A Composable Thermal Solver for Heterogeneous 2.5D/3D-IC Packages
Developed by Cadence Design Systems and the University of Notre Dame
Overview
As 2.5D and 3D-IC designs push integration density ever higher, thermal behavior has become a first-order design constraint rather than a downstream verification concern. Traditional monolithic thermal solvers struggle to scale across the heterogeneous architectures that now dominate advanced packaging—stacked chiplets, interposers, hybrid bonding layers, and mixed-material substrates all coexist within a single package, each with distinct thermal properties and power profiles.
To address this, Cadence and the University of Notre Dame have collaborated on a composable thermal solver architecture tailored specifically for heterogeneous 2.5D/3D-IC packages.
Why Composable Thermal Solving Matters in 2026
By 2026, the industry has broadly shifted toward chiplet-based and 3D-stacked designs, driven by the economics of advanced nodes and the demands of AI accelerators, high-bandwidth memory (HBM) stacks, and disaggregated SoCs. In this landscape:
- Power densities have climbed sharply, particularly in logic-on-logic 3D stacks and HBM-adjacent compute chiplets, making localized hotspots a dominant reliability risk.
- Heterogeneous materials—silicon, interposer glass or organic substrates, underfills, and hybrid-bond interfaces—create complex, non-uniform thermal paths that demand flexible modeling approaches.
- Design iteration speed is critical; thermal analysis must keep pace with rapid package-level floorplanning rather than acting as a late-stage bottleneck.
A composable solver addresses these pressures by allowing thermal models to be assembled modularly from reusable, physics-aware building blocks—each component (e.g., a die, interposer, or TIM layer) can be modeled and coupled independently, then composed into a full-package solution.
Key Concepts
Modular Model Composition
Rather than meshing the entire package as a single monolithic domain, the composable approach decomposes the thermal problem into sub-domains that map naturally to the physical hierarchy of a 2.5D/3D-IC package. Each sub-domain can be solved or reduced independently and coupled through well-defined interfaces, enabling reuse across design variants and faster what-if exploration.
Heterogeneity-Aware Modeling
The solver explicitly accounts for the diverse material stacks and geometries inherent to heterogeneous integration, supporting accurate capture of vertical and lateral heat flow across dies, bond layers, and package substrates.
Scalability for Multi-Die Systems
By composing solutions from smaller, validated blocks, the approach scales to the growing die counts and complexity of chiplet-based packages without requiring full-package re-meshing for every design change.
Significance
The Cadence–Notre Dame collaboration reflects a broader 2026 trend: thermal analysis is being re-architected as a modular, design-integrated capability rather than a standalone sign-off step. For heterogeneous 2.5D/3D-IC packages—where thermal paths are irregular, materials are mixed, and iteration speed determines time-to-market—composable thermal solving offers a practical path toward faster, more accurate thermal-aware package design.
Source: Cadence Design Systems and the University of Notre Dame. Technical details to be presented in the associated paper.