Embedded Files
AI Architecture for Advanced Memory Technologies
AI Architecture for Advanced Memory Technologies
Goal: Architecting AI Systems Incorporating the Next-Generation Memory
We investigate innovative AI system architectures, such as Processing-in-Memory (PIM), which integrates memory and processing units to bypass traditional data bottlenecks. Our design approach also accounts for advanced 2.5D and 3D integration packaging technologies to maximize system-level density and performance.
[DAC 2026] Yuseon Choi, Sangjin Kim, Jungjun Oh, Gwangtae Park, Byeongcheol Kim, Hoi-Jun Yoo
“SliceMoE: Bit-Sliced Expert Caching under Miss-Rate Constraints for Efficient MoE Inference,” in ACM/IEEE Design Automation Conference[ISSCC 2026] Sangwoo Ha, Jingu Lee, Youngjin Moon, Sunjoo Hwang, Wooyoung Jo, Gwangtae Park, Sangjin Kim, Soyeon Um, Junha Ryu, Yurim Jo, Hoi-Jun Yoo
“SMoLPU: 122.1μJ/Token Sparse MoE-based Speculative Decoding Language Processing Unit with Adaptive-Offload NPU-CIM Core,” in IEEE International Solid-State Circuits Conference
Page updated
Google Sites
Report abuse