DeepOHeat-v2 accelerates 3D-IC thermal optimization with self-improving AI

Researchers have developed DeepOHeat-v2, an advanced operator learning surrogate designed to significantly speed up and improve the accuracy of thermal optimization in three-dimensional integrated circuits (3D-ICs). This new version addresses limitations of its predecessor, DeepOHeat-v1, particularly in handling high-contrast geometries and discontinuous conductivities that previously caused errors. DeepOHeat-v2 employs a discretized physics loss and a self-improving framework with a trust gate to refine predictions, reducing mean peak temperature error to 0.55 K and achieving a 56x speedup compared to traditional methods. AI

IMPACT This advancement could lead to more efficient and reliable thermal management in complex chip designs, potentially impacting hardware development and performance.

RANK_REASON The cluster contains a research paper detailing a new AI model for a specific technical application. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• arXiv
• DeepOHeat-v1
• DeepOHeat-v2
• three-dimensional integrated circuit

• paper
• infra

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →