Skip to content

Plasma etch process optimization for photonic-grade diamond-on-insulator substrates and thickness evaluation using colorimetry

FieldValue
JournalDiamond and Related Materials
Published2026-11-01
AuthorsTianyin Chena, Alessio Mirandaa, Leyla Ramia, Ryoichi Ishihara, Salahuddin Nur
DOI10.1016/j.diamond.2026.114313

A 10 µm (100) single-crystal diamond plate was thinned to approximately 300 nm over a 0.5 × 0.5 mm² region while preserving the buried diamond-SiO₂ bond. AFM measured a surface roughness Rq below 0.5 nm. The thinned film was then used to fabricate free-standing photonic chiplets, and microscope colorimetry estimated a 304 nm film within ±3 nm of white-light interferometry.

Substrate fabrication and characterization

Section titled “Substrate fabrication and characterization”

Commercial (100) single-crystal diamond was directly bonded to a 300 nm PECVD SiO₂ layer on a silicon handle wafer. The final demonstration began with a 10 µm diamond plate. Hydrophilic bonding formed the buried diamond-SiO₂ interface. The assembly was mounted on a quartz carrier and masked before ICP-RIE thinning.

Dektak stylus profilometry, WLI, Raman spectroscopy, optical microscopy, SEM, and AFM tracked thickness and surface condition. WLI measured the final film profiles, Raman confirmed single-crystal diamond, and AFM measured roughness. SEM inspected micromasking and particles.

Debonding evidence points to thermal stress

Section titled “Debonding evidence points to thermal stress”

The first high-power recipe debonded during both plasma steps. Interference fringes were consistent with an air gap at the diamond-oxide interface. At 10 µbar, Ar/Cl₂ at 500 W ICP and 200 W RF debonded after 30 min. O₂ at 1100 W ICP and 90 W RF also debonded after 30 min at 10 µbar.

Two causes were considered: lateral oxide removal and plasma-induced thermomechanical stress. A 2.5 µm oxide stack remained bonded for 60 min under moderated Ar/O₂ conditions. The same high-power recipe also preserved the bond on fused quartz.

Diamond has a thermal expansion coefficient of about 1.0 to 1.1 ppm K⁻¹, versus 2.6 ppm K⁻¹ for silicon. Quartz lies closer at 0.5 to 0.55 ppm K⁻¹. These tests supported thermomechanical stress as the main cause of debonding on Si-based carriers. The O₂ debonding argued against lateral oxide removal as the sole cause. Bond survival on quartz did not prevent severe micromasking.

One cycle uses 4 min Ar/Cl₂ for cleaning and smoothing, followed by 4 min Ar/O₂ for diamond removal. A laser-cut fused-quartz top mask limits plasma exposure of the surrounding carrier and reduces redeposition-driven micromasking.

StepGas flowPower and chamber conditionsRate and duration
Ar/Cl₂15 Ar, 7 Cl₂ sccm100 W ICP, 250 W RF; 0.01 mbar; 20 °CAbout 10 nm/min; 4 min
Ar/O₂50 Ar, 13 O₂ sccm200 W ICP, 300 W RF; 0.01 mbar; 20 °CAbout 50 to 55 nm/min; 4 min

Repeated cycles thinned bonded plates from the 10 to 20 µm range. The Ar/Cl₂ step removes particles before they develop into nanopillars or pits. The thinning recipe used a quartz mask and did not transfer directly to SiN hard-mask patterning.

WLI horizontal and vertical profiles through the target area each averaged approximately 300 nm. AFM measured Rq below 0.5 nm, and SEM found few visible contaminants or micromasking structures in the target region. Raman spectroscopy confirmed that the remaining film was single-crystal diamond.

The diamond-oxide interface remained intact after thinning. Thickness variation was about 50 to 70 nm across the 0.5 × 0.5 mm² target area. Surrounding bonded diamond remained as thick as approximately 1 µm.

The authors attributed the variation to quartz-mask etch nonuniformity and thickness variation in the starting plate. The film met the roughness target but remained nonuniform. The data do not establish wafer-scale uniformity or device yield.

After acetone and isopropanol cleaning, PECVD SiN and CSAR62 electron-beam resist were applied. Electron-beam lithography and CHF₃/O₂ RIE transferred the chiplet pattern into the SiN hard mask.

Because Ar-containing thinning chemistry redeposited SiN, pattern transfer required a separate cyclic diamond etch. Each cycle used 15 s of O₂ at 50 sccm and 10 µbar, with 200 W ICP and 300 W RF, followed by 5 min cooling. The carrier table was held at 0 °C, and helium backside flow was 10 sccm.

Room-temperature buffered oxide etchant, a 7:1 v/v mixture of 40 wt% NH₄F and 49 wt% HF, removed the SiN mask and undercut SiO₂. Critical point drying released the structures. The finished chiplets contained waveguides held in a support frame, with no residual pedestal beneath them. A probe lifted one chiplet through van der Waals adhesion, although transfer onto another photonic circuit was not tested.

Colorimetry used a four-layer air, diamond, SiO₂, and silicon model. At 550 nm, the fixed refractive indices were 1, 2.41, 1.46, and 4. The calculation covered diamond thicknesses of 0 to 1200 nm and SiO₂ thicknesses of 0 to 600 nm.

CIE 1931 XYZ and CIE 1976 CIELAB coordinates were calculated under an approximation of CIE Illuminant A. The CIE76 total color difference compared measured and modeled colors. The authors evaluated 5 nm thickness differences and present the microscope method with about 5 nm resolution. Its response was periodic and therefore not globally one-to-one.

Calibration used an image with a known thickness to fit brightness and reference thickness by minimizing color difference. The unknown image was then captured under identical illumination and exposure. Modeled spectra were swept over diamond thickness until their color difference was minimized.

For a DOI sample specified as 300 nm diamond on 300 nm SiO₂, the calibration image used bare SiO₂/Si. Images used a 5X objective with 0.15 numerical aperture and a 2.65 mm field of view. Calibration returned 84% brightness and 295 nm SiO₂ thickness. The algorithm inferred 304 nm diamond thickness, within ±3 nm of WLI.

Calibration errors remain possible because of illuminant mismatch, white balance, literature refractive indices, microscope color processing, and finite objective aperture. The authors present the method as an initial thickness estimate rather than a replacement for calibrated profilometry.

Comparison with prior thin-film approaches

Section titled “Comparison with prior thin-film approaches”

The present work is not the thinnest or smoothest diamond reported. Its distinct result is a larger true direct-bonded DOI area without a metal interlayer.

PlatformArea and roughnessDiamond thicknessSupport and bonding
This work500 × 500 µm²; Rq < 0.5 nmAbout 300 nmTrue direct-bonded DOI on SiO₂/Si; no metal interlayer
Guo et al. [51]200 × 200 µm²; roughness not reportedDown to 10 nmTrue bonded stack on Si, SiO₂, fused silica, sapphire, or LiNbO₃
Ding et al. [37]About 200 × 200 µm²; < 0.3 nm160 nm; about 1 nm variationBonded SiO₂/Si with a Cr/Au frame
Maletinsky group [55]Millimeter scale; < 0.2 nmDown to 70 nm; < 0.35 nm/µmFree-standing membrane; not a true DOI stack
IonQ platform [40]0.5 mm hexagonal membranes; < 0.3 nm160 nm post-bond filmsGold thermocompression bonding; free-standing and not a true DOI stack

Starting diamond thickness matters for material selection because its variation contributed to the final nonuniformity. Device engineers also need separate recipes for quartz-mask thinning and SiN-mask pattern transfer. The released chiplets show compatibility with planar lithography and handling, but the paper reports no cavity Q, insertion loss, or array yield.

The paper’s photonic DOI process needs a single-crystal diamond plate that can be direct-bonded to SiO₂/Si and plasma-thinned from 10 µm to about 300 nm. The matching diamond materials catalog lists MPCVD-grown single-crystal diamond plates and wafers up to 125 mm, with thicknesses from 0.1 to 500 µm. 6CCVD supplies the material and offers custom fabrication and engineering support for the plate preparation and bonding workflow.

View Original Abstract

Diamond qubits integrated with photonic circuits can be initialized, manipulated, entangled, and read out individually with high fidelity, making them attractive for large-scale quantum systems. However, the limited size of heteroepitaxially grown single-crystal diamond and photonic-grade diamond-on-insulator (DOI) remains challenging for integration with existing manufacturing processes. Here, we develop a plasma etch recipe to thin direct-bonded (100) single-crystal diamond ( < 50 μ m ) into large-area, thin-film DOI, and demonstrate free-standing photonic chiplets fabricated from the resulting DOI. The ICP-RIE recipe preserves bonding and provides sufficient micro-masking and surface-quality control. We thin a 10 μ m diamond bonded to SiO 2 / Si and obtain a photonic-grade DOI with diamond thickness ≤ 300 nm over 0.5 × 0.5 mm 2 with roughness < 0.5 nm . Diamond photonic chiplets are fabricated using a standard two-step lithography process, without complex thin-film transfer, under-etching, or pedestal formation. We also present a colorimetric study of diamond visibility and quantify color differences across thicknesses in common colorimetric spaces. This analysis enables automatic diamond-thickness extrapolation from standard optical microscope images with 5 nm resolution, in good agreement with white-light interferometry (WLI) measurements. The DOI substrate and colorimetric thickness-evaluation method provide an effective fabrication platform and reliable validation route for scalable manufacturing of diamond nanophotonic devices.