Diffuse Mismatch Model (DMM)

• The diffuse mismatch model (DMM) is a theoretical framework that estimates thermal boundary conductance via vibrational mismatch between solids.
• It computes phonon transmission probabilities using density of vibrational states and group velocities, assuming incoherent, isotropic scattering at interfaces.
• While effective for disordered interfaces, DMM is limited in epitaxial or low-disorder systems where coherent effects and anharmonicity significantly impact phonon transport.

The diffuse mismatch model (DMM) is a cornerstone theoretical framework for estimating the thermal boundary conductance (TBC) across solid–solid interfaces in the phonon gas regime. The model is built on the hypothesis that incident phonons scatter incoherently and isotropically at an interface, losing all memory of their incoming direction and polarization. Transmission probabilities are determined solely by the vibrational mode density and group velocities on each side, assuming local thermal equilibrium within each solid. While DMM provides tractable, analytic expressions for TBC—including direct integration with full phonon dispersions—it is increasingly scrutinized for quantifying interfacial transport, especially for atomically smooth, epitaxial, or low-disorder systems, and for interfaces where wave effects, selection rules, and anharmonicity play significant roles.

1. Theoretical Foundations and Formulation

At the heart of DMM is the assumption that at an interface, phonon transmission and reflection become completely diffuse. Memory of the phonon's incoming wavevector and polarization is lost; the transmission probability depends only on frequency, density of states (DOS), and group velocity. The local Bose–Einstein population distribution $f(\omega,T)$ is preserved on each side. Under these postulates, the spectral interfacial conductance per unit area is

$$G_{\mathrm{DMM}}(T) = \int_0^{\omega_{\max}} \hbar\omega\, D_1(\omega)\, v_1(\omega)\, \tau_{1 \to 2}(\omega)\, \frac{\partial f_{\mathrm{BE}}(\omega, T)}{\partial T}\, d\omega$$

where $D_1(\omega)$ and $v_1(\omega)$ are the phonon density of states and group velocity perpendicular to the interface in material 1, and $\tau_{1 \to 2}(\omega)$ is the DMM transmission coefficient: $$\tau_{1 \to 2}(\omega) = \frac{\sum_j v_{2, j}(\omega) D_{2, j}(\omega)}{\sum_i v_{1, i}(\omega) D_{1, i}(\omega) + \sum_j v_{2, j}(\omega) D_{2, j}(\omega)}$$ Here, the sums run over all polarizations in both materials. This formally encodes the vibrational mismatch: the transmission is maximized for well-matched densities of vibrational states and diminishes with larger discrepancies.

For anisotropic and low-dimensional materials, DMM transmission coefficients and irradiation integrals can be extended using anisotropic Debye approximations or full-dispersion phonon data, allowing mode- and direction-resolved expressions (Gaskins et al., 2017, Brown et al., 2019).

2. Implementation and Computational Strategies

While early DMM applications relied on isotropic Debye models, contemporary implementations leverage full phonon dispersions and group velocities from density-functional perturbation theory (DFPT) and related lattice dynamics methods. For each branch $p$,

• $v_p(\omega) = d\omega/dk$ is extracted from the phonon dispersion,
• $D_p(\omega) \propto \omega^2 / v_p^3$ under the isotropic approximation, or numerically integrated using dense $\mathbf{q}$-grids for full-band data,
• Debye cutoffs $\omega_D$ are either taken from experiment or calculated by integrating the DOS up to the Brillouin zone boundary.

In systems with strong anisotropy (e.g., layered or 2D crystals), the irradiation integrals and DOS require adjustments for direction-dependent sound velocities and flexural modes, such as in graphene or h-BN, leading to piecewise anisotropic DMM (PWA-DMM) models for an accurate account of the non-linear ZA branch (Brown et al., 2019).

Full-dispersion DMM calculations can also account for near-interface alloying and strain by replacing bulk phonon data with those calculated for the strained or interfacially altered crystal regions (Ye et al., 2016).

3. Experimental Assessment and Limitations

Comprehensive experiments consistently reveal DMM's characteristic strengths and failures:

• In ZnO/GaN, DMM underpredicts TBC by nearly a factor of two (e.g., $G_{\mathrm{DMM}} = 250$ MW m$^{-2}$ K$^{-1}$ vs. measured $G_{exp} = 490$ MW m$^{-2}$ K$^{-1}$ at 300K), and fails especially for long-wavelength (zone-center) phonons that, in reality, transmit with near-unity efficiency because of wave-coherence, a phenomenon neglected by the DMM (Gaskins et al., 2017).
• For metal–silicide/Si interfaces (e.g., CoSi$_2$/Si), full-dispersion DMM reproduces experimental TBC above $\sim$100K (e.g., 480 MW m$^{-2}$ K$^{-1}$ at 300K), but overpredicts for interfaces with large vibrational mismatch or at lower temperatures where coherent effects dominate (Ye et al., 2016).
• At cryogenic temperatures and for specular, atomically-smooth interfaces (as in Si/Al KIDs), DMM is clearly disfavored compared to acoustic mismatch model (AMM), and overestimates transmission; in these cases, specular reflection and phonon caustics persist and dominate the transmission statistics (Martinez et al., 2018).

Mode-by-mode comparisons with atomistic Green's function (AGF) and frequency-domain perfectly matched layer (FD-PML) simulations reveal that even strongly disordered or interdiffused interfaces do not fully randomize phonon momentum or polarization: substantial angular memory and selection rules persist well beyond the single-monolayer scale, contrary to DMM's basic premise (Kakodkar et al., 2016, Song et al., 2021). This results in DMM mispredicting both the absolute TBC and the angular/branch-resolved transmission statistics, especially for high-frequency or low-frequency/extreme-mismatch regimes.

4. Extensions and Modifications: Partial Diffusivity and Intermediate Models

Recognizing the inadequacy of the bare DMM for most realistic interfaces, modified models such as the mixed mismatch model (MMM) have been proposed (Zhang et al., 2016). MMM interpolates between specular (AMM-like) and fully diffuse (DMM-like) behaviors by introducing a frequency-dependent specularity parameter $p(\omega)$, linked to the interface roughness or interfacial density of states (IDOS). The total transmission is given by

$$\alpha^{\mathrm{mix}}_{1\to2,j}(\omega) = p(\omega)\,\alpha^{\mathrm{AMM}}_{1\to2,j}(\omega) + [1-p(\omega)]\,\alpha^{\mathrm{DMM}}_{1\to2,j}(\omega)$$

where $p(\omega)$ is obtained either from Ziman's formula for surface roughness or from the normalized IDOS. Such models recover pure DMM (p=0) for amorphous/rough interfaces and AMM (p=1) for atomically flat ones, providing improved agreement with molecular dynamics and experimental TBC data. The value of $p(\omega)$ can be systematically related to the interfacial atomistic configurations.

In practical terms, DMM can be regarded as a lower bound for TBC in systems with maximum disorder, while interfaces with substantial specular character demand broader frameworks to capture the crossover behavior.

5. Regimes of Applicability and Quantitative Performance

A comparative assessment of various experimental systems demonstrates that:

• DMM is most accurate for highly disordered or amorphous interfaces, or when mass and vibrational spectrum mismatch is substantial, ensuring multiple scattering and isotropization;
• For weakly disordered, smoothly graded, or epitaxial interfaces, DMM systematically underestimates TBC due to neglect of coherent transmission and persistent selection rules; it also overestimates transmission at high frequencies where anti-reflection or mode suppression occurs (Kakodkar et al., 2016, Gaskins et al., 2017);
• The model's accuracy is strongly system- and temperature-dependent. For ZnO/GaN, DMM underpredicts TBC by a factor $\sim$2 in the 77–500 K range. For Si/CoSi$_2$ and similar silicides, DMM matches experiment within $\sim$10% above 100K, but is less accurate for more strongly mismatched or smoother interfaces (Ye et al., 2016). Cryogenic limits and ultra-smooth boundaries, such as in KIDs, invalidate the DMM entirely in favor of AMM (Martinez et al., 2018).

6. Alternatives, Improvements, and Outlook

Hybrid models combining full dispersion, AGF transmission, and anharmonic/inhomogeneous scattering are recognized as critical for truly predictive TBC calculations in epitaxial and high-quality interfaces. AGF methods retain phase, coherence, and full in-plane momentum conservation, addressing the limitations of the DMM for low-disorder systems but remaining confined to elastic transport unless extended for inelastic effects (Gaskins et al., 2017, Song et al., 2021).

Analytical continuum models with explicit Born approximation for interfacial disorder provide insight at low frequencies but break down for higher frequencies or when disorder becomes non-perturbative (Song et al., 2021).

For material and device engineering, DMM offers, at best, a tractable lower bound for TBC, especially when rapid order-of-magnitude estimates or screening calculations are required for highly disordered or polycrystalline interfaces. Accurate device-level predictions (e.g., for UO$_2$/BeO composite fuels) require integration of DMM-predicted interfacial resistance with effective-medium models and careful attention to phase, interface area, and grain size distributions (Zhu et al., 2020).

7. Summary Table: TBC Prediction: DMM vs. Experiment

System	$G_{exp}$ (MW m$^{-2}$ K$^{-1}$)	$G_{DMM}$ (MW m$^{-2}$ K$^{-1}$)	$G_{exp}/G_{DMM}$	Reference
ZnO/GaN, 300 K	490	250	1.96	(Gaskins et al., 2017)
Si/CoSi$_2$, 300K	480	490	0.98	(Ye et al., 2016)
Si/PtSi, 300K	170	210	0.81	(Ye et al., 2016)

This table highlights that DMM may either quantitatively succeed (CoSi$_2$/Si) or systematically under-/over-predict depending on interface quality and spectral match.

---

In summary, the diffuse mismatch model provides an explicit, computationally efficient approach to estimating phonon-mediated TBC in highly disordered regimes, but its core assumptions—complete loss of angular/polarization memory and isotropy—are violated at epitaxial, smooth, or moderately disordered interfaces, as well as where inelastic and wave effects are non-negligible. Future progress in predictive modeling requires hybridizing DMM-like mode-counting with mode-resolved, wave-coherent, and anharmonic treatments, coupled to detailed knowledge of atomic-scale interface structure and dynamics.