T3A: Design and managements of multiprocessor systems-on-chips

Summary form only given. Continuous advances in VLSI technology enable implementation of complex systems ranging from single chip cloud computers to application specific heterogeneous multiprocessor systems-on-chip. Regardless of the application domain, design and implementation of these systems require multidimensional problem solving. Traditional metrics such as thermal design power, processor speed and chip area are not adequate to describe or differentiate modern computing systems. Instead, energy efficiency and performance under specific use cases, or key performance indicators, constitute an array of metrics. Similarly, dynamic power management techniques ensure energy efficiency under workload and parameter variations. This talk will present novel design and management methodologies for VLSI systems. This talk will present a unified methodology for the design and management of state of the art VLSI systems using examples from both high performance CMP and MpSoC domains. More specifically, the first part will focus on a multidimensional optimization methodology which relies on novel analytical power, performance and area models. The second part of the talk will cover variation-adaptive dynamic power management techniques based on feedback control theory.