Skip to content

Atomic layer etching of diamond using argon and oxygen

FieldValue
JournalDiamond and Related Materials
Published2026-10-18
AuthorsAlma Beatrice Sergi, Angelika Kaiser, Jakob Buchheim, Markus Mohr
DOI10.1016/j.diamond.2026.114302

Oxygen dosing followed by argon plasma etching removed 1.909 ± 0.004 Å of {100} single-crystal diamond per cycle within a quasi-self-limiting process window. At 20 °C, 500 W ICP power, and 5 W table RF power, 600 cycles lowered rms roughness from 0.26 ± 0.02 nm to 0.15 ± 0.01 nm. The rate corresponded to about two atomic layers per cycle. In a separate test, shallow NV centers remained observable after 29 cycles removed about 5 nm.

Etch depth was measured by AFM after 600 cycles at each bias point. Thermally evaporated aluminum masks protected selected areas because oxygen dosing required masking. Mask erosion was negligible.

The bias sweep produced three regimes.

DC biasObserved behaviorProcess assessment
Below about 45 VThe modified surface was not fully removed.The process was incomplete.
About 45 to 95 VThe modified layer was removed without additional sputtering.The process formed a quasi-self-limiting plateau.
Above about 95 VDiamond sputtering became dominant.The process lost self-limiting behavior.

At 5 W table RF power, the plateau occurred at about 49 V bias. The upper regime was assigned to dominant diamond sputtering. Residual oxygen would increase sputtering because oxygen has a higher sputter yield than argon.

Each ALE cycle had four phases. Oxygen entered the chamber and reacted in the ICP plasma to modify the diamond. The first purge removed residual dosing gas. The table RF generator then enabled argon bombardment of the modified surface. A final purge removed etch products.

The ICP plasma stayed on, and the argon flow remained constant. An optical emission spectrometer guided adjustment of argon flow and purge duration to minimize oxygen background at the end of the first purge.

The samples were electronic-grade CVD single-crystal plates measuring 2.0 mm × 2.0 mm × 0.5 mm. Their surfaces had {100} orientation and a reported miscut of ±3°. Aluminum was patterned with AZ 5214 E photoresist. The exposed aluminum was etched in H2O:HNO3:H3PO4 at 1:1:8 and 50 °C for about 4 minutes.

Etching used an Oxford PlasmaPro 100 Cobra 300 with an ALE kit. Argon and oxygen each had a stated purity of 6.0. AFM measurements used a Park Systems NX10 in non-contact and tapping modes with an NCHR probe.

With table RF power fixed at 5 W, oxygen dosing was varied from 0 to 0.70 seconds. Etch per cycle rose until a plateau above 0.25 seconds. A 0.25-second dose was therefore selected for later work.

At about 49 V bias, separate 600-cycle runs isolated two unwanted contributions. Oxygen dosing without the argon bias gave 0.20 ± 0.01 Å per cycle. Skipping oxygen dosing gave 0.12 ± 0.01 Å per cycle.

The resulting ALE synergy was 81.01 ± 0.84%. The modest value mainly reflects unintended etching during oxygen dosing.

AFM scans covered five 40 × 40 μm² areas before and after 600 cycles. The comparison used a triacid-cleaned sample with aluminum masking to reproduce mask preparation conditions. Scratches disappeared after ALE, although polishing lines remained. Power spectral density decreased over a broad range of length scales. Oxygen dosing without argon bombardment instead caused severe roughening.

Angle-resolved XPS used detection angles of 45° and 70°, representing about 7 nm and 3 nm depths. The C-C sp3 contribution at 284.8 eV increased by a factor of 1.3 at 45° after etching. The increase at 70° was less pronounced.

Most small elemental residues seen before etching vanished. Phosphorus remained at 0.1 at.%, while fluorine and nitrogen each appeared at 0.3 at.% after etching. The new residues were attributed to earlier processes run in the same chamber.

Oxygen and double-bonded oxygen decreased after the argon step. The relative C-C sp2 and oxygen-bonded-carbon contributions increased, mainly near the surface. A simple XPS overlayer model estimated 1 sp2 monolayer before etching, 2.7 after an argon finish, and 2.1 after an oxygen finish.

The model does not account for surface roughness. The reported data therefore show no extensive sp2 formation, and an oxygen finish reduces surface sp2 carbon.

The measured 1.909 Å per cycle is about two {100} diamond atomic layers. A bias-pulsed Cl2 dosing and Ar etching process reported by Michaels had a plateau near 0.9 Å per cycle, equal to one 0.892 Å monolayer.

By comparison, CF4 dosing with Ar etching by Regnier et al. reached about 7.1 Å per cycle, or roughly two diamond unit cells. The 81.01% synergy was higher than that reported for CF4/Ar. It remained lower than the 97% measured for O2/Ar on amorphous carbon by Kanarik et al.

A proposed mechanism starts with oxygen dereconstructing the clean {100} 2 × 1 surface dimers into ether and carbonyl sites. A stable oxygen-covered surface can place carbonyl groups in the first layer and ether groups in the second.

During argon bombardment, CO desorption from second-layer ether bridges may leave dangling bonds in the third layer. This may weaken the first two layers enough for their removal as roughly two layers per cycle.

This mechanism remains a proposed explanation. Verification would require molecular dynamics simulation of the specific Ar/O2 diamond cycle, which the authors did not know to exist.

One E6 diamond substrate was overgrown with a 100 nm 12C purified diamond layer. The sample received 5 keV 15N+ implantation, with an expected average implantation depth around 8 nm, followed by standard annealing.

After 29 ALE cycles, about 5 nm had been removed. The sequence ended with oxygen dosing. NV centers remained observable, and their Rabi contrast was about 10%, similar to the value before etching.

This test shows that shallow centers survived the selected removal depth in one sample. A detailed study of spin-property changes is still under way. The paper identifies thin isotopically purified layers as a target for surface preparation.

The results apply to electronic-grade {100} single-crystal CVD diamond with a reported ±3° miscut. The paper does not establish the same etch rate or surface chemistry for other orientations. It also does not demonstrate regrowth of a purified layer after etching.

The proposed two-layer removal mechanism is not verified. The XPS overlayer estimate omits roughness, and the origin of the remaining phosphorus is unknown. Fluorine and nitrogen may carry over from earlier chamber processes.

Estimated sputter thresholds for argon bombardment of diamond are about 70 eV and 76 eV. The plasma potential is typically about 5 V under similar conditions. The paper notes that argon-induced amorphization may contribute to etching above 70 V bias.

Below about 45 V, removal was incomplete, while above about 95 V, sputtering took over. The ongoing NV-center work must resolve whether spin properties remain stable across the operating range.

For device engineering, the demonstrated value is controlled removal of a shallow defective layer and smoothing of {100} single-crystal surfaces. Material selection should therefore include surface orientation and finish, not only bulk diamond quality.

Shallow NV-center devices in this work require a polished {100}-oriented single-crystal diamond plate and controlled surface removal. 6CCVD’s diamond materials catalog lists MPCVD-grown single-crystal plates and wafers up to 125 mm, thickness from 0.1 to 500 micrometers, and polishing to Ra below 1 nm. The paper’s 0.5 mm plate format falls within that thickness range, while orientation and the O2 dosing and Ar etch sequence need review through engineering support.

View Original Abstract

The ability to remove surface defects of diamond by dry etching methods is of high importance for applications that make use of shallow color centers in diamond. Here, we present an atomic layer etching process, making use of oxygen as dosing gas and argon as etching gas, in order to achieve controlled etching of the surface without creating additional defects by sputtering. We demonstrate a process window, where quasi-self-limiting atomic layer etching takes place with an etch per cycle of (1.909 ± 0.004) Å/cycle, corresponding to about 2 atomic layers of the {100} oriented diamond sample. The process is able to reduce the rms surface roughness down to (0.15 ± 0.01) nm. Furthermore, X-ray photoelectron spectroscopy indicates only a small amount of sp 2 bonded carbon on the diamond surface after the atomic layer etching process.