Predicting Heat Transport across Solid Interfaces

Enable energy-efficient compute by understanding heat transport across interfaces in devices.

Problem 05OpenUpdated Nov 2026

Thermal transportInterfacesSemiconductors

Written with inputs from

  • Prof. Andrew KummelProfessor of Chemistry, UC San Diego

Solid Interfaces: Where Heat Gets Stuck

The amount of heat a chip produces and the speed at which it dissipates both depend on the materials it is made of. In particular, solid–solid interfaces are a growing contributor to the thermal resistance of devices, so a better understanding of interfaces is critical to more efficient compute.

Standard mismatch models, the acoustic mismatch model of Little (1959) and the diffuse mismatch model of Swartz and Pohl (1989), fix the answer from bulk phonon properties alone and cannot reproduce measured thermal resistances. Atomistic calculations can treat one crystalline interface at a time but are too slow for routine use, and intractable for polycrystalline, amorphous or polymer interfaces. Can we surpass the mismatch models with methods that generalize across interfaces?

Heat crosses the interface as lattice vibrations: a phonon arriving from the hot metal side is partly transmitted and partly reflected, and the transmitted share, together with the electron–phonon coupling on the metal side, sets the thermal boundary conductance G(T).

The problem

Build a computationally tractable framework that predicts the thermal boundary conductance G(T) and resistance RK(T) = 1/G(T) across specified solid–solid interfaces without fitting a conductance to each specimen.

Demonstrate transferable improvement over the acoustic and diffuse mismatch models (AMM/DMM).

Three results to reproduce

All conductances in MW m⁻² K⁻¹. The numbers are the published values from the papers cited.

Temperature dependence at sharp metal/semiconductor interfaces

Reproduce the measured G(T) curves for Al(111)/Si(111) and Al(111)/GaN(0001) over 80–700 K: the room-temperature values and the renewed, approximately linear increase above about 400 K.

379
Al/Si · 300 K
423
Al/GaN · 300 K
>400 K
linear rise

Reference to reproduce: Li et al., Nature Communications 13, 4901 (2022), Fig. 1a–b.

Carbide-assisted heat transfer in Cu/diamond

Reproduce the effect of a highly crystalline, 10 nm TiC interlayer: the conductance rises by 48% relative to the study’s direct Cu/diamond reference, GCu/TiC/diamond / GCu/diamond = 1.48.

1.48
G ratio, with / without TiC
10 nm
crystalline TiC

Reference to reproduce: Chang et al., ACS Applied Materials & Interfaces 11, 26507 (2019).

Submissions

We evaluate submissions on their ability to reproduce accepted literature results that we collate: the three datasets above, the G(T) curves of Li et al. (2022), the TiC interlayer effect of Chang et al. (2019) and the PMMA stack response of Sandell et al. (2020).

We then freeze a model that reproduces them and run it on a series of hidden, industrially relevant solid–solid interfaces, AlN/GaN for example, comparing its predictions with the acoustic and diffuse mismatch models and with slower atomistic methods as the ground truth where they can be run.

Submissions should include:

  • Method: A description of the proposed computational approach and its underlying scientific principles.
  • Implementation: Working code and instructions to run it on a solid–solid interface.

Evaluation considers:

  • Predictive accuracy: Agreement with the three published datasets, and with the ground truth on the hidden interfaces.
  • Transferability: Improvement over the acoustic and diffuse mismatch models across the hidden interfaces, with no loss on the cases they already get right.
  • Reproducibility: The ability to run the code on a stated interface and recover the reported results.

Next: 01 · A Predictive Theory of Cuprate Superconductivity

Next problemAll problems