Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
156 tokens/sec
GPT-4o
7 tokens/sec
Gemini 2.5 Pro Pro
45 tokens/sec
o3 Pro
4 tokens/sec
GPT-4.1 Pro
38 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

DAISM: Digital Approximate In-SRAM Multiplier-based Accelerator for DNN Training and Inference (2305.07376v2)

Published 12 May 2023 in cs.AR and cs.LG

Abstract: DNNs are widely used but face significant computational costs due to matrix multiplications, especially from data movement between the memory and processing units. One promising approach is therefore Processing-in-Memory as it greatly reduces this overhead. However, most PIM solutions rely either on novel memory technologies that have yet to mature or bit-serial computations that have significant performance overhead and scalability issues. Our work proposes an in-SRAM digital multiplier, that uses a conventional memory to perform bit-parallel computations, leveraging multiple wordlines activation. We then introduce DAISM, an architecture leveraging this multiplier, which achieves up to two orders of magnitude higher area efficiency compared to the SOTA counterparts, with competitive energy efficiency.

Definition Search Book Streamline Icon: https://streamlinehq.com
References (23)
  1. Y.-H. Chen et al., “Eyeriss: an energy-efficient reconfigurable accelerator for deep convolutional neural networks,” IEEE JSSC, vol. 52, no. 1, pp. 127–138, Jan 2017.
  2. I. Qiqieh et al., “Energy-efficient approximate multiplier design using bit significance-driven logic compression,” in DATE’17, Mar 2017, pp. 7–12.
  3. Y. Guo et al., “Design of power and area efficient lower-part-OR approximate multiplier,” in TENCON’18, Oct 2018, pp. 2110–2115.
  4. G. Singh et al., “Near-memory computing: past, present, and future,” MICPRO, vol. 71, no. C, Nov 2019.
  5. T.-J. Yang et al., “A method to estimate the energy consumption of deep neural networks,” in ACSSC’17, 2017, pp. 1916–1920.
  6. T. Luo et al., “Energy efficient in-memory integer multiplication based on Racetrack memory,” in ICDCS’20, Nov 2020, pp. 1409–1414.
  7. H. Jin et al., “ReHy: a ReRAM-based digital/analog hybrid PIM architecture for accelerating CNN training,” IEEE TPDS, vol. 33, no. 11, pp. 2872–2884, Dec 2022.
  8. M. Hassanpour et al., “A survey of near-data processing architectures for neural networks,” MAKE, vol. 4, no. 1, pp. 66–102, Jan 2022.
  9. S. Mittal, “A survey of ReRAM-based architectures for processing-in-memory and neural networks,” MAKE, vol. 1, no. 1, pp. 75–114, Apr 2018.
  10. J.-H. Kim et al., “Z-PIM: a sparsity-aware processing-in-memory architecture with fully variable weight bit-precision for energy-efficient deep neural networks,” IEEE JSSC, vol. 56, no. 4, pp. 1093–1104, Jan 2021.
  11. J. Heo et al., “T-PIM: an energy-efficient processing-in-memory accelerator for end-to-end on-device training,” IEEE JSSC, vol. 58, no. 3, pp. 600–613, Nov 2023.
  12. S. Hamdioui et al., “Memristor based computation-in-memory architecture for data-intensive applications,” in DATE’15, Mar 2015, pp. 1718–1725.
  13. S. Shresthamali et al., “FAWS: fault-aware weight scheduler for DNN computations in heterogeneous and faulty hardware,” in ISPA’22, Dec 2022, pp. 204–212.
  14. J. Wang et al., “A 28-nm compute SRAM with bit-serial logic/arithmetic operations for programmable in-memory vector computing,” IEEE JSSC, vol. 55, no. 1, pp. 76–86, Sep 2020.
  15. Q. Dong et al., “A 0.3v VDDmin 4+2t SRAM for searching and in-memory computing using 55nm DDC technology,” in VLSIC’17, Jun 2017, pp. C160–C161.
  16. N. Burgess et al., “Bfloat16 processing for neural networks,” in ARITH’19, Jun 2019, pp. 88–91.
  17. P. Yin et al., “Design and performance evaluation of approximate floating-point multipliers,” in ISVLSI’16, Sep 2016, pp. 296–301.
  18. K. He et al., “Deep residual learning for image recognition,” Dec 2015.
  19. O. Russakovsky et al., “ImageNet large scale visual recognition challenge,” IJCV, vol. 115, no. 3, pp. 211–252, Dec 2015.
  20. R. Balasubramonian et al., “CACTI 7: new tools for interconnect exploration in innovative off-chip memories,” ACM TACO, vol. 14, no. 2, Jun 2017.
  21. N. P. Jouppi et al., “CACTI-IO: CACTI with off-chip power-area-timing models,” IEEE TVLSI, vol. 23, no. 7, pp. 1254–1267, Aug 2015.
  22. Y. N. Wu et al., “Accelergy: an architecture-level energy estimation methodology for accelerator designs,” in ICCAD’19, Nov 2019, pp. 1–8.
  23. Semiconductor Industry Association, “Overall roadmap technology characteristics,” accessed: 2024-01-06. [Online]. Available: https://www.semiconductors.org/wp-content/uploads/2018/08/2003Overall-Roadmap-Technology-Characteristics.pdf
Citations (1)
View on Semantic Scholar

Summary

We haven't generated a summary for this paper yet.

Ai Sparkles Streamline Icon: https://streamlinehq.com Summarize Now