Thermal Sensor Placement based on Meta-Model Enhancing Observability and Controllability

Abstract

On-chip thermal sensor (TS) is essential to prevent a chip from burn out and to increase its reliability. In addition, accuracy of temperature reading can affect to the performance and the power consumption of chip. We propose methodology for optimal TS allocation based on meta-model. Observability and controllability concept are introduced to place the TS nearby hot spot and controllable position. Initially, temperature profile on chip for various operating scenarios is simulated by CFD (Computation Fluid Dynamics) simulation to make thermal resistance matrix. Then thermal resistance matrix is reconstructed based on TS location. Finally, optimal location for TS is recommended depending meta-model reflecting observability and controllability in one space. To validate temperature when dynamic thermal management is working, we developed in-house code working with ROM (Reduced Order Model). Investigation is carried out to evaluate thermal violation for different placement groups of TS. Simulation result shows that the best TS placement control hot spot temperature reasonably, but the worst TS placement make hot spot temperature and power increase dramatically due to increased leakage power. Through meta-model approach, designers can determine the position of the thermal sensor objectively and consistently by mathematical methods.