Microprocessor non-uniform sampling heat distribution reconstruction method based on convolutional neural network

Abstract

The invention relates to a microprocessor non-uniform sampling heat distribution reconstruction method based on a convolutional neural network. The method comprises the following steps: firstly, manufacturing a sample temperature data set by using an oil cooling heat dissipation system capable of transmitting infrared spectrum; secondly, completing model training in a chip design stage or a silicon post-stage, and storing the trained network; thirdly, in the running stage of the processor, judging the category of the workload application program by using a classification network according to the temperature data sampled by the thermal sensor; and finally, obtaining the overall temperature distribution of the chip by using the corresponding heat distribution reconstruction network. Due to the fact that the workload classification network model and the heat distribution reconstruction network model are designed respectively, heat distribution reconstruction is carried out through the non-uniform temperature data sampled by the limited number of heat sensors, and the overall temperature distribution data of the chip obtained through recovery are more accurate.

Description

technical field [0001] The invention belongs to the technical field of chip temperature monitoring, and in particular relates to a method for reconstructing non-uniform sampling heat distribution of a microprocessor based on a convolutional neural network. First, use the oil-cooled heat dissipation system that can transmit the infrared spectrum to complete the production of the sample temperature data set; second, complete the model training in the chip design or post-silicon stage, and store the trained network; third, in the processor running stage, according to The thermal sensor samples the temperature data, and uses the classification network to judge the category of the workload application; finally, uses the corresponding thermal distribution reconstruction network to obtain the overall temperature distribution of the chip. Background technique [0002] In recent years, high-performance multi-core processors generally integrate on-chip thermal sensors to implement con...

AI Technical Summary

Problems solved by technology

However, when the number of thermal sensors is limited, there are obvious errors in the hotspot temperature reconstructed by this method, and accurate temperature sensing cannot be achieved.

Images