Self-Attention to Operator Learning-based 3D-IC Thermal Simulation

2025-10-12

AI Summary

Thermal simulation of 3D integrated circuits faces efficiency bottlenecks due to high power density: conventional PDE solvers offer high accuracy but suffer from prohibitive computational cost, while existing machine learning methods—such as Fourier Neural Operators (FNOs)—often lose high-frequency thermal features and critically depend on scarce, expensive high-fidelity training data. Method: We propose an enhanced FNO framework integrating self-attention mechanisms and a U-Net architecture, coupled with a transfer learning strategy. Self-attention improves modeling of long-range thermal coupling; U-Net preserves fine-grained, high-frequency temperature gradients; and transfer learning enables effective fine-tuning using low-fidelity data, drastically reducing reliance on high-fidelity labels. Contribution/Results: Our method achieves state-of-the-art prediction accuracy while accelerating thermal simulation by 842× over finite element analysis, significantly enhancing design iteration efficiency without compromising fidelity.

Technology Category

• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning
• Search and Optimization: Sampling/Simulation-based Search
• Computer Vision: 3D Computer Vision

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Thermal management in 3D ICs is increasingly challenging due to higher power densities. Traditional PDE-solving-based methods, while accurate, are too slow for iterative design. Machine learning approaches like FNO provide faster alternatives but suffer from high-frequency information loss and high-fidelity data dependency. We introduce Self-Attention U-Net Fourier Neural Operator (SAU-FNO), a novel framework combining self-attention and U-Net with FNO to capture long-range dependencies and model local high-frequency features effectively. Transfer learning is employed to fine-tune low-fidelity data, minimizing the need for extensive high-fidelity datasets and speeding up training. Experiments demonstrate that SAU-FNO achieves state-of-the-art thermal prediction accuracy and provides an 842x speedup over traditional FEM methods, making it an efficient tool for advanced 3D IC thermal simulations.

Problem

• Addressing slow thermal simulation in 3D IC design
• Overcoming high-frequency information loss in ML models
• Reducing dependency on extensive high-fidelity datasets

Innovation

• Self-Attention U-Net FNO captures long-range dependencies
• Transfer learning fine-tunes low-fidelity data efficiently
• SAU-FNO achieves 842x speedup over traditional FEM methods