In this paper, we address the challenge of accurate full-chip power mapping for commercial off-the-shelf CPU and GPU processors, explicitly considering temperature dependence. It is well known that both dynamic and leakage power are strongly temperature-dependent; however, existing power estimation methods for real chips often neglect this critical factor. To mitigate this, we characterize temperature-dependent spatial power maps for the first time on commercial processors, including the AMD Radeon RX 6400 GPU and Qualcomm Snapdragon 680 (SM6225) CPU. Using a back-side cooling infrared (IR) thermal imaging system, we capture full-chip thermal maps under different cooling conditions while running identical workloads. These thermal maps are converted into power maps using first-principles-based methods. By repeating this process across varying cooling environments, we collect power maps corresponding to different average chip temperatures. Our experimental results confirm that both total power and spatial power distributions vary significantly with cooling conditions, even under the same workload. We then train machine learning models using real-time performance and utilization metrics—collected via AMD Adrenalin Edition and Qualcomm Snapdragon Profiler—to capture these thermal effects. Two deep neural network architectures are explored: a transformer-based model, ChipPowerMap, and a CNN-based decoder model. We compare their performance in accurately predicting temperature-aware full-chip power maps. Numerical results highlight the effectiveness of ChipPowerMap in achieving highly accurate thermal map predictions, boasting an RMSE of only 67.88mW/mm2or 0.97% of the full-scale error. It also outperforms the CNN-based method by 1.62x in terms of accuracy on average. Besides, the proposed model offers real-time estimation with a rapid speed of 25ms on the target chip.

In this paper, we propose a novel approach for the real-time estimation of chip-level spatial power maps for commercial Google Coral M.2 TPU chips based on a machine-learning technique for the first time. The new method can enable the development of more robust runtime power and thermal control schemes to take advantage of spatial power information such as hot spots that are otherwise not available. Different from the existing commercial multi-core processors in which real-time performance-related utilization information is available, the TPU from Google does not have such information. To mitigate this problem, we propose to use features that are related to the workloads of running different deep neural networks (DNN) such as the hyperparameters of DNN and TPU resource information generated by the TPU compiler. The new approach involves the offline acquisition of accurate spatial and temporal temperature maps captured from an external infrared thermal imaging camera under nominal working conditions of a chip. To build the dynamic power density map model, we apply generative adversarial networks (GAN) based on the workload-related features. Our study shows that the estimated total powers match the manufacturer's total power measurements extremely well. Experimental results further show that the predictions of power maps are quite accurate, with the RMSE of only 4.98mW/mm2, or 2.6% of the full-scale error. The speed of deploying the proposed approach on an Intel Core i7-10710U is as fast as 6.9ms, which is suitable for real-time estimation.

In this work, we propose a fast full-chip thermal numerical analysis approach based on an enhanced physics-informed neural networks (PINN) framework. The new method, called ThermPINN, leverages both PINN-based DNN optimization framework and analytic solutions of simplified thermal problems for solving thermal partial differential equations (PDE). The resulting ThermPINN leads to more efficient training speed of DNN networks and more scalability for solving large PDE problems. Specifically, we propose to partially enforce physics laws based on closely related analytic solutions to simpler problems. As a result, we are able to significantly reduce the number of variables in the loss function and easily meet boundary conditions. To consider the impact of various ambient temperatures and effective convection coefficients, which are influenced by different design parameters and run-time conditions, we develop a parameterized thermal analysis technique. This technique enables design space exploration and uncertainty quantification (UQ), which are critical for ensuring the reliability of integrated circuits under various operating conditions. The numerical results on alpha21264 processor show that the proposed ThermPINN has 2× speedup and 3× better accuracy over the state-of-the-art thermal simulator, VarSim. The experimental results for 2-D full-chip thermal analysis of 3171 cases show that the proposed parameterized ThermPINN considering both training and inference time can achieve a 6× speedup over commercial COMSOL with an average mean absolute error (AE) of 0.47 K. In terms of training time, the proposed parameterized ThermPINN is 11× faster than the parameterized plain PINN with similar accuracy. The UQ analysis with 5000 samples for maximum temperature propagated from ambient temperature shows that the parameterized ThermPINN and parameterized plain PINN are 113× and 22× faster than COMSOL, respectively.

This paper presents a novel approach for real-time estimation of spatial thermal maps for the commercial AMD Ryzen 7 4800U 8-core microprocessor using a transformer-based machine learning method. The proposed method, called ThermTransformer, leverages real-time performance metrics of the AMD chip, provided by uProf 4.0, to accurately estimate transient thermal maps. These maps can be valuable for dynamic thermal, power, and reliability controls requiring higher accuracy. Unlike traditional Convolutional Neural Networks (CNN) designed for image data or Recurrent Neural Networks (RNN) suitable for transient data, ThermTransformer is based on a modified self-attention architecture. It takes time-series performance metrics information as input and directly generates transient thermal images. Our results demonstrate that this transformer-based method achieves the best of both worlds - surpassing CNN in prediction quality and performing well for transient data. Exper-imental results reveal that ThermTransformer achieves highly accurate predictions of power maps, with an RMSE of only 0.36°C or 0.8% of the full-scale error. Additionally, it outperforms the recently proposed GAN-based thermal map estimation method, ThermGAN, by 1.66x and the LSTM-based thermal prediction method, RealMaps, by 6.09x in terms of accuracy on average. Furthermore, the proposed approach can be efficiently deployed on the target chip, providing real-time estimation with a speed as fast as 14ms.