Methodology for Accurate Hotspot Prediction by Upscale Spatiotemporal Resolution of Temperature Sensing using a Power Meter and Resistor-Capacitor Thermal Model

• Published in: 2021 20th IEEE Intersociety Conference on Thermal and Thermomechanical Phenomena in Electronic Systems (iTherm) pp. 130 - 138
• Main Authors: An, Taekeun, Im, Yunhyeok, Yoo, Jongkyu, Cho, Youngsang, Lee, Heeseok, Heo, Yun
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01-06-2021
• Subjects: hot spot; Meters; mobile; Performance evaluation; Predictive models; spatiotemporal resolution; Sytem-on-chip; Temperature distribution; Temperature measurement; Temperature sensor; Temperature sensors; thermal model; Thermal resistance
• ISSN: 2694-2135
• DOI: 10.1109/ITherm51669.2021.9503200

Accurate temperature monitoring for the entire area of a die with fine time resolution is crucial for maximizing device performance. For mobile devices in particular, it is necessary to control die temperature within operation boundaries to minimize leakage power and operating voltage to maximize benchmark performance within a limited power budget. However, it is difficult to measure the temperature of the entire area of a SOC die with fine time resolution by using conventional temperature sensors because the number of temperature sensors is limited and time interval for sensing with an analog circuit is in the range of several milliseconds. We propose a methodology for predicting the temperature of the entire chip area with high time resolution by upscaling the spatial and temporal resolution of the temperature distribution on the die using a power meter and thermal RC model. Superposition of the temperature response from multiple heat sources is used to obtain a temperature gradient. Interpolation is then used to obtain temperature with high resolution in the time interval by using a thermal RC model and monitored power, which can be obtained for every few nanoseconds. By using a CFD simulation, we validate the proposed method.