Skip to content

Scalable fabrication of diamond-on-silica heterostructures via high-selectivity deep ICP-RIE and room-temperature bonding

FieldValue
JournalDiamond and Related Materials
Published2026-07-24
AuthorsR. Chembra Vasudevan, A. Hammouti, J. Le Pouliquen, T. Batte, P. Pirasteh
DOI10.1016/j.diamond.2026.113992

A 2 µm SiO2 hard mask enabled a complete through-etch of a 16 µm NV-doped single-crystal diamond membrane at 15:1 diamond-to-SiO2 selectivity. The etched pieces were transferred onto SiO2 and bonded at room temperature without detectable parasitic photoluminescence. The measured coherence time also matched the supplier’s value for the starting diamond.

The process combines deep dielectric-mask etching with hydroxide-catalysed bonding at room temperature. It leaves a 1.2 µm SiO2 layer on diamond that serves as the optical interface and bonding surface. The authors present partitioning of a commercial membrane as a way to reduce material cost per device.

The bonded material combines single-crystal diamond with silica. The authors identify integrated quantum photonics and sensing as the main target applications.

The starting piece was a commercially sourced 1.5 x 1.5 mm², 16 µm single-crystal diamond membrane. It contained isotopically pure carbon and approximately 1 ppm NV centers.

Before patterning, the authors cleaned it in 3:1 H2SO4:H2O2 at 75 °C for 45 minutes. The membrane was then rinsed with acetone and isopropanol and dried under nitrogen.

A 2 µm SiO2 hard mask was deposited by PECVD at 280 °C. The SiH4:N2O flow was 40/1000 sccm. The oxide stress was 100 MPa when deposited on silicon and measured by profilometry. The authors could not measure stress on diamond because larger diamond substrates were unavailable.

For handling, the coated membrane was fixed to a silicon carrier with a 1 to 2 µm QuickStick 135 adhesive layer. Thermal contact would later prove important to etch-rate consistency.

UV lithography in S1813 defined the etch pattern. CHF3 transferred the pattern into the oxide. SF6/O2 removed silica residue before the diamond etch. Pure O2 then etched the diamond.

StepGas and flowPressureRF and ICP powerResult
Oxide pattern transferCHF3, 20 sccm5 mTorr25 W RF, 400 W ICPSiO2 rate 177 nm/min; selectivity 5.2:1 to S1813
Residue removalSF6/O2, 25/5 sccm20 mTorr100 W RF, 0 W ICPClears oxide residue before diamond etching
Diamond etchO2, 20 sccm5 mTorr100 W RF, 400 W ICPDiamond rate 166 nm/min; mask erosion 11.3 nm/min; selectivity 15:1

The corresponding RF biases were 44 V, 280 V, and 185 V. The diamond etch ran in 10-minute cycles to limit heating during the full 16 µm depth. The platen was held at 15 °C with helium backside cooling at 20 mTorr.

Across the initial test samples, diamond etch rate ranged from 106 to 220 nm/min. Diamond-to-SiO2 selectivity ranged from 10:1 to 19.5:1. The authors attribute this variation to uneven thermal contact, which changed the effective diamond temperature.

SEM images showed complete penetration, defined sidewalls, and negligible micromasking. Analysis of multiple sidewalls gave an angle of 21.6 ± 1.8°.

AFM measured an RMS sidewall roughness of 3.1 ± 0.2 nm. The bare diamond surface measured 3.94 ± 0.04 nm. The etched sidewalls were therefore slightly smoother than the starting surface at the few-nanometre scale.

After etching, the remaining 1.2 µm SiO2 cap had an RMS roughness of 0.67 ± 0.15 nm. Its roughness before etching was 2.4 ± 0.1 nm. This cap became the interface for transfer and bonding.

A micromanipulator released the etched membrane from the carrier after the QuickStick adhesive was softened at 90 °C. A vacuum tool then placed it on a PDMS stamp with 2.2 MPa stiffness.

The host was silicon with a 300 nm PECVD SiO2 layer. Both silica surfaces received O2 plasma before bonding. The substrate received 50 W RF at 100 mTorr and 100 sccm O2 for 10 minutes. The membrane cap received identical conditions for 1 minute. This shorter exposure was chosen to protect the PDMS stamp and thin membrane.

The sodium silicate had a modulus of 3 to 4 and was diluted 1:6 by volume in deionized water. The solution was spin-coated onto the activated host. The stamp carrying the membrane was contacted for about 1 minute and then peeled away.

The assembly remained in a desiccator at room temperature for four days. This period allowed the silicate layer to dehydrate and polycondense.

Hydroxide first etched silica and released silicate ions. Below pH 11, silicate formed Si(OH)4. The molecules then polymerized into a three-dimensional siloxane network as water evaporated.

Cross-sectional SEM showed a uniform bonding layer a few dozen nanometers thick. All four bonding attempts produced bonded membranes.

All bonded membranes survived the authors’ 75 °C, 45-minute piranha cleaning. Some also survived HF exposure for up to 20 seconds, while others were exposed to SF6 plasma for up to 30 minutes.

Two of the four structures had voids localized at one corner. The authors believe better micromanipulation during placement could avoid these voids.

PL measurements used about 30 mW of 532 nm laser light. The focused spot size was near 50 µm, and the Rayleigh range exceeded 1 mm. After a 1-second exposure, the bonded spectrum from 550 to 900 nm matched the bare diamond spectrum in normalized intensity and signal-to-noise ratio.

The zero-phonon lines and phonon sidebands remained resolved for both NV charge states. No extra fluorescence from the sodium silicate layer, contaminants, or interfacial defects appeared in the NV emission range. No detectable change in NV charge state was found.

ODMR at about 7 mT produced eight resonances with contrast close to 1%. This spectrum was consistent with an NV ensemble distributed across the four possible crystalline orientations. Each frequency point used 3 ms integration and was averaged over 100 scans.

At about 2 mT and very low microwave power, the hyperfine spectrum gave a Gaussian FWHM of 0.49 ± 0.01 MHz. This measurement used 3 ms integration and 250-scan averaging. The resulting inhomogeneous dephasing time, T2*, was 1.07 ± 0.03 µs. This was close to the supplier’s 1 µs rating for the starting diamond.

The results indicate that etching and bonding did not measurably degrade NV optical readout or inhomogeneous dephasing.

Toros et al. previously reached etch depths up to 150 µm with a 7 µm SiO2 mask and 50:1 selectivity. Corazza et al. reported depths exceeding 45 µm with 10 to 22 µm masks. This study used a 2 µm mask for a 16 µm etch and 15:1 selectivity.

The reported result therefore has a lower depth and selectivity than those dielectric-mask studies. It uses a thinner mask and adds a bonded optical interface. Metallic masks have reported selectivities of 11:1 for aluminum and 75:1 for nickel under SF6. The authors selected silica to reduce the redeposition and contamination risks associated with metallic masks.

The evidence remains a proof of concept. The etch recipe had been tested on a few diamond samples before use on the NV-doped membrane. The authors describe it as qualitatively reliable and reproducible, although etch rate and selectivity varied with thermal contact.

Only four membranes were bonded, and two contained corner voids. The paper reports no diamond-to-silica shear-strength value. Prior hydroxide-catalysed bonds exceeded 2 MPa in other material systems, but that figure does not establish the strength of these diamond interfaces.

The oxide stress value could not be checked on the diamond substrate. The authors also identify interface-scattering and paramagnetic-defect tests as future work. Laser excitation from the silica side would test scattering. Sub-micron NV structures could test magnetic noise near the bonding interface.

For quantum photonics, diamond’s visible refractive index of about 2.4 and silica’s index of about 1.45 provide optical confinement. The proposed next steps are sub-micron diamond thinning and photonic patterning. The bonded platform could support compact photonic circuits, high-Q resonators, fiber-tip endoscopes, or radiation detectors. These devices remain future applications rather than demonstrated outputs of this study.

For NV-based sensing, the paper needs a 16 µm NV-doped single-crystal diamond membrane with a retained SiO2 surface. 6CCVD’s diamond materials catalog lists MPCVD-grown single-crystal diamond from 0.1 to 500 micrometers thick, which covers the demonstrated thickness but does not state an NV concentration. Engineering support can be used to check the required membrane grade, geometry, and bonding sequence against that material range.

View Original Abstract

Single-crystal diamond is a leading material platform for high-power electronics and solid-state quantum technologies, yet many device architectures require micrometer-scale membranes with deeply etched features, patterned from commercially available substrates. In this work, we demonstrate a complete through-etch of a 16 μ m -thick NV-doped single-crystal diamond membrane using a single-layer SiO 2 hard mask combined with a multi-step oxygen-based ICP-RIE process. With a diamond-to-SiO 2 selectivity of 15:1, this non-metallic mask strategy can achieve etch depths of few tens of μ m with well-defined sidewalls, conserved surface roughness and negligible micromasking. Furthermore, we use the oxide layer that remains after etching to serve as the bonding surface in a subsequent integration step. The etched microstructures are transferred onto SiO 2 substrates and bonded at room temperature using O 2 plasma surface activation and a sodium silicate interlayer. The resulting siloxane film is optically transparent across the visible spectrum and introduces no detectable parasitic photoluminescence, preserving the optical readout of the embedded NV centers. Together, this deep-etch and room-temperature bonding process provides a scalable and contamination-free route from bulk diamond membranes to diamond-on-silica heterostructures for integrated quantum photonics and sensing applications.

  1. 2022 - A review of ultrawide bandgap materials: Properties, synthesis and devices
  2. 1993 - Optical properties of diamond
  3. 1997 - Scanning confocal optical microscopy and magnetic resonance on single defect centers [Crossref]
  4. 2022 - Diamond integrated quantum nanophotonics: Spins, photons and phonons [Crossref]
  5. 2020 - Magnetometer with nitrogen-vacancy center in a bulk diamond for detecting magnetic nanoparticles in biomedical applications [Crossref]
  6. 2022 - High-precision robust monitoring of charge/discharge current over a wide dynamic range for electric vehicle batteries using diamond quantum sensors [Crossref]
  7. 2025 - High-resolution non-destructive detection of grinding burns with NV diamond quantum magnetometer [Crossref]
  8. 2025 - Endoscopic diamond magnetometer for cancer surgery [Crossref]
  9. 2016 - Scalable fabrication of high-quality, ultra-thin single crystal diamond membrane windows [Crossref]
  10. 2023 - Diamond MEMS: From classical to quantum [Crossref]