← Back to the feed
AI & softwareResearch findings

Flexible AI-Based Thermal Analysis for Heterogeneous IC Packages

Original published 2026.10.19 07:51 KSTAdded to Blackdeck 2026.10.19 14:18 KST

What has changed?

DAIST decomposes package structures into block-level subdomains and uses neural operators, enabling reuse across different topologies without retraining

Key points

  1. Accurate thermal analysis of heterogeneous 2.5D/3D-IC packages is computationally prohibitive
  2. DAIST decomposes the package into block-level subdomains and uses neural operators for iterative coupling
  3. Models can be reused across different package topologies without retraining, offering tunable accuracy-runtime tradeoffs

Connect it to your work

Block-level models can be directly applied to unseen package configurations, avoiding retraining and maintaining accuracy

What remains unproven

Thermal accuracy under real-world material and interface conditions remains unverified

Originals used for this note

Links open the original publisher in a new tab.

  1. semiconductorsA Composable Thermal Solver For Heterogeneous 2.5D/3D-IC Packages (Cadence, Notre Dame)