Thermal Boundary Resistance (TBR)

• Thermal Boundary Resistance is the area-specific resistance defined by the temperature drop across an interface under a normal heat flux.
• It is influenced by factors like vibrational mismatch, electron–phonon coupling, local bonding, and nanostructural effects that govern heat dissipation.
• Measurement and simulation methods such as TDTR, 3ω techniques, and atomistic modeling enable precise TBR evaluation for optimizing thermal management in layered devices.

Thermal boundary resistance (TBR), or Kapitza resistance, is the area-specific resistance associated with heat flow across an interface. It is defined by the temperature discontinuity that develops under a normal heat flux and is central to predicting heat dissipation in multilayer stacks, nanostructures, bonded heterojunctions, and internal interfaces such as domain walls. Its reciprocal is the thermal boundary conductance (TBC). Across current arXiv literature, TBR is treated not as a single universal material constant, but as an interface-specific quantity governed by vibrational mismatch, electron–phonon coupling, local bonding, interfacial disorder, finite-size effects, and the metrology used to infer it (Aller et al., 2024).

1. Formal definition and resistance-network description

The standard definition is

$R_{\mathrm{int}}=\frac{\Delta T}{q},$

where $\Delta T$ is the temperature drop across the interface and $q$ is the heat flux normal to that interface. The reciprocal conductance is

$G=\frac{1}{R_{\mathrm{int}}},$

and is commonly reported in $\mathrm{W\,m^{-2}\,K^{-1}}$, while TBR is commonly reported in $\mathrm{m^2\,K/W}$ or, for convenience, in $\mathrm{m^2\,K/GW}$ (Aller et al., 2024). The same relation is used in studies of atomically thin van der Waals stacks, oxide interfaces, bonded semiconductor–diamond systems, and cryogenic metal–substrate junctions (Choi et al., 2018).

In multilayer transport, TBR enters in series with bulk layer resistances. For a stack with layer thicknesses $t_i$, conductivities $k_i$, and interfaces $j$,

$\Delta T$0

This series form is used explicitly in semiconductor-on-diamond stacks, in $\Delta T$1-on-substrate extraction by $\Delta T$2, and in film-on-substrate heterostructures measured by multi-sensor electrical thermometry (Aller et al., 2024). In atomically thin systems, cross-plane transport is often dominated by interfacial terms to the extent that the inserted layers are treated as additional interfaces rather than as bulk resistors; for metal/WSe$\Delta T$3/Al$\Delta T$4O$\Delta T$5, for example, the total resistance is modeled as a sum of metal–WSe$\Delta T$6, WSe$\Delta T$7–Al$\Delta T$8O$\Delta T$9, and, for $q$0, WSe$q$1–WSe$q$2 contributions (Choi et al., 2018).

A recurrent conceptual point is that TBR is not restricted to chemically abrupt external heterointerfaces. It is also defined for internal structural discontinuities, including ferroelectric $q$3 domain walls in PbTiO$q$4, where an isolated wall has $q$5 at $q$6, and for effective nonequilibrium bottlenecks, such as magnon–phonon disequilibrium in SrRuO$q$7, where the cooling dynamics behave as though an interfacial resistance were present (Seijas-Bellido et al., 2017); (Langner et al., 2010).

2. Microscopic transport channels and theoretical descriptions

The simplest descriptions are the acoustic mismatch model (AMM) and diffuse mismatch model (DMM). AMM emphasizes transmission penalties from acoustic impedance mismatch, whereas DMM emphasizes diffuse scattering and overlap of vibrational phase space or phonon density of states (PDOS). In the language used for diamond/AlGaN, AMM is consistent with high TBR when acoustic impedances differ strongly, while DMM rationalizes TBR reduction when an interlayer introduces intermediate-frequency states and increases mode overlap (Aller et al., 2024). In many systems, however, AMM and DMM are explicitly presented as simplified descriptions.

Landauer-type formulations recast interfacial conductance as a spectral transmission problem. For phonon-mediated transport, the integrand combines mode density, transmission, and the derivative of the Bose occupation. Variants of this form appear in 2D-crystal/substrate theory, DMM calculations for $q$8, and quantum-limit analyses across dimensionalities (Ong, 2017); (Deng et al., 2019); (Ho et al., 13 Oct 2025). In multilayer graphene, the formalism predicts that TBR decreases with the number of layers because higher flexural branches open additional transmission channels; at low temperature, the theory gives $q$9 in few-layer graphene (Ong, 2017). This directly contradicts the common shorthand that TBR is only a property of a geometric interface independent of the adjoining film thickness.

Metal–nonmetal interfaces add another channel structure. A two-temperature treatment of the metal yields three concurrent energy-transfer paths: interfacial phonon–phonon coupling, interfacial electron–phonon coupling, and bulk electron–phonon coupling within the metal followed by phonon–phonon transfer across the boundary. These reduce to an equivalent series–parallel thermal resistor network with

$G=\frac{1}{R_{\mathrm{int}}},$0

where $G=\frac{1}{R_{\mathrm{int}}},$1, $G=\frac{1}{R_{\mathrm{int}}},$2, $G=\frac{1}{R_{\mathrm{int}}},$3, and $G=\frac{1}{R_{\mathrm{int}}},$4 (Li et al., 2014). In Pb–diamond, the interfacial electron–phonon branch dominates; in Ti–diamond both branches matter; in TiN–MgO the phonon branch is reduced by the bulk electron–phonon penalty (Li et al., 2014).

Several recent works move beyond bulk-only mismatch pictures by treating the interface itself as a scatterer with its own structure. A dislocation-grid theory for twist boundaries and semicoherent heterointerfaces separates specular acoustic mismatch from diffractive scattering by interfacial dislocation strain fields. Within that framework, misfit dislocation strain fields approximately double the TBR of Si–Ge heterointerfaces relative to acoustic mismatch alone, while in low-angle Si–Si twist boundaries the dislocation-strain contribution dominates and the AM contribution is only about $G=\frac{1}{R_{\mathrm{int}}},$5–$G=\frac{1}{R_{\mathrm{int}}},$6 of the total (Gurunathan et al., 2021). A different subtlety appears in SrRuO$G=\frac{1}{R_{\mathrm{int}}},$7, where the observed exponential cooling is attributed to an effective boundary resistance arising from magnon–phonon disequilibrium inside the film rather than from imperfect phonon transmission across the physical interface (Langner et al., 2010). This suggests that “boundary resistance” can also be an emergent description of coupled nonequilibrium subsystems.

3. Measurement and inversion methodologies

Because TBR is inferred rather than measured directly, the extraction protocol is part of the phenomenon’s practical definition. Time-domain thermoreflectance (TDTR) remains a standard method, but sensitivity to buried interfaces depends strongly on modulation frequency, thermal penetration depth, and model parameter degeneracy. For nanocrystalline diamond/AlGaN, a hybrid TDTR/SSTR procedure was developed in which TDTR at $G=\frac{1}{R_{\mathrm{int}}},$8 is combined with steady-state thermoreflectance down to $G=\frac{1}{R_{\mathrm{int}}},$9. The two datasets are fit simultaneously to the same multilayer heat-diffusion model, extracting $\mathrm{W\,m^{-2}\,K^{-1}}$0, $\mathrm{W\,m^{-2}\,K^{-1}}$1, and $\mathrm{W\,m^{-2}\,K^{-1}}$2 while minimizing the maximum residual. A $\mathrm{W\,m^{-2}\,K^{-1}}$3 residual threshold defines acceptable fits, and the intersection of TDTR and SSTR solution spaces removes the “limitless contour uncertainty” obtained when either technique is used alone (Aller et al., 2024).

Electrical methods occupy a complementary regime. In the $\mathrm{W\,m^{-2}\,K^{-1}}$4 method applied to $\mathrm{W\,m^{-2}\,K^{-1}}$5-SiC/$\mathrm{W\,m^{-2}\,K^{-1}}$6, the apparent film thermal resistance is plotted against oxide thickness and fit as

$\mathrm{W\,m^{-2}\,K^{-1}}$7

so the intercept yields the interfacial resistance (Deng et al., 2019). A more elaborate three-sensor $\mathrm{W\,m^{-2}\,K^{-1}}$8–$\mathrm{W\,m^{-2}\,K^{-1}}$9 method uses two unequal-width heaters and a central detector, with full 3D finite-element fitting, to extract the film conductivity, substrate conductivity, and film–substrate TBR from a single heterostructure without reference samples. Applied to GaN/SiC and GaN/Si with $\mathrm{m^2\,K/W}$0 GaN, it yielded $\mathrm{m^2\,K/W}$1 for GaN/SiC and $\mathrm{m^2\,K/W}$2 for GaN/Si (Yang et al., 2022).

Transient reflectivity and nanocalorimetry impose further model requirements. For Al/Al$\mathrm{m^2\,K/W}$3O$\mathrm{m^2\,K/W}$4, a multiscale workflow combines non-equilibrium molecular dynamics, a two-temperature model, and finite-element simulation of thermo-reflectance. The room-temperature TBR is $\mathrm{m^2\,K/W}$5, but the paper shows that fitting transient reflectivity with a lumped thermal capacitance model can overestimate the TBR by about a factor of two because the probe senses near-surface temperatures whereas the TBR is defined by interfacial temperatures (Caddeo et al., 2016). At the nanoscale, all-optical nanocalorimetry of glass-embedded Ag particles treats the particle as a lumped thermal mass coupled to a diffusive matrix through an interfacial conductance, and time-resolved magneto-optical Kerr measurements in SrRuO$\mathrm{m^2\,K/W}$6 use the magnetic order parameter as a thermometer to reveal effective TBR-like behavior (Banfi et al., 2012); (Langner et al., 2010).

Atomistic and mesoscopic simulation have become integral to validation. Approach-to-equilibrium MD was used to extract a length-independent bulk TBR of $\mathrm{m^2\,K/W}$7 for a sharp Si/Ge interface for heat flow from Si to Ge, after extrapolating away finite-size effects (Hahn et al., 2015). For a model Si/heavy-Si interface, phonon NEGF with Büttiker probes was calibrated to bulk Si and then shown to match MD TBR over mass ratios from $\mathrm{m^2\,K/W}$8 to $\mathrm{m^2\,K/W}$9, while avoiding the artificial resistances that equilibrium Landauer treatments can assign to virtual interfaces in homogeneous systems (Chu et al., 2019).

4. Interfacial structure, chemistry, and bonding

The dominant recent trend is that interfacial structure and chemistry can outweigh bulk material identity. In nanocrystalline diamond/AlGaN, direct chemical-vapor-deposition growth on AlGaN produces a disordered interface with TBR $\mathrm{m^2\,K/GW}$0 in the abstract and $\mathrm{m^2\,K/GW}$1 in the body text. Introducing sputtered amorphous carbide interlayers lowers the resistance to record values of $\mathrm{m^2\,K/GW}$2 for $\mathrm{m^2\,K/GW}$3 and $\mathrm{m^2\,K/GW}$4 for SiC on Al$\mathrm{m^2\,K/GW}$5Ga$\mathrm{m^2\,K/GW}$6N. STEM shows amorphous interlayers of $\mathrm{m^2\,K/GW}$7–$\mathrm{m^2\,K/GW}$8, while FFTs show that the AlGaN remains crystalline, indicating protection from hydrogen-plasma damage during diamond growth (Aller et al., 2024). The proposed mechanisms combine improved adhesion, suppression of voids and contamination, and vibrational bridging by intermediate PDOS.

A second semiconductor-on-diamond example points in the opposite direction for thicker amorphous layers. In bonded GaN/diamond made by hybrid SiO$\mathrm{m^2\,K/GW}$9–Ar surface-activated bonding, a $t_i$0 heterogeneous amorphous interfacial layer yields $t_i$1, while increasing the thickness to $t_i$2 raises the TBR to about $t_i$3. The paper attributes the strong thickness sensitivity not to simple slab resistance but to the local vibrational structure of the diamond–SiO$t_i$4 interdiffusion region, whose vDOS extends to $t_i$5 and loses low-frequency overlap as the interdiffusion region thickens (Xu et al., 2024). This is a direct counterexample to the misconception that any intermediate amorphous layer necessarily “bridges” mismatch.

Reactive interface formation can also raise TBR. ReaxFF MD for Si/SiO$t_i$6 gives $t_i$7 for the non-reactive interface at $t_i$8–$t_i$9, but heating to induce reaction forms a stable amorphous interlayer and raises the low-temperature value to $k_i$0 upon cooling (Heijmans et al., 2019). In PECVD $k_i$1 on $k_i$2-SiC, TEM shows a $k_i$3–$k_i$4 disordered near-surface SiC region and interfacial defects, and the measured room-temperature TBR is $k_i$5, about four times higher than the Si/SiO$k_i$6 control on the same platform (Deng et al., 2019).

Bonding strength itself is not monotonic in its thermal consequences. In metal/WSe$k_i$7/Al$k_i$8O$k_i$9 stacks, stronger metal–WSe$j$0 bonding from Al to Au to Ti improves the monolayer interface, so $j$1 increases and the referenced monolayer TBR decreases. However, for bilayer WSe$j$2, the additional WSe$j$3–WSe$j$4 resistance rises from $j$5 for Al to $j$6 for Ti because metallization perturbs the phonon DOS of the first layer and increases mismatch with the second (Choi et al., 2018). This suggests that “stronger bonding lowers TBR” is only locally true; in multilayer or composite stacks, improving one junction can worsen another.

5. Temperature, thickness, and dimensionality effects

Temperature dependence is often strong and mechanism-specific. For glass-embedded Ag nanoparticles of radius $j$7, all-optical nanocalorimetry shows that the Kapitza resistivity rises from $j$8 at $j$9 to $\Delta T$00 at $\Delta T$01, while the conductance drops from $\Delta T$02 to $\Delta T$03. The paper interprets the trend as approximately following $\Delta T$04 for the metal nanoparticle (Banfi et al., 2012). In cryogenic Cu films, the interfacial conductance follows the low-temperature form $\Delta T$05; measured $\Delta T$06 is $\Delta T$07 for Cu$\Delta T$08 and $\Delta T$09 when a $\Delta T$10 Ti interlayer is inserted, increasing the TBR by about fourfold (Wang et al., 2019). For plasmonically heated Au nanowires below $\Delta T$11, FEM with

$\Delta T$12

reproduces the observed nonlinear heating, and switching from $\Delta T$13 to sapphire or quartz reduces the local temperature increase by about $\Delta T$14 (Zolotavin et al., 2017).

Thickness and dimensionality modify TBR even when the physical interface is unchanged. The theory for multilayer graphene on SiO$\Delta T$15 predicts that TBR decreases with increasing layer number because additional flexural branches contribute to transmission, with asymptotic convergence for large $\Delta T$16 (Ong, 2017). In GaN/AlN/4H-SiC HEMT stacks, TBR at AlN/4H-SiC decreases substantially with increasing temperature and saturates above $\Delta T$17; the measured room-temperature value is $\Delta T$18, while the calculated GaN/AlN value is $\Delta T$19 (Tran et al., 13 Oct 2025). In the same work, the channel and buffer conductivities are strongly thickness dependent: at $\Delta T$20, $\Delta T$21 increases from $\Delta T$22 at $\Delta T$23 to $\Delta T$24 at $\Delta T$25 (Tran et al., 13 Oct 2025). This means that a single interfacial TBR value can have very different device impact depending on how the adjoining layer thickness changes the local heat flux.

The dimensionality of the transport channels also matters at the level of theoretical bounds. A generalized quantum-limit analysis gives a single expression for the ballistic thermal boundary conductance of acoustic branches in $\Delta T$26 dimensions,

$\Delta T$27

recovering the universal one-dimensional thermal conductance quantum $\Delta T$28 and the three-dimensional low-temperature Kapitza law $\Delta T$29 (Ho et al., 13 Oct 2025). This does not describe real interfaces directly, but it provides a lower bound on TBR and a benchmark for how far experimental systems remain from unity transmission.

6. Device relevance, representative magnitudes, and open questions

TBR is a first-order design variable in high-power electronics. In diamond/AlGaN, reducing $\Delta T$30 from $\Delta T$31 to $\Delta T$32–$\Delta T$33 nearly removes an order of magnitude from the series thermal resistance between the hot junction and the diamond heat spreader, which the paper connects to lower junction temperatures and improved RF power density and reliability in AlGaN HEMTs and RF devices (Aller et al., 2024). In GaN/AlN/4H-SiC HEMTs, TCAD simulations using measured thermal metrics show that increasing the AlN buffer thickness to $\Delta T$34 reduces the peak hotspot temperature by $\Delta T$35, while increasing the GaN channel thickness to $\Delta T$36 reduces it by $\Delta T$37. In the thin stack, the local interfacial temperature drops are $\Delta T$38 at GaN/AlN and $\Delta T$39 at AlN/4H-SiC (Tran et al., 13 Oct 2025).

In cryogenic and nanoscale calorimetric devices, TBR governs operating regime. For copper films, the crossover between electron–phonon-limited and boundary-limited relaxation occurs when

$\Delta T$40

so thickness and interface engineering can place a device in the desired regime (Wang et al., 2019). In plasmonic nanowires, TBR becomes the major constraint on reaching cryogenic local temperatures under optical drive (Zolotavin et al., 2017). In photoacoustic nanotransducers, a high TBR can be useful rather than harmful: for water-immersed gold nanocylinders, the ratio $\Delta T$41 controls the competition between thermophone and mechanophone launching, and a graphene-functionalized Au–water interface with $\Delta T$42 drives mechanophone dominance across the explored pulse durations while keeping the liquid temperature nearly unchanged (Diego et al., 2024). In ferroelectric PbTiO$\Delta T$43, a single $\Delta T$44 domain wall raises the thermal resistance by about $\Delta T$45 in the simulated geometry, and two walls can raise it to about $\Delta T$46, supporting an electrically actuated phononic switch (Seijas-Bellido et al., 2017).

Across the literature, several open questions recur. Device-level electrical consequences of interface-engineering layers remain unresolved in AlGaN-based HEMTs, including effects on surface states, polarization fields, and contact resistance (Aller et al., 2024). Long-term thermal cycling, adhesion, and wafer-scale uniformity remain open for ultrathin carbide and oxide-based bonded interfaces (Aller et al., 2024). The temperature dependence of engineered TBR, especially the distinction between elastic and inelastic interfacial channels, is still actively pursued through atomistic Green’s-function calculations, phonon spectroscopy, and in-device thermometry (Aller et al., 2024); (Tran et al., 13 Oct 2025). A plausible implication is that future TBR optimization will rely less on selecting “high-$\Delta T$47 materials” in isolation and more on co-designing local chemistry, interdiffusion, vibrational spectra, and metrology so that the inferred interfacial resistance corresponds to the physically relevant thermal bottleneck.