Skip to content

Imaging Heat Transport in Suspended Diamond Nanostructures with Integrated Spin Defect Thermometers

FieldValue
JournalPhysical Review Letters
Published2026-02-24
AuthorsValentin Goblot, Kexin Wu, Enrico Di Lucente, Yuchun Zhu, Elena Losero
DOI10.1103/3s96-7ghm
Citations4

Suspended single-crystal diamond cantilevers as narrow as 0.8 µm showed sub-ballistic heat transport, with a size-dependent effective thermal conductivity matched by first-principles modeling. Integrated nitrogen-vacancy thermometers mapped the temperature profiles that exposed this size effect. Diamond’s room-temperature conductivity exceeds 2000 W/m/K, so the cantilevers test the persistence of bulk phonon transport.

Measurements covered 10 µm-long cantilevers with widths of 0.8 to 2.5 µm and cross-sectional areas of about 0.2 to 2.6 µm². At widths below 2 µm, the normalized temperature gradient departed from the width dependence expected for ordinary diffusion.

The scaling fell between Fourier’s (w^{-2}) law and the (w^{-3}) law for boundary-dominated ballistic transport. Effective conductivity fell as cantilever width decreased, but less strongly than in the boundary-only limit.

Prior first-principles work predicted conductivity near half the bulk value at a width of about 0.7 µm. The experiment reached 0.8 µm but did not directly test that predicted width.

The devices used commercial (100) CVD single-crystal diamond enriched above 99.95% in ¹²C. The transport substrate carried a 1 to 2 ppm NV-rich surface layer about 3 µm deep. Samples B and C shared a reported NV concentration. The reported device series included NV concentrations up to 3 ppm.

Electron-beam lithography defined the structures, followed by vertical oxygen reactive-ion etching and focused-ion-beam milling. The final FIB cut used a 53° angle and 30 kV Ga⁺ ions, producing triangular cross-sections. Beam current fell from 1.5 nA for bulk removal to 700 pA for smoother sidewalls.

A 150 nm chromium patch at the tip absorbed the focused 515 nm heating laser, while the bulk diamond served as the heat sink. The suspended structure acted as a one-dimensional heat channel between the patch and the bulk diamond.

Each readout spot addressed about (10^5) negatively charged nitrogen-vacancy centers through continuous-wave optically detected magnetic resonance. The measured zero-field splitting shifted with temperature, so the authors tracked (\Delta D(T)=D(T)-D(22^\circ\mathrm{C})). Absolute (D) varied with position, possibly from local strain, but (\Delta D) was position-independent. A two-parameter fit to the zero-field splitting converted resonance shifts into local temperature.

The lock-in readout achieved a typical sensitivity of 0.27 K/√Hz. At a 3 s integration time, the temperature readout error was below 0.15 K. Standard deviations of power-scan residuals were ±0.25 K, including alignment errors in the heating beam. Measurements used a 0.15 mW readout laser and 41 dBm microwave power. The microwave frequency had a 5 kHz sinusoidal modulation and a 5 MHz modulation depth.

No static magnetic field was applied, making the zero-field-splitting readout insensitive to small magnetic-field fluctuations. FIB-milled cantilevers showed about 1 K of readout-laser heating, attributed to gallium contamination. Subtracting measurements without the heating laser removed that contribution.

Heating power and readout position were controlled independently. The local temperature rise was linear with heating power at each tested position. For a 1.53 µm-wide cantilever, a 0.73 µm narrow section occupied the 4 to 10 µm region beyond the chromium patch. At 3.0 mW incident heating power, the section from 0 to 4 µm had a gentler gradient than the narrow section. The free-hanging section from -2 to 0 µm remained at constant temperature. Each constant-width section produced a linear temperature profile, as expected for a constant local heat flux.

The optical readout had a diffraction-limited spot about 0.44 µm in diameter. A separate patch check used a gold patch with a few-nanometer chromium adhesion layer and 532 nm excitation. The separate measurement found the metal patch at or below 100°C under heating.

Estimated upper bounds were 1.5 × 10^-9 W for radiative loss and 4 × 10^-9 W for convective loss. Both estimates were 5 to 6 orders of magnitude below the applied laser power.

The width study fixed each cantilever length at 10 µm and adjusted heating power so maximum temperature rises stayed below 40 K. Temperature gradients were normalized by absorbed power (P_{\mathrm{abs}}=\alpha P_h). Heat flux followed (q=\alpha P_h/S), with (S=\tan(\theta)w^2/4) and (\alpha) the chromium absorption coefficient.

For constant absorbed power, Fourier’s law requires the normalized gradient to scale as (w^{-2}). Boundary-dominated ballistic transport instead gives (w^{-3}). The measured points departed from (w^{-2}) below 2 µm and remained between the two limits. This result shows that a linear profile alone cannot establish Fourier transport.

The regime is “apparent diffusion”: the profile resembles diffusion while conductivity depends on sample size. FIB-milled and non-FIB cantilevers followed the same trend, indicating negligible transport effects from FIB surface damage. Earlier nanoscale thermal studies centered on graphene and graphite, while low-dimensional single-crystal diamond data were lacking.

First-principles modeling reproduced the measured effective conductivity with no fitted parameter. A full linearized phonon Boltzmann transport equation solution was too costly for repeated geometry calculations. The authors therefore used viscous heat equations for temperature and phonon drift velocity.

Parameter calculations supplied third-order anharmonic phonon-phonon scattering, impurity scattering, and rough-boundary effects. Boundary scattering used a Casimir length equal to one-half of the smallest system dimension. Each simulation imposed a 10 K temperature difference over 10 µm around room temperature.

Full viscous solutions differed only slightly from an apparent-diffusive calculation across the width range. Ballistic effects therefore supplied most of the conductivity reduction, while viscous hydrodynamic effects remained minor. A separate vacancy sensitivity test covered 0, 10, and 20 ppm. The 10 ppm case matched best and was considered a reasonable estimate for the sample. Sub-ballistic transport results from intrinsic normal and Umklapp scattering acting with extrinsic boundary scattering.

The lower width limit came from fewer NV centers under the diffraction-limited spot, which reduced temperature sensitivity. The upper limit came from difficulty fabricating suspended structures above 3 µm because etching took longer. Width uncertainty was set conservatively at 50 nm and dominated the error in effective conductivity.

Heat-flow estimates also depended on the chromium absorption coefficient, calculated as one minus reflectance with transmission and scattering neglected. Cross-section estimates came from SEM images, so the reported conductivities should be read with geometry uncertainty rather than as material constants.

The results show that bulk diamond conductivity cannot be assigned directly to submicrometer diamond channels. For device engineering, this means that models should include geometry-dependent boundary scattering together with intrinsic phonon scattering. Engineers selecting diamond for suspended structures should therefore check defect profile and geometry alongside bulk conductivity.

The authors suggest extending the method to heat vortices, thermal rectification, sidewall roughness, surface termination, and quantized phonon conductance. These are proposed directions, not measurements reported in this paper.

The heat-transport imaging device requires single-crystal diamond containing NV thermometers, with ¹²C enrichment above 99.95% and patterning into suspended cantilevers with a chromium absorber. The diamond materials catalog lists MPCVD-grown single-crystal diamond plates and wafers up to 125 mm. Custom fabrication questions can be directed to engineering support, but the listed specifications do not establish the required ¹²C enrichment, NV concentration, or chromium patch.

View Original Abstract

Among all materials, monocrystalline diamond has one of the highest measured thermal conductivities, with values above 2000 W/m/K at room temperature. Diamond nanostructures are increasingly used in photonics, electronics, and quantum technologies, where heat dissipation is critical. However, predicting nanoscale heat flow is made difficult by the predominance of momentum-conserving “normal” phonon-phonon scattering processes. Here, we use dilute nitrogen-vacancy color centers as in situ, highly precise spin defect thermometers to image temperature inhomogeneities in single-crystal diamond microstructures heated from ambient conditions. We analyze cantilevers with cross sections in the range from about 0.2 to 2.6 μm^{2}, observing a strong reduction in the cantilevers’ conductivity as width decreases. We use first-principles simulations based on the linearized phonon Boltzmann transport equation and viscous heat equations to quantitatively predict the cantilevers’ thermal transport properties, rationalizing how the interplay between intrinsic and extrinsic phonon scattering mechanisms determines the observed nondiffusive behavior. Our temperature-imaging method paves the way for the exploration of unconventional, nondiffusive heat-transport phenomena in devices and nanostructures of arbitrary geometries.