Abstract

As AI models continue to grow in scale and complexity, communication and energy are becoming the dominant bottlenecks in system performance. Chiplet-based architectures present a promising path forward by enabling modular scalability beyond monolithic integration. However, they also introduce heterogeneity in interconnects and non-uniform communication costs, complicating system design. This talk examines the architectural and mapping challenges posed by these trends and highlights the importance of communication-aware dataflow mapping.

Bio.

Ting-Jung Chang is an Assistant Professor in the Department of Computer Science at National Yang Ming Chiao Tung University. She holds a Ph.D. in ECE and an M.A. in EE from Princeton University. Prior to joining the university, she worked at SambaNova Systems on next-generation AI hardware. Her research interests include computer architecture and VLSI design, with a particular focus on heterogeneous systems and emerging technologies.

All faculty and students are welcome to join.