Work

HotPROTON: A System-Level Framework for Thermal-Aware Electromigration Analysis in Manycore Systems

Year: 2026

Type: article

Abstract: Electromigration (EM) has emerged as a critical reliability concern in modern manycore systems, driven by increasing power densities and thermal stress. Existing models and tools typically operate at the device or interconnect level and fail to capture the spatial, temporal, and architectural variations necessary for accurate system-level analysis. This gap becomes especially pronounced in large-scale designs, where the complex interplay between localized current crowding, dynamic thermal profiles, and workload-dependent current patterns collectively accelerate EM-induced failures. To address these challenges, we propose HotPROTON, a holistic approach that connects physics-based EM modeling with system-level thermal profiles and workload-dependent current characteristics. HotPROTON is the first complete physics-based system-level EM analysis framework for manycore architectures. It tightly integrates HotSniper, a manycore thermal and power simulator, with PROTON, a physics-driven EM stress analysis tool, enabling full-stack evaluation from workload execution to material degradation. Importantly, HotPROTON incorporates temperature-aware EM stress evaluation, where thermal variation directly influences the stress computation, replacing traditional empirical models with a physics-driven degradation approach. The framework introduces a novel Power Delivery Network (PDN) generation module that maps simulation outputs into physically accurate PDN models, enabling transient-aware core-level analysis. Moreover, the proposed tool replaces empirical models with physics-based time-to-failure (TTF) computation in order to determine the corresponding architectural reliability metric, i.e., R-value. Experimental validation using PARSEC benchmarks on a manycore system demonstrates the framework’s capability to accurately predict reliability limits and R-value evolution under various workload scenarios.

Source: ACM Transactions on Architecture and Code Optimization

DOI: 10.1145/3844139