Skip to content

Thermal Transport in Atomic Layer Deposition-Grown Beryllium Oxide Thin Films and Implications on Device Cooling

FieldValue
JournalACS Applied Materials & Interfaces
Published2026-10-20
AuthorsHusam M. Walwil, Jonghyun Bae, Seokjun Kim, Dohwan Jung, Nipun Sutradhar
DOI10.1021/acsami.6c16056

PEALD-grown polycrystalline BeO films reached 13.45 ± 1.45 W m^-1 K^-1 at 250 nm, while the BeO/Si interface measured about 200 MW m^-2 K^-1 at room temperature. Device simulation predicted that BeO beneath a top-side polycrystalline diamond heat spreader would lower peak channel temperature rise by an additional 5.2% at 6 W mm^-1 relative to SiO2. The 5.2% value is a simulation result, not a measured device improvement.

Out-of-plane thermal conductivity increased with BeO thickness. The 250 nm film reached 13.45 ± 1.45 W m^-1 K^-1. The 113, 53, and 26 nm films measured 8.7 ± 1.35, 7.2 ± 2.1, and 5 ± 2.2 W m^-1 K^-1. Even the 26 nm film exceeded SiO2 and SiN films with greater thicknesses in the paper’s literature comparison. The 250 nm result agreed with a prior estimate of about 15.6 W m^-1 K^-1 for a 200 nm BeO film. The authors place this performance an order of magnitude above conventional SiO2 and SiN dielectrics.

BeO thickness (nm)Out-of-plane conductivity (W m^-1 K^-1)BeO (0002) rocking-curve FWHM (degree)
265 ± 2.228.17
537.2 ± 2.115.12
1138.7 ± 1.3512.13
25013.45 ± 1.4511.26

The wider rocking curve in thinner films indicates a broader tilt distribution and stronger grain-boundary scattering. Film thickness therefore affects both phonon transport and crystal orientation.

BeO was deposited by plasma-enhanced ALD on boron-doped p-type Si (100) wafers with 1 to 10 Ω·cm resistivity. The process also used AlGaN/GaN-on-Si (111) epitaxial wafers for diamond integration.

Each substrate was ultrasonically cleaned for 5 min each in acetone, isopropyl alcohol, and deionized water. A 1 min buffered oxide etchant immersion removed native oxide. The PEALD reactor operated at 13.56 MHz. Diethyl beryllium was held at 40 °C, while O2 plasma used 50 sccm and 100 W. Argon purge was 100 sccm at 0.25 Torr. The substrate remained at 150 °C, and growth per cycle was 1.5 Å.

Cycles of 170, 350, 750, and 1670 produced films 26, 53, 113, and 250 nm thick. Cross-sectional FIB-SEM confirmed these thicknesses.

XRD showed c-axis-textured polycrystalline wurtzite. A nonbasal orientation dominated the 26 and 53 nm films, while (0002) dominated the 113 and 250 nm films. The phi scan lacked the regular sixfold symmetry expected for single-crystalline BeO, supporting polycrystallinity rather than epitaxy.

Separating conductivity from interface conductance

Section titled “Separating conductivity from interface conductance”

An approximately 85 nm Al transducer supported time-domain thermoreflectance measurements. Pump and probe spots were 28 and 16 μm. Modulation frequencies ranged from 1 to 20 MHz, and pump heating remained below about 5 K.

At 20 MHz, film conductivity and BeO/Si interface conductance remained coupled. Assuming interface conductance G2 between 50 and 500 MW m^-2 K^-1 produced conductivity fits of 12.5 to 16 W m^-1 K^-1. The authors combined measurements at 20, 12, 10, 7, 5, and 3 MHz. Iterative fitting converged at 13.45 W m^-1 K^-1 and G2 of 195 MW m^-2 K^-1.

A dual-frequency ratio at 5 and 20 MHz gave G2 of 197 ± 53 MW m^-2 K^-1. The result did not change when conductivity varied by ±20%. Thinner films used G2 of 197 MW m^-2 K^-1 as a fixed input. One-megahertz effective conductance decreased with thickness for the 26 and 53 nm films. This trend supported diffusive transport and the shared-interface assumption.

The 250 nm film measured 13.45 ± 1.5 W m^-1 K^-1 at 300 K, 12.30 ± 1.2 at 370 K, and 11.20 ± 1.0 at 450 K. Conductivity followed a weak T^-0.45 dependence. The authors associate the decrease with normal and Umklapp phonon scattering. They also identify boundary and grain-boundary scattering as dominant across the temperature range.

The diffuse mismatch model matched the temperature-dependent interface conductance. Its predictions used truncated linear acoustic dispersion and omitted optical phonons. The measured conductance stayed below the elastic radiation limit from 300 to 450 K. These results support predominantly elastic phonon transmission across BeO/Si.

Film conductivity remained well below 370 W m^-1 K^-1 for single-crystalline BeO and 220 to 320 W m^-1 K^-1 for bulk polycrystalline BeO ceramics. The authors attribute this gap to phonon scattering at film boundaries and grain boundaries.

A ~200 nm polycrystalline diamond layer was grown on an AlGaN/GaN-on-Si epitaxial wafer through a 5 nm BeO interlayer. BeO served as the diamond seed and protected device contacts and the two-dimensional electron gas. Its thinness reduced the target thermal resistance between the GaN channel and diamond.

Growth used microwave plasma CVD after an O2 plasma exposure. Conventional H2/CH4 chemistry produced poor carbon deposits below 500 °C, so the process added oxygen to etch sp2-bonded carbon. Staged chemistry first approached the [C] regime, then shifted toward [H] and [O]. The optimized process achieved crystalline diamond at 400 °C with a strong sp3 Raman peak and a 6 to 6.6 cm^-1 FWHM. Its phase purity was close to conventional high-temperature diamond.

STEM showed a disordered 5 nm BeO layer and an atomically abrupt AlGaN/GaN heterojunction after growth. EDS mapped oxygen to the interlayer. EDS could not detect Be directly because of its low atomic number. The diamond contained no detectable O, N, Al, or Ga contamination.

The authors built a 3D COMSOL model of a single-finger AlGaN/GaN-on-Si HEMT. Micro-Raman thermometry with 532 nm excitation validated the baseline model. Measurements used VGS of 2 V and varied VDS to change dissipated power.

The PCD cases were hypothetical. Each assumed a 2 μm diamond layer with conductivity of 638 W m^-1 K^-1. Each also used 10 nm nucleation and 100 nm passivation layers made from the same material.

CasePeak channel temperature rise at 6 W mm^-1 (°C)Reported reduction
No PCD76.4Baseline
PCD with SiO270.9~7.2% versus no PCD
PCD with BeO66.9Additional 5.2% versus PCD with SiO2

Replacing SiO2 with BeO reduced the effective PCD/AlGaN thermal boundary resistance from about 123 to 13 m^2 K GW^-1. The combined nucleation and passivation resistance fell from 122.2 to 12.64 m^2 K GW^-1. The model therefore predicted an additional 5.2% reduction in peak temperature rise compared with the PCD/SiO2 case.

The conductivity values for 26, 53, and 113 nm rely on a shared G2. Identical growth conditions and smooth STEM interfaces support that assumption. The 26 nm uncertainty is ±2.2 W m^-1 K^-1, and ballistic transport may affect films thinner than about 30 nm.

The effective thermal boundary resistance between diamond and the GaN channel was not measured. Roughness on the diamond surface prevented a TDTR measurement. The authors expect process optimization and identify BeO and AlN as possible interlayers. This remains an expectation.

The device result also remains a model prediction. The model assumed a 2 μm PCD layer, while the experiment produced a ~200 nm layer. Interface resistances came from the diffuse mismatch model. The simulated device also used a commercial epitaxial wafer different from the wafer used for diamond growth.

For BeO selection, thickness and film texture govern conductivity, while interface roughness governs integration. The study supports BeO as a thin thermal interlayer, but it does not establish a measured cooling gain for a complete HEMT stack.

The paper’s application requires a ~200 nm polycrystalline diamond heat spreader for an AlGaN/GaN HEMT, integrated through a 5 nm BeO interlayer. 6CCVD offers MPCVD-grown polycrystalline diamond plates and wafers up to 125 mm, with thickness from 0.1 to 500 micrometers, which includes the studied layer. The diamond materials catalog lists polishing below 5 nm for inch-size polycrystalline material and global shipping, while engineering support can support review of the integration constraints left by the rough diamond surface.

View Original Abstract

Thin film beryllium oxide (BeO) is an emerging material that can potentially be used as a high thermal conductivity gate dielectric and/or passivation layer for high power electronic devices. This study presents the measurement of the thermal conductivity and thermal boundary conductance associated with plasma-enhanced atomic layer deposition (PEALD)-grown polycrystalline BeO films on Si substrates with thicknesses ranging from 25 to 250 nm. The measured out-of-plane thermal conductivities of the 250, 113, 53, and 26 nm-thick BeO films are 13.45 ± 1.45, 8.7 ± 1.35, 7.2 ± 2.1, and 5 ± 2.2 W m-1 K-1, respectively, representing an order-of-magnitude improvement over conventional dielectric materials such as SiN and SiO2. The measured thermal boundary conductance across the BeO/Si interface is ∼200 MW m-2 K-1 at room temperature, which approaches the intrinsic interface limit and suggests that interfacial heat transport is predominantly governed by elastic phonon transmission. The growth of a ∼200 nm-thick polycrystalline diamond film on an AlGaN/GaN-on-Si epitaxial wafer using a 5 nm-thick BeO diamond-seeding is demonstrated, where the structural integrity of the underlying AlGaN/GaN heterostructure is fully preserved. Thermal simulations of AlGaN/GaN high electron mobility transistors (HEMTs) integrated with a top-side polycrystalline diamond heat spreader predict that replacing the conventional SiO2 passivation and nucleation layers with BeO helps reduce the channel temperature rise by an additional 5.2%.