Novel accelerator-model co-exploration
Exploration of traditional and emerging compute systems for scalable AI workloads
This research thrust focuses on exploring the next generation of AI system development, including exploring sub-quadratic model architecture and their optimized kernel development, compute-in-memory (CiM) hardware archiecture, noise-aware self-healing algoritthm-architecture co-design. Some of the research outcome of this project includes: CiMNet, LLM performance evaluation framework, Noise-aware adaptive learning.
We explore architecture-algorithm co-design for scalable and self-healing AI inference solutions.