.1997.9.8.1735 ↗
13. Santoro A. · 2016 · 1842
14. 10.1109/TNN.2008.200560510.1109/TNN.2008.2005605 ↗
15. T. B.Brown B.Mann N.Ryder M.Subbiah J.Kaplan P.Dhariwal A.Neelakantan P.Shyam G.Sastry A.Askell S.Agrawal A.Herbert‐Voss G.Krueger T.Henighan R.Child A.Ramesh D. M.Ziegler J.Wu C.Winter C.Hesse M.Chen E.Sigler M.Litwin S.Gray B.Chess J.Clark C.Berner S.McCandlish A.Radford I.Sutskever et al. Language Models are Few‐Shot Learners https://www.arxiv.org/pdf/2005.14165.pdf(accessed: December2020)
16. Raschka S. · 2015
17. 10.1038/387278a010.1038/387278a0 ↗
18. 10.1109/TC.2018.286704810.1109/TC.2018.2867048 ↗
19. Hennessy J. L. · 2019
20. 10.1109/MC.2015.37410.1109/MC.2015.374 ↗
21. Han S. · 2015
22. 10.1147/JRD.2014.238025210.1147/JRD.2014.2380252 ↗
23. Hu M. · 2012 · 498
24. 10.1109/TC.2021.307325510.1109/TC.2021.3073255 ↗
25. 10.1038/s41586-018-0180-510.1038/s41586-018-0180-5 ↗
26. Hu M. · 2016
27. 10.1109/JPROC.2020.300300710.1109/JPROC.2020.3003007 ↗
28. 10.1109/TCSI.2018.284899910.1109/TCSI.2018.2848999 ↗
29. 10.1109/TCSI.2020.298190110.1109/TCSI.2020.2981901 ↗
30. 10.1109/TCSI.2019.290748810.1109/TCSI.2019.2907488 ↗
31. Biswas A. · 2018
32. 10.1109/JETCAS.2020.301425010.1109/JETCAS.2020.3014250 ↗
33. 10.1021/nl904092h10.1021/nl904092h ↗
34. 10.1080/23746149.2016.125958510.1080/23746149.2016.1259585 ↗
35. 10.1063/1.511353610.1063/1.5113536 ↗
36. 10.1109/JPROC.2018.279084010.1109/JPROC.2018.2790840 ↗
37. 10.1109/TED.2020.301517810.1109/TED.2020.3015178 ↗
38. Chen A. · 2015
39. Ankit A. · 2019
40. 10.1109/TVLSI.2019.292924510.1109/TVLSI.2019.2929245 ↗
41. Shafiee A. · 2016
42. 10.1109/LSSC.2021.309335410.1109/LSSC.2021.3093354 ↗
43. Chakraborty I. · 2020
44. 10.3389/fnins.2019.0075310.3389/fnins.2019.00753 ↗
45. 10.1109/TCAD.2020.300018510.1109/TCAD.2020.3000185 ↗
46. Ankit A. · 2017
47. Rasch M. · 2021
48. Sharma T. · 2021 · 1
49. 10.1038/s41928-019-0270-x10.1038/s41928-019-0270-x ↗
50. B.Murmann ADC Performance Survey 1997–2021 https://www.web.stanford.edu/~murmann/adcsurvey.html(accessed: December2022).
51. 10.3389/fnins.2017.0053810.3389/fnins.2017.00538 ↗
52. 10.3389/fnins.2016.0033310.3389/fnins.2016.00333 ↗
53. 10.3389/fnins.2020.0040610.3389/fnins.2020.00406 ↗
54. Zhang B. · 2022
55. 10.3389/fnins.2018.0074510.3389/fnins.2018.00745 ↗
56. 10.3389/frai.2021.69914810.3389/frai.2021.699148 ↗
57. Burr G. W. · 2014
58. 10.1109/JPROC.2018.287105710.1109/JPROC.2018.2871057 ↗
59. Mackin C. · 2020
60. 10.1002/adma.20160431010.1002/adma.201604310 ↗
61. Kim S. · 2019
62. H.Kim M.Rasch T.Gokmen T.Ando H.Miyazoe J. J.Kim J.Rozen S.Kim Zero‐shifting Technique for Deep Neural Network Training on Resistive Cross‐point Arrays arXiv:1907.10228 2019 https://doi.org/10.48550/arXiv.1907.10228.
63. 10.3389/frai.2022.89162410.3389/frai.2022.891624 ↗
64. 10.1109/TED.2021.311599310.1109/TED.2021.3115993 ↗
65. 10.1038/nnano.2016.7010.1038/nnano.2016.70 ↗
66. 10.1038/s41598-018-35542-w10.1038/s41598-018-35542-w ↗
67. 10.1109/TBCAS.2016.252582310.1109/TBCAS.2016.2525823 ↗
68. 10.1021/acs.nanolett.9b0420010.1021/acs.nanolett.9b04200 ↗
69. Lin J. · 2016
70. Zhang J. · 2016
71. Dong Q. · 2020
72. Khwa W. S. · 2018
73. 10.1109/JSSC.2020.300579810.1109/JSSC.2020.3005798 ↗
74. 10.1038/nphoton.2015.18210.1038/nphoton.2015.182 ↗
75. 10.1038/nphoton.2017.9310.1038/nphoton.2017.93 ↗
76. 10.1038/s41467-020-16108-910.1038/s41467-020-16108-9 ↗
77. Sebastian A. · J. Appl. Phys. · 2019 · 52
78. 10.1038/s41928-018-0092-210.1038/s41928-018-0092-2 ↗
79. 10.1038/s41563-019-0291-x10.1038/s41563-019-0291-x ↗
80. 10.1038/nature1444110.1038/nature14441 ↗
81. Jain P. · 2019
82. 10.1038/s41467-018-04484-210.1038/s41467-018-04484-2 ↗
83. Chen W. H. · 2018
84. Xue C. X. · 2019
85. 10.1109/LSSC.2020.301079510.1109/LSSC.2020.3010795 ↗
86. 10.1038/s41586-020-1942-410.1038/s41586-020-1942-4 ↗
87. Tang J. · 2018
88. Solomon P. M. · 2021
89. 10.1109/TVLSI.2017.277695410.1109/TVLSI.2017.2776954 ↗
90. Ranjan A. · 2015
91. 10.1109/TMAG.2010.207592010.1109/TMAG.2010.2075920 ↗
92. Guo Q. · 2013
93. Kang W. · IEEE Trans. Magn. · 2017 · 53 · 1
94. 10.1557/mrc.2020.7110.1557/mrc.2020.71 ↗
95. 10.1109/TED.2022.314223910.1109/TED.2022.3142239 ↗
96. Trentzsch M. · 2016
97. Dünkel S. · 2017
98. Yu S. · 2016
99. 10.1109/JPROC.2012.219036910.1109/JPROC.2012.2190369 ↗
100. 10.1038/nnano.2012.24010.1038/nnano.2012.240 ↗