On-Line Temperature Estimation for Noisy Thermal Sensors Using a Smoothing Filter-Based Kalman Predictor

Area of Science:

• Computer Engineering
• Electrical Engineering
• Materials Science

Background:

• Dynamic thermal management (DTM) relies on embedded thermal sensors for real-time monitoring of multi-core processors.
• Embedded thermal sensors are prone to noise from environmental factors and process variations, leading to inaccurate temperature readings.
• Inaccurate temperature data compromises the efficiency of DTM strategies, affecting processor performance and longevity.

Purpose of the Study:

• To develop an accurate temperature estimation technique for noisy sensor readings in multi-core processors.
• To enhance the efficiency of Dynamic Thermal Management (DTM) by improving the reliability of thermal sensor data.
• To propose a novel multi-sensor synergistic calibration algorithm (MSSCA) for improved simultaneous prediction accuracy.

Main Methods:

• A smoothing filter-based Kalman prediction technique is employed to estimate temperatures from noisy sensor data.
• Spatial correlations among different sensor locations are leveraged for multi-sensor estimation.
• A multi-sensor synergistic calibration algorithm (MSSCA) is developed to enhance simultaneous prediction accuracy.
• Infrared imaging is used to capture real-time thermal traces of an Advanced Micro Devices (AMD) quad-core processor for validation.

Main Results:

• The proposed synergistic calibration scheme reduced the root-mean-square error (RMSE) by 1.2 °C.
• Signal-to-noise ratio (SNR) was increased by 15.8 dB with minimal runtime overhead.
• The average false alarm rate (FAR) of corrected sensor readings decreased by 28.6%.
• The approach demonstrated improved accuracy in estimating processor temperatures from noisy sensor inputs.

Conclusions:

• The developed Kalman prediction and MSSCA significantly improve the accuracy of temperature estimation from noisy thermal sensors.
• Accurate temperature estimation enables more appropriate and timely adjustments of processor voltages, frequencies, and cooling fan speeds.
• The proposed methods enhance the overall efficiency and reliability of Dynamic Thermal Management (DTM) in multi-core processors.