Session

· Contributors · Organizations · Search Program · Flagged · Happening Now

Research Manuscript

: Breaking Barriers: Compute-in-Memory for Transformer Acceleration

Session Chairs

DescriptionThis session highlights groundbreaking innovations in Compute-in-Memory (CIM) architectures designed to accelerate large-scale transformer models and alleviate data transfer bottlenecks. It covers various optimization strategies, from CIM-based solutions for attention computation in large language models (LLMs) to efficient memory management for long-context inference. The session delves into software-hardware co-design techniques, including irregular attention sparsity, outlier-aware quantization, Mixture-of-Experts (MoE) approaches, 3D hybrid bonding, and novel asynchronous execution methods.

Event Type

Research Manuscript

TimeWednesday, June 253:30pm - 5:30pm PDT

Location3001, Level 3

Topics

Tracks

Similar Sessions

Storage Meets Computing Power for Advancing AI and Data Processing Efficiency

Spike It, See It, Say It: Next-Gen AI Processing

Unleashing the Power of Accelerators: ASICs, FPGAs, and PIMs

Presentations

3:30pm - 3:45pm PDT	DIAS: Distance-based Attention Sparsity for Ultra-Long-Sequence Transformer with Tree-like Processing-in-Memory Architecture Authors Zekai Chen Yiming Chen Teng Wan Tianyi Yu Yu Wang Huazhong Yang Xueqing Li	DES2B: In-memory and Near-memory Computing Architectures, Applications and Systems
3:45pm - 4:00pm PDT	AttenPIM: Accelerating LLM Attention with Dual-mode GEMV in Processing-in-Memory Authors Liyan Chen Dongxu Lyu Zhenyu Li Jianfei Jiang Qin Wang Zhigang Mao Naifeng Jing	DES2B: In-memory and Near-memory Computing Architectures, Applications and Systems
4:00pm - 4:15pm PDT	SplitSync: Bank Group-Level Split-Synchronization for High-Performance DRAM PIM Authors Yoon Byungkuk Sanghyeok Han Gyeonghwan Park Jae-Joon Kim	DES2B: In-memory and Near-memory Computing Architectures, Applications and Systems
4:15pm - 4:30pm PDT	PIMoE: Towards Efficient MoE Transformer Deployment on NPU-PIM System through Throttle-Aware Task Offloading Authors Lizhou Wu Haozhe Zhu Siqi He Xuanda Lin Xiaoyang Zeng Chixiao Chen	DES2B: In-memory and Near-memory Computing Architectures, Applications and Systems
4:30pm - 4:45pm PDT	Supporting Register-based Addressing Modes for in-DRAM PIM ISAs Authors Seok Young Kim Byung Ho Choi Seokwon Kang Yongjun Park Seon Kim	DES2B: In-memory and Near-memory Computing Architectures, Applications and Systems
4:45pm - 5:00pm PDT	OutlierCIM: Outlier-Aware Digital CIM-Based LLM Accelerator with Hybrid-Strategy Quantization and Unified FP-INT Computation Authors Zihan Zou Shikuang Chen Chen Zhang Xing Wang Zhichao Liu Haoran Du Xin Si Hao Cai Bo Liu	DES2B: In-memory and Near-memory Computing Architectures, Applications and Systems
5:00pm - 5:15pm PDT	Near-Memory LLM Inference Processor based on 3D DRAM-to-logic Hybrid Bonding Authors Sanghyeok Han Yoon Byungkuk Gyeonghwan Park Choungki Song Dongkyun Kim Jae-Joon Kim	DES2B: In-memory and Near-memory Computing Architectures, Applications and Systems
5:15pm - 5:30pm PDT	SeIM: In-Memory Acceleration for Approximate Nearest Neighbor Search Authors Chaoqiang Liu Dan Chen Yu Huang Wenjing Xiao Haifeng Liu Yi Zhang Huize Li Xiaofei Liao Hai Jin	DES2B: In-memory and Near-memory Computing Architectures, Applications and Systems

62dac.conference-program.com