Skip to content

Rapid surface smoothing of monocrystalline diamond with bias-pulsed atomic layer etching

FieldValue
JournalJournal of Vacuum Science & Technology A Vacuum Surfaces and Films
Published2026-06-22
AuthorsJulian A. Michaels, Yeghishe Tsaturyan, Russ Renzas, D. D. Awschalom, J. G. Eden
DOI10.1116/6.0005347

Bias-pulsed atomic layer etching removed monocrystalline diamond at 0.90 ± 0.02 Å per cycle with a 6 s cycle. The process also cut large-area RMS roughness on one (100) diamond substrate from 2.23 to 0.058 nm after 4000 cycles. Measured ALE synergy reached 98%.

Atomic layer removal occupied a 15 to 25 V plasma DC bias window in both purged ALE and BP-ALE. Below this window, the modified surface layer was incompletely removed during each cycle. Above it, energetic Ar species removed the unmodified layer beneath the modified surface. The authors classified this excess removal as sputtering.

The authors associate the middle regime with enough ion energy to remove only the chemically weakened layer. Etch per cycle, or EPC, was defined as etch depth divided by completed cycles. The bombardment step also had at least a 3 s t_on window for single-layer removal. The surface-smoothing recipe used 23 V DC bias within the monolayer regime.

Shortening t_off raised the etch rate, but the authors say the cycle became continuous ICP-RIE rather than ALE. Lengthening t_off beyond about 4 s appeared to amorphize the second atomic layer.

The experiments used electronic-grade (100) single-crystal diamond substrates. A Shipley S1805 photoresist pattern with 1 mm long rectangles created steps for AFM depth measurements. The Oxford Instruments PlasmaPro 100 Cobra inductively coupled plasma reactive ion etching system held the chamber at 60 °C and the substrate at 25 °C.

Chamber pressure was 7 mTorr, with 100 SCCM Ar and 40 SCCM Cl2. The reported recipes used 1000 W ICP power and 10 W CCP power. DC bias was varied from 0 to 40 V during process development.

BP-ALE varied the plasma bias state while both gas flows continued. The authors’ process model assigns chlorine and argon species to the off step. This step chemically modifies the surface, forms weakened carbonaceous chlorine compounds, and amorphizes the upper diamond layer. The on step ejects that modified layer through bombardment by energetic argon and other species.

The 6 s cycle used t_off = 4 s and t_on = 2 s. Etch depth came from 35 × 35 μm² AFM scans on a Bruker FastScan. Each data point averaged at least five scans taken at locations at least 100 μm apart.

Roughness used 30 × 30 μm² and 1 × 1 μm² scans on an Asylum Cypher. Their resolutions were 1024 × 1024 pixels and 512 × 512 pixels, respectively. Five large-area scans and ten small-area scans were averaged for each reported data point.

Etch depth was nearly linear from 100 to 400 cycles, with nearly the same amount removed at every tested count. Each point in that comparison used a different diamond substrate. The optimized EPC value was reported from a 600-cycle etch.

BP-ALE used no gas purge. Purged ALE used 30 s total, split into 15 s for Cl2 stabilization and 15 s for evacuation. A second purged recipe used 50 s total, split into 25 s for each step. All three recipes produced the same etch rate within one standard deviation after 100 cycles.

The isolated Cl2 step produced no measurable etch, with α = 0 Å per cycle. The isolated Ar step produced β = 0.02 Å per cycle. These values produced the reported synergy close to the ideal value of unity.

Until this work, BP-ALE had been demonstrated only on 4H-SiC. Diamond ALE dates to a 1988 patent, but the paper notes little development after that disclosure.

Surface roughness across different starting conditions

Section titled “Surface roughness across different starting conditions”

Surface smoothing depended on both the starting specimen and the cycle count. Most large-area Rq values fell, but the reduction was not universal or monotonic. Selected values from Table I show the largest reduction and one specimen that became rougher.

SubstrateBP-ALE cycles30 × 30 μm² Rq (nm)1 × 1 μm² Rq (nm)
A102.580.546
A140000.4480.226
B102.230.804
B120000.1270.088
B140000.0580.056
C200.2220.671
C22000.5830.709

The representative sample types were as-received, fine polished with suspected handling residue, and strain released with subsurface damage removed by ICP-RIE. After 4000 cycles, A1’s large-area Rq fell from 2.58 to 0.448 nm, while B1 fell from 2.23 to 0.058 nm. B1’s small-area Rq also fell from 0.804 to 0.056 nm after 4000 cycles.

C2’s large-area Rq rose from 0.222 to 0.583 nm after 200 cycles. A2 and C1 also showed nonmonotonic large-area values. Cycle count alone therefore did not predict the final roughness.

The authors calculated power spectral density curves from the same AFM images to compare surface features across spatial frequencies. They concluded that BP-ALE smoothed surfaces regardless of starting roughness or processing state. For diamond-on-diamond epitaxy, the paper identifies low Rq, controlled morphology, and defined step terraces as seed-quality factors.

Optical emission spectroscopy and short-cycle behavior

Section titled “Optical emission spectroscopy and short-cycle behavior”

Optical emission spectroscopy was performed in situ through a chamber porthole with an Ocean Optics SR4 spectrometer. Acquisition was limited to 25 cycles because the time resolution depended on total cycle duration.

The plasma spectrum was dominated by Ar transitions. Cl signals were not observable with this setup. The 763 nm and 811 nm Ar lines changed directly with DC bias and tracked the off and on portions of each cycle.

For t_off = 4 s and t_on = 2 s, the signals saturated near 4900 at 811 nm and 2600 at 763 nm. With t_off = 6 s and t_on = 2 s, saturation was lower at approximately 4700 and 2500. The t_off = 8 s and t_on = 4 s recipe reached slightly higher levels of about 5000 and 2700.

The authors suggest that startup depends on t_on and the duty cycle. They state that the data are not conclusive. The steady state may affect etching fewer than 25 layers.

For a 1.5 s programmed cycle set to 1.0 s off and 0.5 s on, plasma feedback likely made operation closer to 0.75 s off and 0.75 s on. The authors judged that cycle too fast for consistent operation on their system. More research is needed to reach steady state from the first cycle rather than after the first few dozen.

The study did not exhaustively optimize plasma energy, chemistry, or chamber design. These unresolved factors limit the direct transfer of the reported cycle timing to other reactors.

The measured roughness reduction supports evaluating BP-ALE as a post-process for quantum-information surfaces and homoepitaxial growth seeds. The authors argue that smoother surfaces can reduce optical or electrical scattering. They also note that existing plasma etches can leave damage and roughness that impede quantum systems.

Surface selection should account for initial Rq, scan scale, and cycle count because the smoothing data vary by specimen. Run-up behavior also matters when fewer than 25 layers are needed because the Ar emission reached steady state only after the first cycles.

The paper reports etch depth, roughness, synergy, and plasma emission, but not device-level optical, electrical, or spin-qubit performance. It also gives no direct roughness benchmark against chemical mechanical polishing or conventional RIE. Further work must resolve the low-cycle plasma response and specimen-dependent smoothing before BP-ALE is treated as a general diamond finish.

For quantum-information devices and homoepitaxial diamond growth, the paper used (100) single-crystal diamond and identifies low roughness and controlled morphology as material requirements. 6CCVD’s diamond materials catalog lists MPCVD-grown single-crystal plates and wafers up to 125 mm, with polishing to Ra below 1 nm for single-crystal material. The catalog uses Ra rather than the paper’s AFM Rq metric, and the (100) orientation should be confirmed through engineering support before material selection.

View Original Abstract

Atomic layer etching (ALE), an etch method capable of precisely removing atomic monolayers, is gaining renewed interest as an enabling technology across a wide range of applications. However, the long cycle times in conventional ALE, imposed by the need to purge between reactant doses, can make the process time-intensive. Recently, bias-pulsed atomic layer etching (BP-ALE) has been shown to improve throughput while maintaining atomic precision in material removal. BP-ALE differs from conventional ALE by pulsing only the plasma DC bias in an inductively coupled plasma-reactive ion etching reactor, thereby reducing cycle duration. Here, we demonstrate the application of BP-ALE to diamond etching using Ar/Cl2 plasma, reporting an etch rate of 0.90±0.02AA/cycle with 6 s cycles. We study the transition from BP-ALE to conventional ALE as a function of purge time. Furthermore, we observe a smoothing effect on diamond surfaces, with clear reductions in surface roughness across multiple single-crystal diamond substrates. Additionally, we use optical emission spectroscopy to elucidate plasma behavior in the BP-ALE process. This work presents a new perspective on reactive ion etching of diamond, one of the premier materials in semiconductor and quantum technologies.