Samvel Abrahamyan, Sargis Stepanyan, Hayk Sargsyan, 2026. "Computing-in-Memory Architectures for Deep Learning Acceleration: Recent Advances and Future Trends", International Journal of Review Computing and Information Technology (IJRCIT) 1(1): 105-121.

Deep Learning has seen rapid advancements in the state-of-the-art across numerous fields: computer vision, natural language processing, healthcare, robotics, autonomous systems and scientific computing. Nonetheless, the increasing depth of deep neural networks has dramatically increased their computational requirements, memory needs and energy consumption. Conventional von Neumann computing architectures eventually run into the memory wall problem — that is, data movement between processing elements and memory becomes a significant performance bottleneck. This can be counterproductive for the most modern artificial intelligence systems especially those involving large-scale neural networks and edge AI applications. Computing-in-Memory (CIM) has been established as a breakthrough computing paradigm for in-situ computation supplemented with the necessary data read and write movements within memory arrays, having better computational efficiency by minimizing the data transfer overhead.
The Computing-in-Memory (CIM) architectures employ the emerging technologies in memory devices (such as SRAM, DRAM, Resistive RAM (ReRAM), Phase Change Memory (PCM), Magnetic RAM (MRAM), and memristive devices) to perform arithmetic and neural network computations inside the memory building blocks itself. The architectures benefit them with energy efficiency, throughput, latency reduction and hardware coverage. Most of the ongoing works on CIM systems for deep learning workloads due to recent advances in analog computing, neuromorphic engineering and in memory AI accelerators. Great strides have been made, yet challenges such as device variability, accuracy limits, scalability concerns, reliability issues and software-hardware co-design still remain.
This paper provides an overview of Computing-in-Memory architectures for the Acceleration of Deep Learning. This review covers the architectural foundations, memory technologies, neural network mapping strategies, hardware accelerators, energ/efficiency optimizations and security concerns of in-memory computing based deep learning accelerators for emerging applications. Based on this, the paper then discusses the current research challenges and future directions that will influence the evolution of CIM-enabled artificial intelligence systems. These findings emphasises that Computing-in-memory provides a unique avenue towards providing next generation AI hardware with sustainable, scalable and energy efficient end-to-end deep learning systems.

1. 1. M. Horowitz, “Computing's Energy Problem (And What We Can Do About It),” Ieee International Solid-State Circuits Conference (Isscc), 2014.
2. 2. Song Han, Huizi Mao, And William J. Dally, “Deep Compression: Compressing Deep Neural Networks With Pruning, Trained Quantization And Huffman Coding,” Iclr, 2016.
3. 3. Yann Lecun, Yoshua Bengio, And Geoffrey Hinton, “Deep Learning,” Nature, 2015.
4. 4. Ian Goodfellow, Yoshua Bengio, And Aaron Courville, Deep Learning, Mit Press, 2016.
5. 5. Alex Krizhevsky, Ilya Sutskever, And Geoffrey Hinton, “Imagenet Classification With Deep Convolutional Neural Networks,” Neurips, 2012.
6. 6. Kaiming He Et Al., “Deep Residual Learning For Image Recognition,” Cvpr, 2016.
7. 7. Ashish Vaswani Et Al., “Attention Is All You Need,” Neurips, 2017.
8. 8. Norman P. Jouppi Et Al., “In-Datacenter Performance Analysis Of A Tensor Processing Unit,” Isca, 2017.
9. 9. Yu-Hsin Chen Et Al., “Eyeriss: An Energy-Efficient Reconfigurable Accelerator For Deep Convolutional Neural Networks,” Ieee Journal Of Solid-State Circuits, 2017.
10. 10. Ping Chi Et Al., “Prime: A Novel Processing-In-Memory Architecture For Neural Network Computation In Reram-Based Main Memory,” Isca, 2016.
11. 11. Sheng Lin Et Al., “Fpga-Based Acceleration For Deep Neural Networks,” Ieee Design & Test, 2018.
12. 12. Abu Sebastian Et Al., “Memory Devices And Applications For In-Memory Computing,” Nature Nanotechnology, 2020.
13. 13. Geoffrey W. Burr Et Al., “Neuromorphic Computing Using Non-Volatile Memory,” Advances In Physics: X, 2017.
14. 14. Abhronil Sengupta Et Al., “Spintronic Devices For Neuromorphic Computing,” Nature Electronics, 2018.
15. 15. Shihui Yin Et Al., “A High-Efficiency Reram-Based Cim Accelerator For Deep Neural Networks,” Ieee Transactions On Circuits And Systems, 2021.
16. 16. Abu Sebastian, Manuel Le Gallo, And Evangelos Eleftheriou, “Computational Memory-Based Inference And Training Of Neural Networks,” Nature Communications, 2020.
17. 17. Diederik P. Kingma And Max Welling, “Auto-Encoding Variational Bayes,” Iclr, 2014.
18. 18. Tom Brown Et Al., “Language Models Are Few-Shot Learners,” Neurips, 2020.
19. 19. Jacob Devlin Et Al., “Bert: Pre-Training Of Deep Bidirectional Transformers For Language Understanding,” Naacl, 2019.
20. 20. Manuel Le Gallo Et Al., “Mixed-Precision In-Memory Computing,” Nature Electronics, 2018.
21. 21. Priyadarshini Panda Et Al., “Toward Scalable, Energy-Efficient, And Robust Neuromorphic Computing,” Proceedings Of The Ieee, 2020.
22. 22. Mark Horowitz, “1.1 Computing’s Energy Challenge,” Isscc, 2014.
23. 23. John L. Hennessy And David A. Patterson, Computer Architecture: A Quantitative Approach, Morgan Kaufmann, 2019.
24. 24. Krste Asanović Et Al., “The Landscape Of Parallel Computing Research,” University Of California Report, 2006.
25. 25. Abu Sebastian Et Al., “Tutorial And Survey On In-Memory Computing For Deep Learning,” Nature Electronics, 2023.