Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

3D-ICE 4.0: Accurate and efficient thermal modeling for 2.5D/3D heterogeneous chiplet systems

Kai Zhu, Darong Huang, Luis Costero, David Atienza

TL;DR

The paper tackles the challenge of accurately and efficiently modeling heat in 2.5D/3D chiplet systems with complex, anisotropic layouts. It introduces 3D-ICE 4.0, a framework that preserves material heterogeneity, employs adaptive vertical-layer division, and uses temperature-aware non-uniform grids, all accelerated by OpenMP and a parallel sparse solver. Key innovations include automatic layout-based model generation via quadtree tiling, layer-division driven by vertical resistance variance, and gradient-guided grid refinement. Empirical results show substantial speedups over PACT, close agreement with COMSOL for both steady-state and transient scenarios, and substantial reductions in grid counts for a given accuracy, making it well-suited for design-space exploration and runtime thermal management in heterogeneous chiplet architectures.

Abstract

The increasing power densities and intricate heat dissipation paths in advanced 2.5D/3D chiplet systems necessitate thermal modeling frameworks that deliver detailed thermal maps with high computational efficiency. Traditional compact thermal models (CTMs) often struggle to scale with the complexity and heterogeneity of modern architectures. This work introduces 3D-ICE 4.0, designed for heterogeneous chip-based systems. Key innovations include: (i) preservation of material heterogeneity and anisotropy directly from industrial layouts, integrated with OpenMP and SuperLU MT-based parallel solvers for scalable performance, (ii) adaptive vertical layer partitioning to accurately model vertical heat conduction, and (iii) temperature-aware non-uniform grid generation. The results with different benchmarks demonstrate that 3D-ICE 4.0 achieves speedups ranging from 3.61x-6.46x over state-of-the-art tools, while reducing grid complexity by more than 23.3% without compromising accuracy. Compared to the commercial software COMSOL, 3D-ICE 4.0 effectively captures both lateral and vertical heat flows, validating its precision and robustness. These advances demonstrate that 3D-ICE 4.0 is an efficient solution for thermal modeling in emerging heterogeneous 2.5D/3D integrated systems.

3D-ICE 4.0: Accurate and efficient thermal modeling for 2.5D/3D heterogeneous chiplet systems

TL;DR

The paper tackles the challenge of accurately and efficiently modeling heat in 2.5D/3D chiplet systems with complex, anisotropic layouts. It introduces 3D-ICE 4.0, a framework that preserves material heterogeneity, employs adaptive vertical-layer division, and uses temperature-aware non-uniform grids, all accelerated by OpenMP and a parallel sparse solver. Key innovations include automatic layout-based model generation via quadtree tiling, layer-division driven by vertical resistance variance, and gradient-guided grid refinement. Empirical results show substantial speedups over PACT, close agreement with COMSOL for both steady-state and transient scenarios, and substantial reductions in grid counts for a given accuracy, making it well-suited for design-space exploration and runtime thermal management in heterogeneous chiplet architectures.

Abstract

The increasing power densities and intricate heat dissipation paths in advanced 2.5D/3D chiplet systems necessitate thermal modeling frameworks that deliver detailed thermal maps with high computational efficiency. Traditional compact thermal models (CTMs) often struggle to scale with the complexity and heterogeneity of modern architectures. This work introduces 3D-ICE 4.0, designed for heterogeneous chip-based systems. Key innovations include: (i) preservation of material heterogeneity and anisotropy directly from industrial layouts, integrated with OpenMP and SuperLU MT-based parallel solvers for scalable performance, (ii) adaptive vertical layer partitioning to accurately model vertical heat conduction, and (iii) temperature-aware non-uniform grid generation. The results with different benchmarks demonstrate that 3D-ICE 4.0 achieves speedups ranging from 3.61x-6.46x over state-of-the-art tools, while reducing grid complexity by more than 23.3% without compromising accuracy. Compared to the commercial software COMSOL, 3D-ICE 4.0 effectively captures both lateral and vertical heat flows, validating its precision and robustness. These advances demonstrate that 3D-ICE 4.0 is an efficient solution for thermal modeling in emerging heterogeneous 2.5D/3D integrated systems.

Paper Structure

This paper contains 11 sections, 3 equations, 12 figures, 2 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related work
  3. 3D-ICE 4.0 framework
  4. Model generation with detailed layouts
  5. Adaptive layer division
  6. Local grid refinement
  7. Experimental results
  8. Case1: Accuracy and efficiency validation
  9. Case2: Modeling with detailed material distribution and accuracy gains by adaptive layer division
  10. Case3: Efficiency improvement by local refined grids
  11. Conclusion

Figures (12)

  • Figure 1: Workflow of compact thermal modeling framework. Newly added/updated stages in 3D-ICE 4.0 are highlighted in orange.
  • Figure 2: Thermal map for a 4-chiplet system with (a) abstract homogeneous model, (b) heterogeneous model with detailed material distribution.
  • Figure 3: Extraction of detailed material distribution, (a) GDSII layout of one layer, (b) generated tiles, (c) overlapped ratio for each tile.
  • Figure 4: Workflow of the generation of thermal-aware non-uniform grids and simulation.
  • Figure 5: (a) Temperature map from 3D-ICE 4.0 with a uniform $256\times256$ grid, (b) the absolute difference map between 3D-ICE 4.0 and PACT.
  • ...and 7 more figures