Abstract

Heterogeneous computing is widely applied for performance/efficiency optimization, ranging from high-performance computing like data centers and supercomputers, to mobile and energy-efficient computing. In the era of post-Dennard scaling, processing chips face the new challenges to maintain safe temperature range and allowable power budget due to the exponentially increasing power density. In this talk, I will discuss some recent advances in system-level design for heterogeneous many-core systems, which features the optimization of chip reliability, system energy efficiency and application performance simultaneously through hardware-software co-design approaches. Extensive evaluations on the PARSEC, DSP-stone benchmarks and realistic applications demonstrate significant improvement over the state of the art on the three design dimensions. Moreover, temperature monitoring of individual cores is an essential technique for thermal-aware chip management. If time allows, I will introduce a brand-new thermal sensing technique for many-core systems based on optical network-on-chip. This interdisciplinary research takes advantages of the thermo-optic effect of optical routers, and derives thermal information through lightweight software computation based on inter-core data communications. Physical simulations validate the high accuracy of the thermal sensitivity model, and large sets of experiments with randomly generated applications show the feasibility of the approach.

Speaker's Biography

Dr. Weichen Liu is currently a professor at College of Computer Science, Chongqing University, China. He received the PhD degree from the Hong Kong University of Science and Technology, Hong Kong, and the MEng and BEng degrees from Harbin Institute of Technology, China. Dr. Liu served as the chairs, technical program committee members, editors and reviewers for over 20 premier international conferences and journals. He authored and co-authored more than 60 research papers in peer-reviewed journals, conferences and books, and received the best paper candidate awards from ASP-DAC 2016, CASES 2015, CODES+ISSS 2009, the best poster award from AMD-TFE 2010, and the most popular poster award from ASP-DAC 2017. His research interests include embedded and real-time systems, multiprocessor systems, reconfigurable systems, reliable and fault-tolerant systems, network-on-chip, hardware/software co-design, mobile computing, and formal methods.