EUVIMEDEUVIMED entdecken
Crossref

SNAP-FPCIM: A 28-nm Embedded-Serial-Alignment and Non-2's-Complement-MAC Floating-Point Compute-in-Memory Macro With Adjustable Computation Precision

Xing Wang, Tianhui Jiao, Yucheng Du, Miaoyu Tang, Shaochen Li, Yi Yang, An Guo, Yuchen Tang, Zhichao Liu, Yuchen Ma, Yingxuan Zhou, Weiwei Shan und 2 weitere

IEEE Transactions on Circuits and Systems I: Regular Papers · 2026-11 · 73 (11) · 7214-7227

DOI: 10.1109/tcsi.2026.3706764 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Transactions on Circuits and Systems I: Regular Papers, 2026-11. Verfasst von Xing Wang und 13 weiteren Autor:innen. Die Arbeit nennt 29 Literaturangaben. Gefördert von: National Natural Science Foundation of China, National Science and Technology Major Project, Key Research and Development Program of Jiangsu Province.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Digital compute-in-memory (DCIM), particularly floating-point CIM (FPCIM) has emerged as a promising technique to enhance energy efficiency with higher accuracy for artificial intelligence (AI) applications. However, previous works incurred substantial hardware overhead for FP multiplication and accumulation (MAC) and failed to leverage the bit-level sparsity across data formats. This work presents SNAP-FPCIM, an embedded-serial-alignment, non-2’s-complement-MAC and adjustable computation precision FPCIM based on 6T static random-access memory (SRAM). The contributions of this work include: 1) a broadcast input and embedded light-convertor structure to enable BF16/INT8 MAC operations with improved input reusability; 2) an embedded area-efficient serial-alignment scheme with dual-bit-serial MAC; 3) a format-mixed N2CMAC flow to reduce circuit dynamic activity and signed computation overhead; 4) a mode-reconfigurable design to accommodate diverse computational precision requirements. A fabricated 28nm 64Kb SNAP-FPCIM macro supports FP-MAC operations of BF16 and INT8, achieving an energy efficiency of 62.84TFLOPS/W in BF16 mode and 90.15TOPS/W in INT8 mode.

Förderung

  • National Natural Science Foundation of China
  • National Science and Technology Major Project
  • Key Research and Development Program of Jiangsu Province
  • Fundamental Research Funds for the Central Universities

Zitieren

APA

Wang, X., Jiao, T., Du, Y., Tang, M., Li, S., Yang, Y., Guo, A., Tang, Y., Liu, Z., Ma, Y., Zhou, Y., Shan, W., Yang, J., & Si, X. (2026). SNAP-FPCIM: A 28-nm Embedded-Serial-Alignment and Non-2's-Complement-MAC Floating-Point Compute-in-Memory Macro With Adjustable Computation Precision. IEEE Transactions on Circuits and Systems I: Regular Papers, 73(11), 7214-7227. https://doi.org/10.1109/tcsi.2026.3706764

Vancouver

Wang X, Jiao T, Du Y, Tang M, Li S, Yang Y, et al. SNAP-FPCIM: A 28-nm Embedded-Serial-Alignment and Non-2's-Complement-MAC Floating-Point Compute-in-Memory Macro With Adjustable Computation Precision. IEEE Transactions on Circuits and Systems I: Regular Papers. 2026;73(11):7214-7227. doi:10.1109/tcsi.2026.3706764

BibTeX
@article{Wang2026,
  author = {Wang, Xing and Jiao, Tianhui and Du, Yucheng and Tang, Miaoyu and Li, Shaochen and Yang, Yi and Guo, An and Tang, Yuchen and Liu, Zhichao and Ma, Yuchen and Zhou, Yingxuan and Shan, Weiwei and Yang, Jun and Si, Xin},
  title = {SNAP-FPCIM: A 28-nm Embedded-Serial-Alignment and Non-2's-Complement-MAC Floating-Point Compute-in-Memory Macro With Adjustable Computation Precision},
  journal = {IEEE Transactions on Circuits and Systems I: Regular Papers},
  year = {2026},
  volume = {73},
  number = {11},
  pages = {7214--7227},
  doi = {10.1109/tcsi.2026.3706764}
}
Kostenlos starten →