🔬 materials science

The conductance filtration: sink-relative persistence and exact edge sensitivity for heat transport in particle-filled composites

This paper introduces a conductance filtration framework with sink-relative persistence to model heat transport in particle-filled composites, demonstrating that dissipation shares correspond to the logarithmic sensitivity of effective conductance and enabling exact closed-form predictions for single-edge perturbations across contact, near-contact, and bond mechanisms.

Original authors: Daisuke Yasufuku (U-MAP Co., Ltd.)

Published 2026-09-28

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Daisuke Yasufuku (U-MAP Co., Ltd.)

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Heat is the silent enemy of modern electronics. As computers become faster and more powerful, they pack more energy into smaller spaces, generating immense amounts of waste heat. In the vast server farms that power the internet, this heat has grown so intense that engineers are moving from simple air cooling to complex liquid systems. Yet, even with powerful pumps and coolants, the bottleneck often lies not in the liquid, but in the tiny, invisible layers of material that sit between the hot computer chip and the cooling system. These layers, known as thermal interface materials, are composites made of a soft, insulating glue filled with hard, heat-conducting particles. The goal is simple: pack as many of these conductive particles as possible into the glue to create a highway for heat to escape. However, simply adding more particles does not always work. Sometimes, making the particles more conductive or packing them tighter yields no improvement, leaving engineers guessing which part of the mixture is actually holding back the flow.

For decades, the design of these materials has relied on a trial-and-error approach. Engineers adjust the size of the particles, the type of filler, and the surface chemistry, hoping to hit a target for how well the material conducts heat. But the process that turns these choices into a final result is a black box. The particles form a complex, random network, and heat must navigate a maze of tiny gaps, direct touches, and the surrounding glue. Until now, there has been no way to look inside this maze and see exactly where the traffic jams are. A new study by Daisuke Yasufuku changes this by creating a precise map of the heat's journey through these composites, revealing exactly which connections matter and which ones are dead ends.

The researcher approached the problem by treating the composite not as a solid block, but as a network of connections. Imagine the particles as islands and the heat paths as bridges between them. Some islands touch directly, creating a strong bridge. Others are separated by a microscopic gap filled with the glue, creating a weaker bridge. Still others are part of the same continuous chunk of material, forming a solid road. The study assigns a specific value to the strength of each of these bridges based on the distance between particles and the materials involved. By calculating these values for every single pair of particles in a simulated mixture, the researcher built a complete graph of the thermal landscape. This map allows for a new kind of analysis called a filtration, which sorts these connections from strongest to weakest, revealing how the network comes together to carry heat.

The most significant finding of this work is the discovery of a precise mathematical relationship between the strength of a connection and the heat it actually carries. Intuitively, one might assume that the strongest bridges carry the most heat. The study proves this is often true: in a single simulation, the top 1% of the strongest connections carried nearly half of the total heat flow. However, the researcher demonstrated that the share of heat dissipated by any single connection is exactly equal to how sensitive the total heat flow is to a change in that specific connection's strength. This means that if you want to improve the material, you cannot just look for the strongest particles or the tightest packing. You must identify the specific connections that are currently limiting the flow. The study provides a method to calculate this "dissipation share" for every single connection in the network with a single, efficient calculation, pinpointing the exact bottlenecks that need fixing.

This method also reveals that the type of filler material changes the rules of the game. When using a highly conductive filler like aluminum nitride, the heat flows mostly through the points where particles touch each other. In this scenario, the resistance at the surface where the particles meet is the main obstacle, and improving the surface treatment of the particles yields a massive improvement in performance. However, when using a less conductive filler like silica, the heat is forced to travel through the surrounding glue matrix for a significant portion of its journey. In this case, polishing the particle surfaces does almost nothing, because the bottleneck has shifted to the glue itself. The study shows that a surface treatment that boosts performance by nearly 75 percent for one type of filler might increase performance by only 1 percent for another, simply because the limiting mechanism has changed.

The researcher tested these ideas on eight different scenarios, varying the particle size, the filler type, and the surface treatment. The results showed that the critical factor is not just how many particles are packed in, but how they are arranged and what they are made of. In some cases, the most effective way to improve the material was to treat the particle surfaces; in others, it was to change the glue. The study also introduced a way to predict the exact outcome of a change without having to rebuild the entire simulation. By using a closed-form calculation, the researcher showed that one could predict how much the total heat flow would increase if a specific connection were improved, allowing for a rapid design loop where engineers can test ideas instantly.

This work does not claim to have solved the problem of thermal management forever, nor does it provide a single magic formula for all materials. The simulations are based on specific conditions, and the model has limits, particularly when the packing of particles is very loose or when the particles are long fibers rather than spheres. However, the study successfully rules out the idea that a single measure of particle conductivity or packing density is enough to predict performance. It establishes that the internal structure of the material is far more complex, and that the path of least resistance is not always the strongest path. By making the internal heat transfer mechanisms visible and quantifiable, this research offers a new tool for engineers. Instead of guessing which variable to change, they can now look at the map of the material's internal network, see exactly where the heat is getting stuck, and make a targeted change to the specific mechanism that is holding them back.

Technical Summary: Conductance Filtration and Sink-Relative Persistence in Particle-Filled Composites

Problem Statement
As electronic packaging density increases, thermal management relies heavily on thermal interface materials (TIMs), which are particle-filled composites. Designers currently face a "black box" problem: they can manipulate four variables (packing fraction, particle size distribution, filler species, and surface treatment), but standard evaluation returns only a single scalar output—the effective thermal conductivity ( $k_{eff}$ ). When performance falls short, the internal mechanisms limiting heat transport (e.g., interface resistance vs. matrix conduction) remain unidentified. Existing models either provide effective properties without mechanistic breakdown (Effective Medium Theory) or rely on idealized unit cells that cannot accept arbitrary packed structures. Furthermore, recent topological approaches using persistent homology typically use geometric axes (like distance) rather than physical quantities (like conductance), missing the material-specific physics of the transport process.

Methodology
The paper proposes a unified framework that bridges structural geometry and physical transport properties using a graph representation and topological data analysis.

Branch-Conductance Model:
The authors map the packed structure onto a weighted graph $G=(V, E, g)$ where vertices are particles and edges represent heat exchange paths. They formulate the thermal conductance $g_e$ of each particle pair into three distinct branch types within a single construction:
- Near-contact: Particles separated by a thin matrix film. Conductance is derived from an annular integral incorporating interfacial resistance (Kapitza radius) and saturation effects due to diffusion limits within the particles.
- Contact: Particles touching directly. Conductance is the parallel sum of a solid spot path (including constriction resistance and interfacial resistance) and a remaining matrix annulus.
- Bond: Particles belonging to the same continuous solid (e.g., an aggregate). Conductance is purely axial bulk conduction, excluding constriction resistance to avoid discretization artifacts.
- Through-particle resistance: Added in series to edges to account for heat diffusion through the particle interior, ensuring vertices remain isothermal nodes.
Conductance Filtration and Sink-Relative Persistence:
Instead of using geometric distance as the filtration axis, the paper introduces the conductance filtration, where edges are added in decreasing order of thermal conductance $g_e$ .
- Sink-Relative Persistence: Standard persistent homology tracks connectivity among particles. Here, the authors define persistence relative to a "sink boundary" (the cold side of the specimen). A connected component "dies" when it connects to the sink. This yields a barcode where the death time of the component containing the source boundary defines the critical conductance level ( $\theta^*$ ), representing the bottleneck for percolation.
Sensitivity and Dissipation Share:
The paper establishes a mathematical link between the network solution and sensitivity analysis:
- Dissipation Share ( $P_{share}$ ): Defined as the fraction of total energy dissipation occurring on a specific edge.
- Logarithmic Sensitivity: Using the envelope theorem, the authors prove that the dissipation share is exactly equal to the logarithmic sensitivity of the effective conductance with respect to that edge's conductance ( $P_{share} = \partial \log G_{eff} / \partial \log g_e$ ).
- Exact Finite Change: A closed-form expression is derived to predict the new effective conductance after a finite change to a single edge, requiring only one sparse solve and one matrix factorization.

Key Contributions

Unified Branch Model: A single construction treating contact, near-contact, and bond branches with explicit validity conditions, avoiding the need for separate models or idealized lattices.
Physical Filtration Axis: The introduction of thermal conductance as the filtration axis, allowing the topological descriptor to reflect material properties (filler species, surface treatment) rather than just geometry.
Exact Sensitivity Metric: The proof that the dissipation share is the exact logarithmic sensitivity of the system's performance. This allows for the precise identification of limiting edges without heuristic approximations.
Computational Efficiency: The framework requires only one sparse linear solve and one factorization to compute the full dissipation profile and predict finite changes, making it feasible for large-scale structures (up to 8,000 particles with near-linear scaling).

Results and Numerical Demonstration
The framework was applied to eight conditions varying filler species (AlN vs. SiO2), particle size (5µm vs. 25µm), and surface treatment (untreated vs. treated) at a fixed packing fraction of 0.70.

Limiting Mechanisms: The analysis revealed that the limiting mechanism shifts with material. For high-conductivity fillers (AlN), contact branches dominate, making interfacial resistance the bottleneck. For low-conductivity fillers (SiO2), near-contact branches through the matrix carry significant heat, rendering interfacial improvements less effective.
Design Loop: The "dissipation share" correctly identified which edges to modify, whereas ranking by raw conductance ( $g_e$ ) performed no better than random selection (Spearman correlation of 0.315).
Idealization Ratios: The ratio of effective conductance under idealized mechanisms (zero interface resistance, infinite matrix conductivity, etc.) quantified the potential gain for each mechanism, showing that a single design change can shift which mechanism is limiting.
Critical Conductance Level: $\theta^*$ was shown to correlate with $k_{eff}$ better than packing fraction alone, spanning 1.7 decades in variation compared to the 1.0 decade span of $k_{eff}$ at a fixed packing fraction.

Significance and Claims
The paper claims to close the design loop for particle-filled composites by providing a single graph representation that accepts arbitrary packed structures and returns:

The identification of limiting mechanisms (interface, matrix, or particle).
The exact sensitivity of the system to specific edge changes.
A prediction of the response to finite changes in closed form.

The authors explicitly state that the framework does not claim to predict absolute $k_{eff}$ values with higher accuracy than existing methods (agreement with Effective Medium Theory is within ~25% median relative difference), nor does it claim to rank structures perfectly. Instead, its significance lies in elucidating the internal heat transfer mechanisms and providing an exact sensitivity analysis that guides how to change design variables (e.g., which surface treatment or filler size to prioritize) rather than just reporting a final performance number. The method is limited to steady-state heat transport, assumes isothermal particles, and currently underestimates conductivity at very low packing fractions where parallel matrix paths are not captured.

Technical Summary: Conductance Filtration and Sink-Relative Persistence in Particle-Filled Composites

More like this