ICANEWS

Engineering & Technology

Transient Multiscale Workflow for Thermal Analysis of 3DHI Chip Stacks with BEOL Heterogeneity

In the rapidly evolving landscape of modern package designs, technologies such as backside power delivery (BSPD) and 3D stacked chiplets are becoming increasing

Advancing Thermal Analysis for Modern Chip Stacks with a Transient Multiscale Workflow

In the rapidly evolving landscape of modern package designs, technologies such as backside power delivery (BSPD) and 3D stacked chiplets are becoming increasingly prevalent. These advanced designs necessitate a deeper understanding and more accurate prediction of thermal behavior, particularly concerning hot-spots. A critical challenge in this area involves accounting for the inherent heterogeneity found within back end of the line (BEOL) structures. While multiscale homogenization strategies have demonstrated effectiveness for steady-state simulations, the development of accurate 3D transient simulations that incorporate BEOL structures has remained an open challenge for researchers and engineers.

Addressing Heterogeneity in BEOL Structures

The complexity introduced by heterogeneous BEOL structures in 3D stacked chiplets poses a significant hurdle for traditional thermal analysis methods. These structures, comprising various materials and geometries at fine scales, exhibit non-uniform thermal properties. Precisely predicting hot-spots and overall thermal management in such intricate designs requires a method capable of capturing these fine-scale variations and their transient effects. The inability to accurately model these characteristics can lead to underperformance, reliability issues, and even failure of integrated circuits.

Previous research has shown promise with multiscale homogenization techniques for steady-state scenarios. However, the dynamic nature of thermal events in operating chip stacks demands transient analysis. This transient analysis, which considers temperature changes over time, is essential for understanding short-duration thermal peaks and managing dynamic power consumption. The transition from steady-state to transient often introduces significant computational and methodological complexities, further emphasizing the need for innovative solutions.

Research Goal: Accurate Transient Thermal Workflow for 3DHI Chip Stacks

A new research effort focuses on demonstrating a transient thermal workflow specifically designed to account for the 3D heterogeneous structures within the BEOL. This workflow targets problems characterized by strong and weak temporal scale separation. A fundamental assumption underpinning this research is the use of temperature-independent constitutive properties.

"Our workflow, based on Bloomfield et. al. 2025, automatically extracts, meshes, and homogenizes thermal properties from GDSII and OASIS files to construct thermal property maps."

This goal directly addresses the identified gap in current thermal analysis capabilities, moving beyond steady-state approximations to provide a more dynamic and intricate understanding of thermal behavior in advanced chip architectures.

Methodology for Thermal Property Extraction and Homogenization

The core of this transient thermal workflow relies on an automated process for extracting, meshing, and homogenizing thermal properties. This process is initiated from industry-standard GDSII and OASIS files, which are commonly used formats for integrated circuit layout data. The workflow leverages a methodology previously introduced by Bloomfield et. al. in 2025, indicating a building upon established foundational work.

  • Automatic Extraction: The workflow automatically extracts relevant geometrical and material information from the provided GDSII and OASIS design files. This automation is crucial for handling the immense complexity of modern chip layouts, which would be impractical to process manually.
  • Meshing: Once extracted, the structures are meshed. Meshing involves discretizing the continuous physical domain into a finite number of elements, a prerequisite for numerical simulation. The quality and type of mesh are critical for the accuracy and efficiency of thermal simulations.
  • Homogenization: After meshing, the thermal properties are homogenized. Homogenization is a technique used to derive effective macroscopic properties for heterogeneous materials based on their microscopic structure and properties. This process simplifies the computational model while retaining the effects of fine-scale heterogeneity, allowing for multiscale analysis.

These automated steps culminate in the construction of thermal property maps. These maps serve as a simplified, yet accurate, representation of the complex thermal characteristics of the BEOL structures within the chip stack, essential for subsequent thermal simulations.

Key Findings: Generation of Property Maps and Impact of Transient Effects

The research successfully generated property maps, specifically for heat capacity and conductivity. These maps were developed for a 1 mm by 1 mm SoC-style model die. The model die was constructed using LibreLane, a design tool, and was analyzed under two different grid resolutions:

  • 100 by 100 grids with 5 micron by 5 micron representative volume elements (RVEs)
  • 50 by 50 grids with 10 micron by 10 micron RVEs

The focus on different RVE sizes indicates an investigation into how the scale of homogenization impacts the resulting property maps and the overall accuracy of the thermal analysis. Smaller RVEs, such as the 5 micron by 5 micron elements, allow for a more detailed representation of heterogeneity, while larger RVEs, like the 10 micron by 10 micron elements, offer computational efficiency at potentially reduced resolution.

Expressions for Transient Effective Conductivity and Impact Demonstration

A significant aspect of the findings involves the provision of expressions for a transient effective conductivity. This is a crucial advancement, as it moves beyond static effective properties and allows for the capture of time-dependent thermal responses. The concept of transient effective conductivity acknowledges that the effective thermal properties of a heterogeneous material can vary over time, especially during short thermal events or rapid changes in power dissipation.

The research includes a demonstration of the impact of these transient effects for a single RVE. This single-RVE demonstration is vital for illustrating the conceptual difference between steady-state and transient homogenization. By showing the time-dependent behavior within a representative volume element, the research underscores the necessity of the new workflow for accurate hot-spot prediction in dynamic scenarios. The difference in thermal response over time when considering transient effective properties versus static ones can be substantial, affecting peak temperature predictions and thermal cycling behavior.

Provision of Transient Conductivity Maps

Further demonstrating the practical application of the workflow, transient conductivity maps have been provided for a specific time integration timestep, $dt=0.001$. The provision of these maps is a tangible output of the workflow, offering a detailed spatial and temporal representation of thermal conductivity across the model die. The choice of $dt=0.001$ signifies a granular temporal resolution, enabling the capture of fast thermal transients critical for modern high-performance chips.

These maps would show how the effective thermal conductivity at different locations within the chip changes over time, influenced by the underlying heterogeneous BEOL structures and the current thermal state. Such detailed mapping is an instrumental input for subsequent, more comprehensive transient thermal simulations of the entire chip stack. It allows for the accurate distribution of heat and prediction of temperature fields under dynamic operating conditions.

What's Next: Future Directions and Further Applications

While the study successfully demonstrates a transient thermal workflow and provides key insights into the generation of property maps and the impact of transient effects, the described research focuses on establishing the methodology and initial results. The current work assumes temperature-independent constitutive properties. Future research could explore the integration of temperature-dependent properties, which would add another layer of fidelity to the thermal models, especially for materials experiencing significant temperature excursions.

Extending the Workflow to Full Chip-Stack Simulations

The current demonstration applied the workflow to a 1 mm by 1 mm SoC-style model die. The logical next step for this research would be to apply the generated transient conductivity maps and the developed workflow to full 3DHI chip stack simulations. This would involve integrating the BEOL property maps into larger-scale package models, allowing for comprehensive thermal analysis of multi-chiplet systems. Such an extension would directly address the stated open challenge of accurate 3D transient simulations that include BEOL structures.

Furthermore, evaluating the computational efficiency and accuracy of this transient multiscale workflow when scaled to commercially relevant 3DHI designs would be critical. Benchmarking against experimental data or more detailed, albeit computationally expensive, fine-grained simulations for complex chip stacks would validate the practical utility of this approach for real-world hot-spot prediction and thermal management strategies.

Key Findings