Dial: An Energy-Efficient Dram in-Memory Computing Accelerator with Compact Partial Product Lut and Twisted Differential Adc
This paper presents DIAL, an energy-efficient DRAM inmemory computing accelerator with partial product LUTbased architecture for advanced tasks. The compact LUT reduces 53% of the LUT area and 45% of power by eliminating extended zeros. The twisted differential ADC reduces ADC area by 37 % and power by 45 % through comparators sharing and improves CSNR by using a 2.3fF computation capacitor and increasing ADC resolution by <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$3 \times$</tex>. The dual-mode sense amplifier reduces 33% of LUT power by adjusting operation modes based on data patterns. Fabricated in 28 nm CMOS technology, DIAL occupies a <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$20.25 \text{mm}^{2}$</tex> die area and achieves 55.4 TFLOPS/W energy efficiency on the GPT-2 benchmark, demonstrating a <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$4.1 \times$</tex> higher efficiency than prior IMC designs.
Authors
- Junha Ryu (ORCID: https://orcid.org/0000-0001-8147-3085)
- Kyomin Sohn (ORCID: https://orcid.org/0000-0002-8094-9843)
- Soyeon Um (ORCID: https://orcid.org/0000-0002-8526-2047)
- Hoi‐Jun Yoo (ORCID: https://orcid.org/0000-0002-6661-4879)
- Sangwoo Ha (ORCID: https://orcid.org/0000-0002-9191-6326)
- Byeongcheol Kim (ORCID: https://orcid.org/0000-0003-1079-7220)
- Min‐Sung Kim (ORCID: https://orcid.org/0000-0002-2369-9956)
- Sang-Jin Kim (ORCID: https://orcid.org/0000-0001-9317-0479)
- Chang-Jin Jeong
- Sunjoo Whang
Institutions
- Korea Advanced Institute of Science and Technology (KR)
- Samsung (South Korea) (KR)
Publication Details
- Published
- 2025-06-08
- DOI
- https://doi.org/10.23919/vlsitechnologyandcir65189.2025.11075215
- Citations
- 1
- Primary Topic
- Parallel Computing and Optimization Techniques
- Type
- article
- Field-Weighted Citation Impact
- 2.27
Funders
- Institute for Information and Communications Technology Promotion