Steady-State Thermal Modeling for Embedded Applications
Sarah Azaizeh, Olivia Marsh, Daniel Whitman, Robert A. Taylor, Shi Sha · 2023
The ever-increasing application complexity drives a higher level of hardware integration in the embedded and mobile systems' design, which causes soaring onboard power density and temperature. To this end, a series of numerical and analytical system-level power and thermal modeling methodologies have been developed for thermal analysis on different system scales and architectures. In this work, we studied the steady-state temperature modeling problem for embedded system-on-chip architectures. First, we studied the commonly used RC-lumped thermal models for single-core and multi-core platforms. Then, we developed a practical power and temperature measurement testbed using Xilinx ZYNQ Z7 system-on-chip platforms. Next, we extracted the thermal prediction parameters using different combinations of synthetic benchmarks, system speed levels, and system utilization. Lastly, we verified the temperature prediction accuracy by comparing it with the onboard sensor readings using convolutional neural network applications. The experimental results showed that our steady-state temperature modeling is highly effective with 0.62°C prediction deviation and 2.79% error on average.