Skip to content

Van der Waals β-Ga2O3 thin films on polycrystalline diamond substrates

FieldValue
JournalNature Communications
Published2025-08-31
AuthorsJing Ning, Zhichun Yang, Haidi Wu, X.-Y. Dong, Yaning Zhang
DOI10.1038/s41467-025-63666-x
Citations1

Monolayer graphene enabled mist-CVD growth of a 350 nm VdW-β-Ga2O3 film on polycrystalline diamond. At 760 °C and 600 sccm O2, its (201) rocking-curve FWHM reached 0.18°, with 6.71 nm RMS roughness. A photodetector made from the film reached a photo-to-dark current ratio of 10⁶ and 210 A/W responsivity. The paper reports a 2.82 m² K/GW effective thermal boundary resistance, although detailed TDTR results assign that value to bare diamond rather than the graphene-interlayer sample.

Monolayer graphene screened the lattice mismatch between β-Ga2O3 and randomly oriented diamond grains. Calculations for the [100] surface showed oxygen adsorption energy changing from -1.33 eV to -2.99 eV with graphene and -3.09 eV with h-BN. Both systems had their strongest calculated binding with one insertion layer, so the authors selected monolayer graphene for the experiment.

Growth itemReported condition
SubstrateElement Six TM180 polycrystalline diamond
InterlayerWet-transferred monolayer graphene
PrecursorGallium acetylacetonate above 99.99% purity; 0.02 to 0.08 mol/L; HCl from 1% to 10%
Mist-CVD700 to 800 °C; O2 from 300 to 1000 sccm; diluting Ar at 3000 sccm; 30 to 40 min
Selected condition760 °C, 600 sccm O2, 350 nm film

Graphene was CVD-grown on 25 μm copper foil and wet transferred after copper removal. The diamond was cleaned with acetone, alcohol, and deionized water, then treated in dilute HF for five minutes to remove surface oxides. Its specified thermal conductivity was above 1800 W/m·K at 300 K and above 1500 W/m·K at 425 K, with polished Ra below 20 nm.

A 1.7 MHz nebulizer delivered 25 to 35 mL of precursor solution during mist-CVD. The authors used XRD, TEM, XPS, AFM, SEM, in situ Raman spectroscopy, and TDTR to assess structure, chemistry, morphology, strain, and heat transport.

Crystallographic control and film structure

Section titled “Crystallographic control and film structure”

TEM resolved monolayer graphene and two β-Ga2O3 orientations, with (201) predominant. The graphene carbon spacing was 0.492 nm, close to the theoretical 0.496 nm oxygen spacing of the (201) orientation. Their calculated lattice mismatch was 4.9%. The two tilted planes had about 7% mismatch, above the authors’ 6% theoretical threshold.

XRD confirmed β-phase reflections along the (201), (401), and (202) orientations. At 760 °C and 600 sccm O2, the 350 nm film produced the smallest FWHM of 0.18°. The authors associated this condition with the highest crystallinity and fewest interfacial defects. They cite prior β-Ga2O3 on (111) single-crystal diamond with FWHM of 3° to 4°, compared with 0.18° here.

XPS fitted the O1s signal into lattice oxygen at 530.7 eV and oxygen vacancies at 531.7 eV. The vacancy fraction fell as O2 flow rose. At 300 and 1000 sccm, other gallium compound peaks indicated insufficient oxygen at the low flow and incomplete precursor reaction at the high flow.

AFM showed roughness falling and then rising with temperature. The authors linked this pattern to an island, coalescence, then new island growth cycle. Grain boundaries remained visible at 760 °C. The authors associated them with graphene defects, stress release, or gas flow.

Graphene alleviated thermal expansion stress, but bare diamond produced the lowest measured thermal boundary resistance.

Raman cycling from 0 to 200 °C produced reversible peak shifts. At 200 °C, β-Ga2O3 on diamond reached Δω of 10.8 cm⁻¹. Its shift coefficient was about -0.05 cm⁻¹/K, versus -0.024 cm⁻¹/K for films on Si and sapphire. At 100 °C, Δω was 2.418 cm⁻¹ on Si and 2.413 cm⁻¹ on sapphire. The authors attributed the smaller Si-sapphire difference to their closer thermal expansion coefficients. They interpreted the larger diamond shift as strong van der Waals coupling and stress release. Graphene slightly reduced this stress.

TDTR used 100 nm Al transducers, a 12 μm pump radius, and a 4 μm probe radius, each defined at 1/e² intensity. The modulation frequency was 1 MHz at 25 °C. All three samples received identical β-Ga2O3 epitaxy conditions.

SampleInterface treatmentEffective TBReff
1Graphene on polycrystalline diamond6.08 m² K/GW
2Bare polycrystalline diamond2.82 m² K/GW
3O2-plasma-treated graphene on polycrystalline diamond, 50 W for 30 s6.02 m² K/GW

The 350 nm film had a thermal conductivity of 7.19 W/m·K. Plasma treatment added sparse covalent bonds, but random nucleation lowered film crystallinity. The paper places the 2.82 m² K/GW result about one order below prior bonding and interlayer approaches. It also states that this value meets kW-class device requirements, although no kW device was tested.

Interdigitated electrodes were deposited and etched on the film. Under 0.795 mW/cm² illumination and biases from -3 V to 3 V, dark current was 5.83 × 10⁻⁶ μA and photocurrent was 7.92 μA. These values gave a photo-to-dark current ratio near 10⁶. The illuminated area was 6 × 10⁻⁴ cm², and responsivity was 210 A/W.

Comparison with devices reported during the same period found higher PDCR than the cited devices and higher responsivity at comparable PDCR. A 1 Hz optical pulse test ran for 2000 seconds, with a 54 ms rise time and 4 ms decay time.

Conductivity changed negligibly as optical frequency increased. The authors associated this result with efficient carrier extraction, passivated defects and interface states, and effective heat management. These interpretations were not established by a separate power-device test.

The authors identify limited large-area single-crystal diamond availability and full-wafer monolayer graphene uniformity as scalability constraints. Their solution uses polycrystalline diamond, but full-wafer control of the interlayer remains unresolved. Visible grain boundaries at 760 °C and 6.71 nm RMS roughness also leave room for process improvement.

The reported 7.19 W/m·K film conductivity means heat spreading within the Ga2O3 layer remains a design consideration. The photodetector stability test covered optical cycling, not power-device thermal cycling or breakdown. No high-power transistor or diode was demonstrated.

The mismatch between the abstract’s broad interface label and the detailed TDTR sample assignment also requires clarification. The 2.82 m² K/GW value should not be assigned to graphene-interlayer devices without resolving that attribution.

The results support polycrystalline diamond as a practical heat spreader when surface preparation and interlayer transfer are controlled. Material selection should account for grain structure, surface roughness, thermal conductivity, and interface quality rather than diamond conductivity alone.

Growth control should track O2 flow and temperature because they changed phase, orientation, vacancies, and roughness. Device qualification should measure thermal boundary resistance on the finished stack and include thermal cycling. A conductive substrate alone does not establish performance for a power device.

For Ga2O3 thermal-management devices, the process needs a cleaned polycrystalline diamond heat-spreader substrate and a transferred monolayer graphene interlayer. 6CCVD supplies polycrystalline diamond grown by MPCVD as plates and wafers up to 125 mm through its diamond materials catalog, and offers custom fabrication and engineering support for the diamond material.