On-chip power density of electronic systems are increasing every year and extraction/management of the generated heat is essential to continue performance scaling. The challenge for thermal management in high-performance many-core processor is to manage the spatiotemporally varying thermal field. Increasing performance demand and small volume is making thermal design of mobile systems highly challenging as well; high-power RF modules makes thermal control more intriguing. GREEN lab explored methods to understand, characterize, and address the thermal challenges in next generation electronics.
Thermal System Identification (TSI) for Post-silicon, On-line, and Real-time Thermal Analysis: We are developing, Thermal System Identification (TSI) – a unique signal processing and control theory based method and circuit for on-chip characterization of thermal system and on-line temperature predictions for individual ICs after fabrication and packaging. The mathematical models are developed and verified by design/measurement of CMOS test-chips. Our approach helps capture the effect of chip-to-chip variation in thermal properties and leakage-temperature interaction—a key deficiency of existing design-time thermal simulators.
On-demand Integrated Active Cooling: We are exploring integrated thermoelectric cooling for enhanced management of thermal hotspots in microprocessor and heterogeneous systems. The on-chip circuits for on-line control of TECs are demonstrated; we are exploring control circuits that will support an energy autonomous TEC operation where energy required for TE cooling is harvested by the TE module itself. Thermal modeling as well as design/tape-out/measurement of test-chip/test-board are being pursued to develop integrated on-demand cooling solutions.
Global Thermal Management of Many-Core Processors: We are developing a coherent modeling environment for to understand the interaction between computation power, architecture and execution model, cooling methods, and the generated thermal field. We are developing engineering solutions for energy and thermal management using distributed sensor-driven real-time feedback to efficiently manage system power.