Title :
Cooperative On-Chip Temperature EstimationUsing Multiple Virtual Sensors
Author :
Jun Yong Shin ; Kurdahi, Fadi ; Dutt, Nikil
Author_Institution :
Center for Embedded & Cyber-Phys. Syst., Univ. of California, Irvine, Irvine, CA, USA
Abstract :
A variety of dynamic thermal management (DTM) schemes have been proposed to address the adverse effects of high temperatures on a chip. These DTM schemes rely on on-chip thermal sensors to get accurate temperature information over a die, and typically assume on-chip thermal sensors give accurate temperature readings. However, on-chip thermal sensors with small footprint and low power consumption, such as popular ring-oscillator (RO) based sensors, have relatively low accuracy. We address the challenging question of how to acquire accurate temperature information of a chip during its runtime using small, low-power on-chip thermal sensors. We propose a novel approach of using multiple virtual thermal sensors to increase the accuracy of temperature readings; the virtual thermal sensors are generated from a small low-power physical thermal sensor by adaptively switching its calibration points on the run. Simulation results show that the RMS error of temperature readings can be reduced by up to 91.1% with the use of four virtual thermal sensors as compared with a single thermal sensor case.
Keywords :
oscillators; sensor fusion; temperature sensors; thermal management (packaging); virtual instrumentation; cooperative on-chip temperature estimation; dynamic thermal management; low power consumption; multiple virtual sensors; on-chip thermal sensors; ring-oscillator based sensors; small footprint; Accuracy; Calibration; Estimation; System-on-chip; Temperature sensors; Calibration; dynamic thermal management (DTM); temperature estimation; thermal sensors; virtual sensors;
Journal_Title :
Embedded Systems Letters, IEEE
DOI :
10.1109/LES.2015.2400992