Congratulations to Zhantong (senior-year undergrad). We demonstrate CIM has great efficiency benefits for the generative model inference in TPU architecture.
Congratulations to Yanchi. A power modeling and analysis framework for LLM chiplets and a SoC thermal optimization solution (collaborated with Prof. Yibo Lin) will be presented.
Congratulations to Yiqi and Mingxuan (senior-year undergrad). A CIM-based chiplet design for Diffusion and sparsity-aware training accelerator are introduced.
A MHA ViT accelerator paper will also be presented (collaborated with Prof. Yufei Ma).
One paper introduces a flexible systolic CIM macro (collaborated with Prof. Le Ye), and the other paper shows a heterogeneous CIM-based processor (collaborated with Prof. Yufei Ma).
One 12nm agile-designed SoC is designed with decentralized power management (work done at Harvard), and a 22nm 0.26nW/synapse SNN accelerator will be presented (collaborated with Prof. Le Ye).
