AI Computing Platform Laboratory

As AI models continue to grow in size, they require vast amounts of energy, making sustainable AI unfeasible if current trends persist. Consequently, the importance of robust computing HW/SW infrastructure underpinning AI will become critical.

Our main research goal is to compute AI models in a faster and energy-efficient way through HW/SW co-design. Specifically, our research interests include:

• Neural Processing Unit (NPU), domain-specific hardware, FPGA
• Quantization, pruning, and knowledge distillation
• Hardware-aware neural architecture search (HW-Aware NAS) and neural architecture accelerator search (NAAS)
• Processing-in-memory (PIM)
• Efficient LLM serving including KV Caching and other optimizations
• On-Device AI