技术领域

Other

290 篇相关论文 (2008–2026)

ISSCC 2026 Session 31 Other
ALPhA-Vision: A Real-Time Always-On Vision Processor with 787μs Face Detection Latency in <5mW
Ben Keller1, Rangharajan Venkatesan1, Steve Dai1, Jason Clemons2, Matthew Fojtik3, Muya Chang1, Thierry Tambe4, Nathanie
Qijing Huang1, Shalini De Mello1, Brucek Khailany2 Nvidia, Santa Clara, CA, 2Nvidia, Austin, TX, 3Nvidia, Durham, NC, 4Stanford University, Stanford, CA 1 Abstract ALPhA-Vision is an always-on low-power subsystem for DNN
ISSCC 2026 Session 31 Other
LUT-SSM: A 99.3TFLOPS/W LUT-Based State-Space Model Accelerator Using Energy-Efficient Element-Wise Layer Fusion and LUT-Friendly Weight-Only Quantization
Sunwoo Yoo*1,2, Dongyun Kam*3, Gunho Park4, Soonhyun Kwon1, Dongsoo Lee4, Youngjoo Lee1
Ulsan National Institute of Science and Technology, Ulsan, Korea, 4Naver Cloud, Seongnam, Korea *Equally Credited Authors (ECAs) 1 3 Abstract State-space models (SSMs) and weight-only quantization alleviate huge external
Korea Advanced Institute of Science and Technology, Daejeon, Korea, 2P
ISSCC 2026 Session 31 Other
VARSA: A Visual Autoregressive Generation Accelerator Using Performance-Scalable Multi-Precision PE-LUT and Grid-Similarity Attention Compression
Jiaqi Zhou, Hongou Li, Keyao Jiang, Yiyang Sun, Tianyu Jia
Abstract This paper presents VARSA, a 22nm visual autoregressive accelerator for efficient text-toimage generation, featuring: 1) a performance-scalable hybrid PE-LUT core; 2) multi-precision parallel processing with run
Peking University, Beijing, China
ISSCC 2026 Session 31 Other
A 51.6μJ/Token Subspace-Rotation-Based Dual-Quantized Large-Language-Model Accelerator with Fused Scale-Activation INT Datapath and Rearranged Bit-Slice LUT Computation
Bo Liu, Zihan Zou, Xinming Yan, Xilong Kang, Xinyang Chen, Bo Hu, Jiaming Lin, Haoran Du, Jun Yang, Xin Si, Hao Cai
Abstract A 51.6μJ/token accelerator for rotation-based dual-quantized LLMs is presented. A subspace-rotation method with parallel Hadamard transposer reduces on-chip rotation power by 62.3% and area by 59.7%. A fused sca
Southeast University, Nanjing, China
ISSCC 2026 Session 31 Other
Revolver: Low-Bit GenAI Accelerator for Distilled-Model and CoT with Phase-Aware-Quantization and Rotation-Based Integer-Scaled Group Quantization
Sangjin Kim, Jungjun Oh, Byeongcheol Kim, Yuseon Choi, Gwangtae Park, Hoi-Jun Yoo
Abstract Revolver is a low-bit GenAI accelerator that enables reasoning and multi-turn chat on edge devices under tight memory and power budgets. It introduces Phase-Aware Precision Selection (PAPS) with Multi-Precision
KAIST, Daejeon, Korea
ISSCC 2026 Session 31 Other
A 14.08-to-135.69Token/s ReRAM-on-Logic Stacked Outlier-Free Large-Language-Model Accelerator with Block-Clustered Weight-Compression and Adaptive Parallel-Speculative-Decoding
Pingcheng Dong1,2, Yonghao Tan1,2, Xuejiao Liu2, Peng Luo2, Yu Liu2, Di Pang2, Songchen Ma1,2, Xijie Huang1, Shih-Yang L
System, Hong Kong, China, 3Hefei Reliance Memory, Hefei, China, Zhejiang University, Hangzhou, China 1 4 Abstract This work presents a 55nm speculative decoding-based LLM accelerator with bumpingbased face-to-face ReRAM-
Luhong Liang2, Chi-Ying Tsui1,2, Fengbin Tu1,2, Liang Zhao4, Kwang-Tin
ISSCC 2026 Session 28 Other
Body-Interfaced Biosensors Wei Gao
California Institute of Technology, Pasadena, CA, Abstract
The rise of personalized medicine is transforming healthcare through predictive and tailored strategies. I will present our progress on wearable, implantable, and ingestible biosensors for real-time molecular analysis ac
ISSCC 2026 Session 28 Other
Importance of GaN for 5G and Solid-State mm-Wave Circuits of the Future
Umesh K. Mishra, Christopher J. Clymore, Emre Akso, Matthew Guidry
Abstract GaN has played an important role in both RF and power devices. GaN made the rapid development and deployment of 5G base stations possible with use of its Ga-polar face. Meanwhile, the N-polar face of GaN has sho
University of California, Santa Barbara, CA
ISSCC 2026 Session 22 Other
A 500kGy Radiation-Hardened 2.4GHz Wi-Fi Receiver for Innovative Nuclear Power Plant Decommissioning
Yasuto Narukiyo1, Sena Kato1, Kotaro Yanaka1, Yuya Takahashi1, Masaya Miyahara2, Jill Mayeda1, Atsushi Shirane1
Abstract This work presents a 2.4GHz Wi-Fi receiver in 65nm CMOS, designed with radiationhardened techniques to sustain operation under total ionizing dose to 500kGy. Such tolerance addresses the requirements of nuclear
Institute of Science Tokyo, Tokyo, Japan, 2High Energy Accelerator Res
ISSCC 2026 Session 22 Other
A Radiation-Hardened Self-Healing CMOS Imager with Online Pixel/Logic Annealing and Tile-Adaptive Compression for Space Applications
Quan Cheng*1,2, Zhengke Yang*1, Haoyuan Li2, Qiufeng Li1, Zhen Kong1, Gaoqiang Niu1, Yuan Liang1, Jiamin Li1, Jerald Yoo
Abstract CMOS imagers in space suffer radiation-induced faults and downlink bandwidth limits. This work presents an 180nm self-healing imager with localized online thermal annealing for both pixels and logic, as well as
Southern University of Science and Technology, Shenzhen, China, 2Kyoto
ISSCC 2026 Session 22 Other
A Multi-Qubit Cryo-CMOS SoC with Polar-Based Electron-Spin and PDM-Based Nuclear-Spin Controllers for Color Centers in Diamond
Niels Fakkel*, Luc Enthoven*, Mohamed Abdelrahman Elbadry, Hendrik Benjamin van Ommen, Margriet van Riggelen, Jiwon Yun,
nitrogen-vacancy (NV) centers in diamond enables a scalable quantum platform. This work introduces a combined Class-DE RFDAC and classD PDM driver for multi-qubit electron- and nuclear-spin control. A switch allows share
Fabio Sebastiano**, Masoud Babaie** TU Delft, Delft, The Netherlands *
ISSCC 2026 Session 22 Other
A 16-Channel Low-Power Cryo-CMOS Flux Control Pulse Generator ASIC in 14nm FinFET Technology
Kevin Tien1, David J. Frank1, John F. Bulzacchelli1, Pat Rosno2, Daniel Moertl2, John Timmerwilke1, Ari Noori1, Devin Un
Daniel Ramirez2, Timothy J. Schmerbeck2, Bryce Snell2, Jeremy Ekman2, Ryan Black2, Mark Yeck1, Tom Haselhorst2, Emma Erickson2, Kevin Demsky2, Christian W. Baks1, Jonathan Kaus2, Andrea Ruffino3, Pier Andrea Francese3, A
ISSCC 2026 Session 22 Other
A Cryo-CMOS Color-Center Quantum Controller with Diamond Waveguide Micro-Chiplet Integration
Jinchen Wang, Yuyang Han, Isaac B. Harris, Eunseok Lee, Yong Hu, Di Liu, Dirk Englund, Ruonan Han
Abstract We present a scalable cryo-CMOS controller for color-center-based quantum processors. A diamond waveguide micro-chiplet with NVs is pick-and-placed on CMOS with a 3D-printed prism for scalable photonic readout.
Massachusetts Institute of Technology, Cambridge, MA
ISSCC 2026 Session 18 Other
A 28nm 47.3TFLOPs/W 894mJ/Inference Visual Autoregressive Accelerator with Differential-Amplifier Speculation and Chain-Reaction-Like Parallel Generation
Zhiheng Yue*, Xujiang Xiang*, Jiamu Fu, Shaojun Wei, Yang Wang, Yang Hu, Shouyi Yin
*Equally Credited Authors (ECAs) Abstract To accelerate Visual Autoregressive (VAR) applications, this work implements a 28nm VAR accelerator achieving 47.3TFLOPs/W and <0.6% FID loss. A differential visual attention amp
Tsinghua University, Beijing, China
ISSCC 2026 Session 18 Other
SMoLPU: 122.1μJ/Token Sparse MoE-Based Speculative Decoding Language Processing Unit with Adaptive-Offload NPU-CIM Core
Sangwoo Ha1, Jingu Lee1, Youngjin Moon1, Sunjoo Whang1, Wooyoung Jo1, Gwangtae Park1, Sangjin Kim1, Soyeon Um2, Junha Ry
decoding LLM processor with an NPU-CIM core. It has 3 features: 1) Token-adaptive expert refinement removes redundant expert activations and schedules expert load order, achieving 2.3×/4.2× energy efficiency improvement
KAIST, Daejeon, Korea, 2Massachusetts Institute of Technology, Cambrid
ISSCC 2026 Session 18 Other
A 3.19pJ/b Electro-Optical Router with 18ns Setup Frame-Level Routing and 1-to-6 Wavelength-Flexible Link Capacity for Photonic Interposers
Yvain Thonnart1, Christian Bernard2, Stéphane Bernabé2, Myriam Assous2, Laura Boutafa2, Benoit Charbonnier2, Rémi Frania
Vincent Josselin2, Stéphane Malhouitre2, Laurent Mendizabal2, André Myko2, Damien Saint Patrice2, Rémi Vélard2, Jean Charbonnier2 CEA-List, Grenoble, France, 2CEA-Léti, Grenoble, France 1 Abstract Global interconnect on
ISSCC 2026 Session 17 Other
Maia: A Reticle-Scale AI Accelerator
S. Xu1, G. Mandal1, S. Jahagirdar1, S. Tripathy2, P. Parthasarathy1, S. Srinivasan1, A. Levin1
Microsoft, Mountain View, CA, 2Microsoft, Bengaluru, India 1 Abstract In Paper 17.4, the architecture and implementation of Microsoft’s MAIA AI silicon, a reticlescale 750W AI SoC, is presented. Innovation across power d
ISSCC 2026 Session 17 Other
ARIES and REGULUS: A Unified and Scalable Hardware-Software Co-Designed NPU SoC Family for On-Device and On-Premises Multimodal Inference
D. Shin, H. Yang, S. Jeon, J. Park, S. Han, J. Park, J. Lee, J. Kim, H. Kim, Y. Oh, M. Kim, C. Jung, W. Kim, S. Kim, H.
Y. Min, C. Song, A. Kanybek, Y. Jung, J. Song, S. Cho, H. Na, J. Park, D. Si, B. Lee, B. Park, H. Jeon Mobilint, Seoul, Korea Abstract We present a scalable NPU architecture, proven in two SoCs (ARIES and REGULUS), desig
ISSCC 2026 Session 17 Other
The STM32N6 Microcontroller: Enabling Intelligent Edge AI for IoT and Beyond
G. Desoli1, J-F. Agaësse2, N. Chawla3, E. Hilkens2, T. Boesch4, V. Taufour2, M. Ayodhyawasi3, P. Ravenhill2, S. Pal-Sing
Abstract The STM32N6 microcontroller meets the growing need for intelligent edge devices in IoT, wearables, industrial automation, and smart home systems supporting real-time, energyefficient AI processing at the edge, r
STMicroelectronics, Cornaredo, Italy, 2STMicroelectronics, Grenoble, F
ISSCC 2026 Session 17 Other
NVIDIA GB10: SoC Built for AI Acceleration
A. Skende1, G. Rosseel2, N. Pinckney3
Nvidia, Westford, MA, 2Nvidia, Santa Clara, CA, 3Nvidia, Austin, TX 1 Abstract This paper details the GB10 SoC that powers the recently launched DGX™ Spark workstation. A dual-die solution, fabricated in TSMC’s 3nm proce
ISSCC 2026 Session 14 Other
Self-Programmable Twin PUFs via Photovoltaic Energy Harvesting During the Pre-Wafer-Dicing Stage
Eunseok Lee, Jaehong Jung, Maitreyi Ashok, Anantha P. Chandrakasan, Ruonan Han
Abstract We present a CMOS self-programmable twin PUF which can be formed at the pre-waferdicing stage via photovoltaic harvesting. Two adjacent PUFs share entropy during oxide breakdown, enabling mutual authentication w
Massachusetts Institute of Technology, Cambridge, MA
ISSCC 2026 Session 14 Other
Highly-Integrated Light-Sensing System with RF Harvesting and Transmission in Commercial N-Type IGZO Flexible Technology
Marco Privitera1,2, Muhammad Zahid Naveed1, Andrea Ballo1,2, Gianluca Giustolisi1, Alfio Dario Grasso1, Massimo Alioto2
Abstract A sticker-like light sensing system is demonstrated in flexible N type-only IGZO TFT. It reuses RF signals for harvested battery charging and backscattered communications, and repurposes resistors as light senso
University of Catania, Catania, Italy, 2National University of Singapo
ISSCC 2026 Session 14 Other
A 40Gb/s 8mW-OMA 1-to-N VCSEL Driver for Parallel and Wireless Optical Links Using 150nm GaN HEMT
Shuo Sarah Feng1, Fuzhan Chen1, Ruitao Matthew Ma1,2, Hongyu Bruce Bao1, Kangping Zhong3, Pak Tao Alan Lau3, Chik Patric
Abstract This paper presents a 40Gb/s 1-to-N VCSEL driver using 150nm GaN HEMT with fT/fMAX of 50GHz/154GHz. The proposed topology can drive VCSEL arrays in series or parallel without the need of pre-driver, using a comm
Hong Kong University of Science and Technology, Hong Kong, China, 2Hig
ISSCC 2026 Session 14 Other
A Single-Chip Laser Diode Driver with Built-In Frequency-Sweep Linearization for FMCW LiDAR
Wenshuo Zhu1, Jianqiang Jiang1, Xuan Sun1, Zhenhao Li1, Tingyi Gu2, Xin Zhang3, Cheng Huang1
Abstract A laser diode driver with supply-intrinsic current shaping for FMCW LiDAR is presented. By combining a fast optical loop and an efficient switching loop, and with the LD fitted model realized in analog circuits,
Iowa State University, Ames, IA, 2University of Delaware, Newark, DE,
ISSCC 2026 Session 14 Other
An 8λ×38Gb/s/λ 106fJ/b Optical WDM Transmitter in 45nm CMOS SOI
Amirreza Shoobi, Kaisarbek Omirzakhov, Zhehao Yu, Ali Pirmoradi, Firooz Aflatouni
Abstract This paper presents an ultra-energy-efficient MRM based 8-channel optical WDM NRZ transmitter, employing a scalable driver topology combined with monolithic electronic– photonic integration, which minimizes pack
University of Pennsylvania, Philadelphia, PA
ISSCC 2026 Session 14 Other
THz-TSI: A 0.33pJ/b 264Gb/s Through-Silicon Interconnect Module for 3D Integration Utilizing Terahertz Coupling
Chen Jiang, Xiaodi Feng, Xiaohan Shen, Chixiao Chen, Qi Liu, Ming Liu, Ningsheng Xu
Abstract A through-silicon interconnect module for 3D integration utilizing THz coupling is presented, which achieves a record high data rate of 264Gb/s and efficiency of 0.33pJ/b. Both bidirectional point-to-point link
Fudan University, Shanghai, China
ISSCC 2026 Session 13 Other
A Nonintuitively Frequency-Staggered Wideband mm-Wave Low-Noise Amplifier
Vinay Chenna, Hossein Hashemi
Abstract An algorithmic topology optimization framework is presented to autonomously synthesize nonintuitive, multilayered wideband mm-Wave LNAs with arbitrary stage count. Cooptimization of actives and passives directly
University of Southern California, Los Angeles, CA
ISSCC 2026 Session 13 Other
An Inverse-Designed Passively Coupled N-Path Filter with gm-Boosted Active HBT Switches
Vinay Chenna, Hossein Hashemi
Abstract A 0.8-to-2.6GHz N-path filter with gm-boosted HBT switches and inverse-designed nonresonant passive networks is presented to enhance dynamic range and tunability of passively coupled higher-order N-path filters.
University of Southern California, Los Angeles, CA
ISSCC 2026 Session 13 Other
Medusa: A Quantum-Inspired 200-Variable 1016-Clause Analog k-SAT Solver
Luke D. Wormald, Ying-Tuan Hsu, Evangelos Dikopoulos, Wei Tang, Benjamin Datsko, Ali Hammoud, Zhengya Zhang, Michael P.
Abstract A quantum-inspired analog variable k-SAT solver supporting up to 200 variables and 1016 clauses. Enabling techniques include make/break feedback, distributed k-SAT logic, digital macro coupling, and feedback opt
University of Michigan, Ann Arbor, MI
ISSCC 2026 Session 13 Other
AI-Enabled End-to-End Design in RFICs with Controllable Architectural Style from ‘Classical’ to ‘Non-Intuitive’ for mm-Wave/sub-THz LNAs
Jonathan Zhou*1, Emir Ali Karahan*2, Juho Park1, Sherif Ghozzy1, Kaushik Sengupta1
*Equally Credited Authors (ECAs) 1 Abstract This paper introduces a unified algorithmic design flow for low-noise amplifiers (LNAs), spanning specifications to layout and integrating topology, architecture, circuit, and
Princeton University, Princeton, NJ, 2now with Marvell, Irvine, CA
ISSCC 2026 Session 13 Other
HYDAR: A 390K QPS, 1574K QPS/W Hybrid Analog/Digital Compute-in-RRAM Accelerator for Efficient Recommendation System
Jiaming Li1, Peng Yao1, Xueqi Li1, Zhenqi Hao1, Dabin Wu1, Zhouzheng Li1, Haishu Xianyu2, Lin Li3, Shujuan You4, Taiwei
China, 4China Mobile Research Institute, Beijing, China, Xiamen Industrial Technology Research Institute, Fujian, China, 6Bytedance China, Beijing, China, 7Huawei Technologies, Shenzhen, China 1 5 Abstract We present the
Qingtian Zhang1, Jianshi Tang1, He Qian1, Bin Gao1, Huaqiang Wu1 Tsing
ISSCC 2025 Session 33 Other
A 224GHz 19.9% TR Varactor-less VCO Utilizing a Multi-Section Switch-Loaded Coupled-Line Resonator
Ahmed Elmenshawi1, Sriram Muralidharan2, Mona M. Hella1
Analog Devices, Beaverton, OR 1 2 THz and sub-THz waves are key enablers for novel sensing and imaging solutions. THz gasphase spectroscopy, hyperspectral imaging, and high-speed communication are among many applications
Rensselaer Polytechnic Institute, Troy, NY
ISSCC 2025 Session 33 Other
A Wideband Bidirectional Calibration-Free Frequency/Switching-Staggering 360° D-Band Phase Shifter with Frequency-Invariant Codes Achieving <2.38°/0.63dB RMS-Errors Over 24% Bandwidth
Basem Abdelaziz Abdelmagid, Yuqi Liu, Hua Wang
With the increasing need for high data-rate and channel throughput, the D-band (110 to 170GHz) has been actively explored for beyond-5G and 6G wireless communication, sensing, and radar applications [1,2]. To overcome th
ETH Zürich, Zürich, Switzerland
ISSCC 2025 Session 33 Other
A 125-to-170GHz Power-Efficient Phase Shifter in SiGe BiCMOS with Outphasing Gain and Phase Corrections
Lorenzo Piotto, Guglielmo De Filippi, Andrea Mazzanti
Advancements in silicon technologies are opening the way to sub-THz phased-array transceivers, enabling high-resolution radar sensors, and wireless communications with a fiber-like transport capacity. Programmable phase
University of Pavia, Pavia, Italy
ISSCC 2025 Session 33 Other
A 216-to-226GHz Watt-Level GaN Solid-State Power Amplifier with Multiband Large-Signal Impedance Correction and Circuit-Package Co-Design Technique
Weibo Wang*1,2, Zhe Li*3, Kai Li3, Haifeng Cheng1,2, Fangjin Guo1,2, Yibin Zhang1,2, Keping Wang3
Tianjin, China 1 2 *Equally Credited Authors (ECAs) Compact and integrated 220GHz solid-state power amplifiers (SSPAs) are important in enabling future high-data-rate wireless communication, imaging, and radar systems. S
National Key Laboratory of Solid-State Microwave Devices and Circuits,
ISSCC 2025 Session 33 Other
A 232-to-260GHz CMOS Amplifier-Multiplier Chain with a
Low-Cost, Matching-Sheet-Assisted Radiation Package and, 11.1dBm Total Radiated Power
Jinchen Wang, Daniel Sheen, Xibi Chen, Steven F. Nagle, Ruonan Han Massachusetts Institute of Technology, Cambridge, MA Terahertz (THz) signal sources and radiators are essential for a variety of future applications, suc
ISSCC 2025 Session 25 Other
A 99.5mW/port DC-to-40GHz Integrated Channel Analyzer for High-Density Signal Integrity Measurement in 28nm CMOS
Guangdong Wu*1, Yuanliang Li*1, Bingyi Ye*1,2, Fangzhu Li1, Xin Liu1, Haowei Niu1,
Beijing, China 1 2 *Equally Credited Authors (ECAs) The growing demand for higher network bandwidth has led to a significant rise in channel density within Ethernet switches and high-performance computers. As data rates
Ruixu Wang1, Dunshan Yu1, Weixin Gai1,3 Peking University, Beijing, Ch
ISSCC 2025 Session 25 Other
A Micromachined Heterogeneously Integrated Active-Probe Enabling Non-Disruptive In-Situ Measurements from DC to 50GHz
Jesse Moody, Tyler Liebsch, Patrick Finnegan, Stefan Lepkowski,
integrated microsystems becomes increasingly difficult as these systems scale in size and speed. These interfaces are tight pitch and parasitic sensitive, limiting the use of traditional (50Ω) test equipment. A microscal
Robert Costanzo, Mieko Hirabayashi, Matt Jordan, Travis Forbes, Christ
ISSCC 2025 Session 25 Other
AI-Enabled Design Space Discovery and End-to-end Synthesis for RFICs with Reinforcement Learning and Inverse Methods Demonstrating mm-Wave/sub-THz PAs between 30 and 120GHz
Jonathan Zhou*1, Emir Ali Karahan*1, Sherif Ghozzy1, Zheng Liu1,2, Hossein Jalili1, Kaushik Sengupta1
AI-enabled algorithmic flow for architecture discovery, circuit topology and parameter optimization for RFICs, particularly exploring design spaces beyond human intuition. RF and mmWave IC design is a complex iterative d
Princeton University, Princeton, NJ Texas Instruments, Dallas, TX 1 2
ISSCC 2025 Session 25 Other
A 4GS/s Fully Analog 256×256 MP-Based Cross-Correlator with 1000TOPS/W Compute Efficiency and 1.3TOPS/mm2 Compute Density in 22nm SOI CMOS
Aswin Undavalli1, Kareem Rashed2, Gert Cauwenberghs3, Shantanu Chakrabartty1,
Louis, Saint Louis, MO Oregon State University, Corvallis, OR 3 University of California, San Diego, CA 4 Northeastern University, Oakland, CA 1 2 Multi-lag cross-correlations (X-Corr) are essential building blocks in ra
Arun Natarajan2, Aravind Nagulu1,4 Washington University in St
ISSCC 2025 Session 25 Other
A Physics-Inspired Oscillator-Based Mixed-Signal Optimization Engine for Solving 50-Variable 218-Clause 3-SAT Problems with 100% Solvability and 31.7μs Solution Time
Evangelos Dikopoulos, Ying-Tuan Hsu*, Luke Wormald*, Wei Tang,
Zhengya Zhang, Michael P. Flynn University of Michigan, Ann Arbor, MI *Equally Credited Authors (ECAs) The Boolean satisfiability (SAT) problem is a fundamental NP-complete problem, and efficiently solving it would revol
ISSCC 2025 Session 23 Other
Slim-Llama: A 4.69mW Large-Language-Model Processor with Binary/Ternary Weights for Billion-Parameter Llama Model
Sangyeob Kim, Jungwan Lee, Hoi-Jun Yoo
Recently, multiple ASICs [1-6] have been proposed to accelerate large language models (LLMs). However, the enormous number of LLM parameters leads to significant energy consumption due to external memory access (EMA). Wh
KAIST, Daejeon, Korea
ISSCC 2025 Session 23 Other
An 88.36TOPS/W Bit-Level-Weight-Compressed Large-Language-Model Accelerator with Cluster-Aligned INT-FP-GEMM and Bi-Dimensional Workflow Reformulation
Yubin Qin, Yang Wang, Jiachen Wang, Zhiwei Lin, Yushu Zhao, Shaojun Wei,
range of natural language processing (NLP) tasks, becoming an essential part of modern society [1-4]. This exceptional performance can be attributed to huge model size and autoregressive computation [5,6]. However, these
Yang Hu, Shouyi Yin Tsinghua University, Beijing, China Large language
ISSCC 2025 Session 23 Other
BROCA: A 52.4-to-559.2mW Mobile Social Agent System-on-Chip with Adaptive Bit-Truncate Unit and Acoustic-Cluster Bit Grouping computation. The ACE PE processes input data in a bit-serial manner, where four 8b weights are accumulated in parallel, leading to a decrease in compute energy proportional to the input bitwidth. As a result, an average reduction of 4.4b in input bitwidth and a 4.7× speedup can be achieved, reducing computation energy by 44.5% at the RG.
Wooyoung Jo, Seongyon Hong, Jiwon Choi, Beomseok Kwon, Haoyang Sang,
Dongseok Im, Sangyeob Kim, Sangjin Kim, Taekwon Lee, Hoi-Jun Yoo Figure 23.7.4 illustrates the proposed ACBU, which achieves runtime bitwidth reduction of the vocoder input feature map to reduce computation energy in the
ISSCC 2025 Session 23 Other
MEGA.mini: A Universal Generative AI Processor with a New Big/Little Core Architecture for NPU
Donghyeon Han1,2, Anantha P. Chandrakasan1
Chung-Ang University, Seoul, Korea 1 2 The global AI market is growing explosively with the rise of generative AI applications, such as image manipulation and text-to-text/image/video creation. AI was primarily expected
Massachusetts Institute of Technology, Cambridge, MA
ISSCC 2025 Session 23 Other
MAE: A 3nm 0.168mm2 576MAC Mini AutoEncoder with Line-based Depth-First Scheduling for Generative AI in Vision on Edge Devices
Shih-Wei Hsieh, Chia-Hung Yuan, Ming-Hung Lin, Ping-Yuan Tsai, You-Yu Nian,
Chia-Yuan Cheng, Hung-Wei Chih, Po-Han Chiang, Ming-Hsuan Chiang, Yuan-Jung Kuo, Yu-Wei Wu, Yi-Syuan Chen, Po-Heng Chen, Sandy Huang, Ming-En Shih, Chia-Ping Chen, Abrams Chen, ShenKai Chang, Chih-Ming Wang, Po-Yu Yeh, J
ISSCC 2025 Session 23 Other
Nebula: A 28nm 109.8TOPS/W 3D PNN Accelerator Featuring
Adaptive Partition, Multi-Skipping, and Block-Wise Aggregation
Changchun Zhou1, Tianling Huang1, Yanzhe Ma1, Yuzhe Fu1, Xiangjie Song1, Siyuan Qiu1, Jiacong Sun1, Min Liu1, Ge Li1, Yifan He2, Yuchao Yang1,3, Hailong Jiao1 Peking University, Shenzhen, China Reconova Technologies, Xia
ISSCC 2025 Session 23 Other
EdgeDiff: 418.4mJ/Inference Multi-Modal Few-Step Diffusion Model Accelerator with Mixed-Precision and Reordered Group Quantization
Sangjin Kim, Jungjun Oh, Jeonggyu So, Yuseon Choi, Sangyeob Kim,
need for high-performing image-generative models, including the diffusion model (DM) [2, 3]. A conventional DM requires numerous UNet-based denoising timesteps (~50), leading to high computation and external memory acces
Dongseok Im, Gwangtae Park, Hoi-Jun Yoo KAIST, Daejeon, Korea The incr
ISSCC 2025 Session 23 Other
HuMoniX: A 57.3fps 12.8TFLOPS/W Text-to-Motion Processor with Inter-Iteration Output Sparsity and Inter-Frame Joint Similarity
Jaehoon Heo, Adiwena Putra, Sungwoong Yune, Jieon Yoon, Hangyeol Lee,
media applications, such as film production and AR/VR. This process involves creating human joint movements and constructing detailed 3D meshes, like human skin, for each joint (see Fig. 23.10.1). It used to require hour
Jihoon Kim, Joo-Young Kim KAIST, Daejeon, Korea Recently, 3D human mot
ISSCC 2025 Session 13 Other
A Via-Programmable DNN-Processor Fabrication Toward 1/40th Mask Cost
Jaewon Shin, Rei Sumikawa, Dongzhu Li, Mototsugu Hamada, Atsutake Kosuge
Growing interest in healthcare has led to the development of many wearable battery-powered artificial-intelligence internet-of-things (AI-IoT) devices for continuous monitoring a wide variety of vital signs [1, 2] (Fig.
University of Tokyo, Tokyo, Japan