Dynamic On-Chip Thermal Sensor Calibration Using Performance Counters

Numerous sensors are currently deployed in modern processors to collect thermal information for fine-grained dynamic thermal management (DTM). Due to process variation and silicon aging, on-chip thermal sensors require periodic calibration before use in DTM. However, the calibration cost for thermal sensors can be prohibitively high as the number of on-chip sensors increases. In this paper, a model which is suitable for online calculation is employed to estimate the temperatures of multiple sensor locations on the silicon die. The estimated sensor and actual sensor thermal profile show a very high similarity with correlation coefficient
${\sim}{\rm 0.9}$
for most tested benchmarks. Our calibration approach combines potentially inaccurate temperature values obtained from two sources: temperature readings from thermal sensors and temperature estimations using system performance counters. A data fusion strategy based on Bayesian inference, which combines information from these two sources, is demonstrated along with a temperature estimation approach using performance counters. The average absolute error of the corrected sensor temperature readings is
${<}{1.5}^{\circ}{\rm C}$
and the standard deviation of error is less than
${<}{\rm 0.5}^{\circ}{\rm C}$
for tested benchmarks.