Computing in Memory

  • No Comments

SMoLPU: An Energy-Efficient Sparse Mixture-of-Experts-Based Speculative Decoding Language Processing Unit

SMoLPU: An Energy-Efficient Sparse Mixture-of-Experts-Based Speculative Decoding Language Processing Unit 150 150
Author(s): Sangwoo Ha, Jingu Lee, Youngjin Moon, Sunjoo Whang, Wooyoung Jo, Gwangtae Park, Sangjin Kim, Soyeon Um, Junha Ryu, Yurim Jo, Hoi-Jun Yoo

Abstract:

Reasoning strategies such as chain-of-thought improve the accuracy of large language models (LLMs) but lengthen decoding sequences, inflating external memory access (EMA) on mobile devices. Numerous optimizations have been explored to reduce EMA, from conventional techniques such as group quantization and channel-wise pruning to sparse mixture-of-experts (MoE) and speculative decoding (…

Published in: IEEE Journal of Solid-State Circuits
Page(s): 1 – 13
Year of Publication: 2026
Electronic ISSN: 1558-173X
DOI: 10.1109/JSSC.2026.3726852
Publisher: IEEE

View on IEEE Xplore

  • No Comments

A 194.6-TOPS/W Pipelined All Current-Domain Mixed-Signal Compute in Memory in 28-nm CMOS

A 194.6-TOPS/W Pipelined All Current-Domain Mixed-Signal Compute in Memory in 28-nm CMOS 150 150
Author(s): Saideep Cherukuri, N. V. Kidambi, W. C. Wang, I. Chakraborty, S. Zhang, L. Shamieh, J. B. Steinberg, S. Mukhopadhyay

Abstract:

Mixed-signal CIM (MS-CIM) faces bit-cell nonlinearity, poor linearity at high frequency, and throughput limits. We present a hybrid pipelined current-domain MS-CIM macro featuring bit-cell matched linearization interface (BMLI) and loop-unrolled successive approximation refinement (SAR) ADC fabricated in 28-nm CMOS. A $256{\,}\times {\,}256$ SRAM array with 8-bit inputs, 8-bit weights achieve 10.16-TOPS …

Published in: IEEE Solid-State Circuits Letters
Page(s): 174 – 177
Year of Publication: 2026
Electronic ISSN: 2573-9603
DOI: 10.1109/LSSC.2026.3697990
Publisher: IEEE

View on IEEE Xplore

Search this website