Skip to content

High‐Quality Free‐Standing Diamond Micromembranes for Nanoscale Quantum Sensing

FieldValue
JournalAdvanced Optical Materials
Published2026-03-01
AuthorsAlexander C Pakpour‐Tabrizi, Artur Lozovoi, Sean Karg, Tecla Bottinelli Montandon, Melody Leung
DOI10.1002/adom.202503864

Free-standing diamond nanobeam micromembranes achieved a fabrication yield above 90% and a photon collection gain as high as about sevenfold. The process preserved shallow nitrogen-vacancy spin and charge performance within nanometers of the diamond surface.

The authors began with (100) electronic-grade single-crystal diamond and used the surface preparation described in prior work. Nitrogen implantation used 3 keV ions at an average depth of 6 nm and a dose of 10⁹ cm⁻². An 800 °C vacuum anneal created shallow NV centers. They deposited 150 nm of SiNₓ at 250 °C by plasma-enhanced chemical vapor deposition. The SiNₓ hard mask was patterned by electron-beam lithography with a conductive polymer layer. An anisotropic ICP-RIE step used 14 sccm CHF₃, 10 sccm SF₆, 5 mTorr, 200 W RF, and 60 W ICP.

Plasma etching then defined the mesa in pure oxygen at 10 mTorr. The diamond step used 110 W ICP power, 1000 W RF bias, and 165 s. It removed 820 nm with near-vertical sidewalls, no edge collapse, and little micromasking. Another conformal 150 nm SiNₓ layer followed the diamond etch. A directional nitride etch removed the bottom mask while retaining protection on the top and sidewalls.

The critical undercut ran for 20 h at 140 °C with 1000 W ICP, 0 W RF, and 20 mTorr. At a 1 μm spacing, the lateral etch reached about 600 nm and undercut the targeted beam range. The authors lowered the temperature and ICP-RIE power relative to their reference process to control this geometry. That choice reduced aspect-ratio-dependent variation across dense structures. Widths from 120 to 720 nm were tested in a shared fabrication format. FDTD calculations targeted beams about 700 nm wide and 350 nm thick.

The suspended layout used a double-layer trellis, many narrow tethers, and a 1 μm gap around each frame. The tethers were about 100 nm wide to provide mechanical flexibility. Each tether could be broken with a micromanipulator during transfer. The frame buffered faster corner etching and limited frame collapse or delamination. Microscopy showed smooth structures, while FIB cross-sections revealed differing bottom profiles across beam widths.

After undercut, the sample remained in 49% HF for 24 h to strip SiNₓ. A refluxing nitric, perchloric, and sulfuric acid clean followed. A 450 °C thermal oxygen anneal then provided high-quality oxygen termination. X-ray photoelectron spectroscopy checked surface cleanliness and termination. The process produced more than 10 substrates with a yield above 90% across the wafer. Yield required full undercut, transferability, no edge erosion, and resistance to standard diamond processing. One 4.5 × 4.5 mm² host sample contained thousands of undercut device sites.

Two pulled glass probes broke the tethers, picked up each membrane, and held it electrostatically. The probes moved, flipped, and placed the membrane on sapphire. The transferred membranes survived refluxing acids, HF, piranha, solvents, and nitrogen blow-drying. Zero-field ODMR data showed 63% unstrained centers on sapphire and 54% in suspended beams. The similar fractions indicated no significant transfer-induced strain or electric field at the interface.

Proton nuclear magnetic resonance measured NV center depth for the coherence measurements. Hahn echo T₂ decreased as the centers approached the diamond surface. Beam measurements fell within the host distribution and agreed with prior shallow-NV results. The best prior points used a 1200 °C anneal, compared with 800 °C in this work. The authors suggest that implantation-related multivacancy clusters cause the shorter values seen with the 800 °C anneal. A higher-temperature anneal is proposed as a possible improvement.

XY-N dynamical decoupling increased T₂ without saturation through 96 pulses. The scaling departed from N^(2/3), which the authors link to broadband surface-defect noise. XY-8 spectra also showed broadband noise and a ¹³C Larmor peak at 360 G. Single-quantum and double-quantum T₁ values ranged from 100 μs to several ms, with no clear depth dependence. Double-quantum T₁ values from 100 μs to 1 ms matched the prior shallow-NV results. Several beam centers had single-quantum T₁ from 200 μs to 1 ms, while the reference reported values above 1 ms. The authors flag possible extra high-frequency magnetic noise at the surface. The beam and host lifetime distributions were otherwise similar.

Charge populations were measured after green initialization under 594 nm illumination at about 3 μW. A double-Poisson fit separated the bright NV⁻ and dark NV⁰ photon distributions. The sample included 116 centers in beams and 148 in host diamond. Fifty percent of beam centers and 45% of host centers had more than 60% NV⁻ population. The similar distributions support adequate steady-state charge properties after fabrication.

Time-resolved photoluminescence measured ionization across several orange illumination powers. The ionization coefficient scaled with optical power squared, as expected for two-photon ionization and recombination. The mean coefficient was 5.4 Hz/μW² in beams and 3.6 Hz/μW² in host diamond. About 20% of beam centers had coefficients above 10 Hz/μW². The authors attribute the higher beam values to stronger optical excitation. Higher excitation therefore increased charge switching for some centers.

Rabi oscillations at 25 G measured spin contrast for 53 beam centers and 65 host centers. About 52% of beam centers and 56% of host centers exceeded 20% contrast. The beam set had about 50% more centers with no detectable contrast. The authors connect this subset to lower NV⁻ populations in the beam group.

Photonic collection and sapphire integration

Section titled “Photonic collection and sapphire integration”

Measured collection efficiency averaged about twofold above host diamond in air and threefold on sapphire. Individual centers reached gains as high as about sevenfold. The suspended-beam saturation power was 0.49 ± 0.29 mW, versus 0.79 ± 0.29 mW for host diamond. Excitation efficiency rose 1.5-fold on average and more than threefold for 20% of centers. Lower excitation power can reduce illumination of the sensing target.

FDTD simulations included beam geometry, NV depth, dipole orientation, and environmental refractive index. The model assumed a 6 nm implantation depth without straggle and rectangular beam sections. Collection varied strongly with lateral position because the result depended on proximity to a sidewall-air interface. NV dipole orientation produced both bright and dark cases. Measured lateral displacement stayed below the optical diffraction limit. The broad simulated range could therefore not be checked center by center. Measured and simulated average gains nevertheless agreed. On sapphire, the implanted side faced downward and sidewall total internal reflection aided collection. The paper reports lower average collection efficiency than diamond nanopillars, but the comparison differs in orientation, localization, and collection mode.

Engineering implications and unresolved issues

Section titled “Engineering implications and unresolved issues”

Pick-and-place integration could place shallow NV centers on superconducting, magnetic, and other condensed-matter targets. This geometry can support interrogation of optically opaque materials that limit single-crystal diamond sensors. Front-side access also avoids the optical aberrations associated with back-side imaging. Arrays can support wide-field imaging, multiplexing, and covariance magnetometry.

The 20 h etch remains aspect-ratio-sensitive and produces profile asymmetry near the wafer edge. Beam centers show faster ionization and more zero-contrast centers than host centers. The authors propose patterned implantation or nanoimplantation to control lateral NV placement. They also identify (111) substrates as a route to preferential dipole alignment. The authors state that controlled lateral implantation could make collection efficiency more reliable.

The device requires nitrogen-implanted, oxygen-terminated single-crystal diamond formed into free-standing micromembranes for NV quantum sensing. 6CCVD’s diamond materials catalog lists MPCVD-grown single-crystal diamond. The engineering support page can be used to specify the paper’s nanobeam geometry, undercut process, and pick-and-place sequence.

View Original Abstract

ABSTRACT Deploying nitrogen vacancy (NV) centers in diamond as nanoscale quantum sensors for condensed matter and materials physics requires placing the NV centers close to the sensing target. One solution is to fabricate diamond nanostructures and integrate them with materials and devices. However, diamond etching and ion milling can introduce subsurface damage and surface defects that degrade the charge stability and spin coherence of NV centers near the surface. Here, we report a procedure for fabricating low‐damage free‐standing diamond micromembranes, and we show that this fabrication scheme preserves the optical and spin properties of state‐of‐the‐art shallow NV center quantum sensors, within nanometers of the diamond surface, while providing significant photonic enhancement. Furthermore, we demonstrate a pick‐and‐place transfer method, which enables integration with diverse sensing targets.

  1. 2023 - Imaging Ferroelectric Domains with a Single‐Spin Scanning Quantum Sensor