Skip to content

Clean Up Behind You ‐ Novel Patterning Approach for Solid Immersion Lenses

FieldValue
JournalAdvanced Functional Materials
Published2026-02-21
AuthorsAleksei Tsarapkin, Stefan Facsko, Wentao Zhang, Anna Mogilatenko, Kilian Unterguggenberger
DOI10.1002/adfm.202531985
Citations1

A sixfold increase in nitrogen-vacancy light collection was achieved with a hemispherical solid immersion lens patterned in single-crystal diamond. The fabrication used a 5 keV gallium focused ion beam and no lens post-processing. TEM measured a 10 to 15 nm Ga-containing damage layer above intact diamond. The sidewall ripples agreed with continuum predictions and did not change the modeled optical collection.

The authors replaced discrete ring milling with a continuous Archimedean spiral. Radially varying dwell time encoded the hemispherical shape through a varying local ion flux. The path used 20 nm radial and tangential pitches at 2.1 nA, about 5% of the spot diameter. The beam completed 1000 outward and inward spiral repetitions, ending with the inward pass. That sequence removed most redeposition from earlier passes. Total milling time was 2 h 13 min, comparable to other strategies at the same ion energy.

The substrate was a 2 by 2 by 0.5 mm³ single-crystal CVD diamond with nitrogen below 5 ppb. NV centers were formed through ¹²C implantation at 11 MeV, a dose of 10⁹ ions/cm², and a 7° incidence angle. Vacuum annealing at 1050 °C for 16 hours at 10⁻⁷ hPa enabled vacancy diffusion and nitrogen-vacancy pairing. An acid clean removed graphite before lens fabrication and left an NV density of 10⁷ cm⁻².

The diamond surface remained pristine during milling, with no conductive coating. Earlier ring-based protocols used discrete rings and boiling acid mixtures to remove contamination after milling. Redeposited amorphous carbon contained Ga and could absorb or scatter light on the lens sidewalls.

The authors adapted a Kuramoto-Shivashinsky surface-evolution model to a hemisphere with a changing local incidence angle. The numerical integration used a 2000 by 2000 grid and fourth-order Runge-Kutta steps. Rotated derivative operators accounted for the changing azimuth of the projected ion beam.

The model predicted a surface instability at a local angle of about 40° to the surface normal. The upper surface remained smooth, while steeper sidewalls developed periodic corrugations. Only the curvature-dependent sputtering coefficient varied with local angle. Tilt-dependent sputtering and higher-order terms were not varied because the model was intended to describe ripple formation qualitatively. Microscopy confirmed the predicted onset location and matched the simulated surface modifications.

Published Ga FIB protocols used about 30 to 40 keV ions, while this study used 5 keV. At 2 keV, Ga accumulated as surface droplets while the underlying diamond remained largely unaffected. The authors attributed the droplets to Ga accumulation and Ostwald ripening at the low penetration depth. The 5 keV setting reduced damage depth and ripple modulation compared with higher energies.

TEM and STEM showed intact single-crystal diamond below a 10 to 15 nm Ga-containing amorphous layer. EDXS mapped Ga into the damaged layer and suggested slightly shallower damage on steep sidewalls. The dynamic IMSIL model combined binary ion and recoil trajectories with continuum flow from implantation, relocation, and sputtering.

The model applied a homogeneous 5 keV Ga flux to digitized TEM topographies and ran each case to sputter equilibrium. The predicted amorphous depth was about 14 nm at 46.9° and about 8 nm at larger inclinations up to 68.5°. The measured top-surface depth agreed with that prediction, but projection artifacts prevented a direct check of the sidewall trend. The model underestimated ripple stability, but it did not predict a thicker damage layer on steep walls.

The paper cites a Ga stopping range of about 7 to 22 nm in diamond for 5 to 30 keV. Straggling contributes to the resulting damage layer. The remaining Ga-rich layer is intrinsic to ion-beam processing and was not removed by post-processing.

Finite-element Maxwell calculations compared fiber collection from a planar surface, an ideal hemisphere, and the rippled hemisphere. The JCMsuite model used Bayesian placement of the SIL and fiber, with 680 nm selected as the reference wavelength. The simulation covered 600 to 800 nm and reported average efficiencies over 650 to 800 nm.

Modeled geometryCollection efficiency, 650 to 800 nmMean collection factor, 600 to 800 nm
Planar diamond2.5(1)%Baseline
Ideal SIL16.3(1)%6.46(9)
Rippled SIL16.3(1)%6.34(9)

Ray optics predicted a 6.1-fold increase for a perfect hemisphere at numerical aperture 0.82. The ideal and rippled collection factors differed slightly, with overlapping reported uncertainties. The authors therefore found the optical effect of surface corrugations negligible. Modeled oscillations arose from quantum interference caused by reflections from the SIL surface.

Optical tests used a confocal microscope with an air objective of NA 0.82 and a single-mode fiber pinhole. A 525 nm laser provided off-resonant excitation, and collection covered the 650 to 800 nm phonon sideband at room temperature.

Saturation metricNV in SILRepresentative NV in bulk diamond
Count rate172 kc/s29 kc/s
Optical power0.31 mW1.8 mW

The SIL produced an 8.5-fold higher maximum count rate than bulk diamond. The measured collection and excitation improvement was 6.0, close to the 6.1 ray-optics expectation. The NV appeared 1.3(3) μm from the SIL center in the image. Magnification by diamond’s refractive index of 2.4 placed its true center distance at 0.5(1) μm.

A three-level model fit gave g²(0) = 0.09, confirming single-photon emission from one NV center. The lens also reduced the confocal point spread function and improved the signal-to-background ratio. The authors attributed the small difference from theory to the off-center NV position.

The process leaves a 10 to 15 nm Ga-containing amorphous layer, which the authors describe as process-inherent damage. Moderate redeposition in the lowest parts of the hemisphere could not be excluded. If present, it would reduce effective erosion and could increase the amorphous-zone thickness. Projection artifacts prevented direct measurement of the shallower damage predicted on steep walls.

The binary collision model also underestimated the stability of the observed ripple patterns. Its radiation-induced viscosity used a silicon value because no carbon value was available. All optical measurements were conducted at room temperature, so cryogenic optical performance was not tested. The paper does not report a production yield for the new spiral protocol. The optical comparison used one NV inside a SIL and a representative NV in bulk diamond.

Diamond’s refractive index of 2.4 makes total internal reflection a dominant loss for NV emission. When the emitter lies at the SIL center, its light reaches the surface near perpendicular incidence. The hemispherical geometry also enables spherical-aberration-free imaging.

FIB imaging supports registration around a prelocated NV, but the measured emitter offset shows why center placement matters. The sidewall ripples need not reduce collection for the modeled and fabricated geometry. Device designs should still include the 10 to 15 nm damaged layer and the 0.5(1) μm emitter offset in their optical budget.

Lens radius followed the simulated implantation depth, linking SIL geometry to defect-layer placement. For material selection, the demonstrated host was low-nitrogen single-crystal diamond with nitrogen below 5 ppb before implantation.

For this work, the relevant material family is MPCVD-grown single-crystal diamond for solid immersion lenses and NV photonics. 6CCVD’s diamond materials catalog lists plates and wafers up to 125 mm, with single-crystal thickness from 0.1 to 500 micrometers. The paper’s 2 by 2 by 0.5 mm³ host needs NV formation and FIB lens patterning, with process questions directed to engineering support.

View Original Abstract

ABSTRACT Solid‐state immersion lenses have the potential to become a key component in photonic quantum technologies, as they enable efficient collection of quantum light. Although they are conceptually simple and powerful at the same time, they are not yet widely used today. This is a consequence of the challenges in the controlled processing of the typically ultra‐hard crystalline materials such as diamond in combination with the deterministic positioning of the lens relative to the position of the quantum emitter. Here we present a novel patterning approach for the fabrication of solid‐state immersion lenses using focused ion beam milling, which eliminates the need for post‐processing. The hemispherical lens shape is defined by a radially varying ion flux applied in a continuous Archimedean spiral beam path. In line with the prediction from continuum modeling, we observe periodic surface corrugations at the side walls of the lenses indicating that the ion flux is locally perfectly homogenized by our approach. Microstructural investigations reveal a damage layer of only 10 - 15 nm above an otherwise intact diamond crystal. This corresponds to the expected beam damage obtained from atomistic binary collision modeling, and proves that redeposition is indeed negligible. To demonstrate their performance, lenses with nitrogen vacancy centers are optically examined, revealing a sixfold enhancement in light collection efficiency in combination with reduced background noise.