CPINN-ABPI: Physics-Informed Neural Networks for Accurate Power Estimation in MPSoCs

TL;DR

This work addresses the challenge of accurate fine-grained power estimation in MPSoCs by showing ABPI, while efficient and temperature-independent, suffers from real-world accuracy gaps. It introduces CPINN-ABPI, a hybrid approach that augments ABPI with a Custom Physics-Informed Neural Network, including a two-branch architecture, a three-component loss with data, physics, and guidance terms, and NSGA-II for multi-objective optimization. The method initializes ABPI parameters, learns residual corrections, and enforces physical constraints, achieving substantial MAE and MSE improvements across real hardware (Jetson Xavier AGX) and simulated heterogeneous SoCs while maintaining real-time inference. These results demonstrate CPINN-ABPI’s potential to enable dependable, fine-grained power management and more responsive thermal control in next-generation MPSoCs and heterogeneous architectures.

Abstract

Efficient thermal and power management in modern multiprocessor systems-on-chip (MPSoCs) demands accurate power consumption estimation. One of the state-of-the-art approaches, Alternative Blind Power Identification (ABPI), theoretically eliminates the dependence on steady-state temperatures, addressing a major shortcoming of previous approaches. However, ABPI performance has remained unverified in actual hardware implementations. In this study, we conduct the first empirical validation of ABPI on commercial hardware using the NVIDIA Jetson Xavier AGX platform. Our findings reveal that, while ABPI provides computational efficiency and independence from steady-state temperature, it exhibits considerable accuracy deficiencies in real-world scenarios. To overcome these limitations, we introduce a novel approach that integrates Custom Physics-Informed Neural Networks (CPINNs) with the underlying thermal model of ABPI. Our approach employs a specialized loss function that harmonizes physical principles with data-driven learning, complemented by multi-objective genetic algorithm optimization to balance estimation accuracy and computational cost. In experimental validation, CPINN-ABPI achieves a reduction of 84.7\% CPU and 73.9\% GPU in the mean absolute error (MAE) relative to ABPI, with the weighted mean absolute percentage error (WMAPE) improving from 47\%--81\% to $\sim$12\%. The method maintains real-time performance with 195.3~$μ$s of inference time, with similar 85\%--99\% accuracy gains across heterogeneous SoCs.

Figures (9)

• Figure 1: Experimental setup overflow for the simulation and the practical validation
• Figure 2: Simplified Jetson AGX Xavier Floorplan
• Figure 3: Jetson Xavier AGX with CPU and GPU Power and Temperature Sensors
• Figure 4: Variation in CPU and GPU temperature and Power over Time
• Figure 5: Custom Dataset Generation Workflow for a Heterogeneous SoC