Skip to content

Quality factor enhancement in polycrystalline diamond MEMS resonators by post-deposition plasma treatment

FieldValue
JournalJournal of Applied Physics
Published2026-01-22
AuthorsD.A. Huber, M. Stöger-Pollach, M. Gülly, Christoph Schallert, D. Steinmüller-Nethl
DOI10.1063/5.0303958
Citations1

Back-thinning the growth-side face of 2.2 μm polycrystalline diamond resonators raised the mean quality factor from 25,472 to 73,924. The nearly threefold gain followed 28 min of treatment, based on laser Doppler vibrometry below 10⁻⁵ mbar. Measurements across 142 devices covered out-of-plane modes from 20 to 500 kHz, with individual Q values up to 315,000.

The experiment tracked the same resonator set through eight sequential backside treatments. Mean Q rose through the 28 min dataset, while detailed mode trends reached a plateau at about 24 min. The 32 min dataset produced a decrease. The maximum measured Q increased by a factor of four across the 142-device set. The first Euler-Bernoulli mode did not show a significant change in mean or maximum Q. Anchor and thermoelastic damping dominated that mode, whereas other out-of-plane modes tracked the improvement in the defective interface layer. The authors attribute that contrast to mode-dependent loss mechanisms rather than a uniform Q response across all modes.

Polycrystalline diamond was deposited by modified HFCVD on a 4 in. (100) silicon wafer. Electrostatic seeding used 5 nm detonation diamond nanoparticles dispersed in isopropyl alcohol. Deposition ran at 800 to 850 °C and 5 mbar, with a CH₄/H₂ ratio of 0.1% for micrometer-sized grains. The final device layer was 2.2 μm thick.

SEM showed smaller grains near the former silicon interface, with grain size becoming nearly constant within the first 500 nm of growth. Cross-sectional EFTEM set the carbon K-edge filter at 284 eV. The brighter substrate-near region indicated non-diamond carbon clusters extending roughly 500 nm into the film. Electrons scattered from sp³ diamond bonds lost slightly more energy at 290 eV. This imaging identifies the candidate defect zone, while the resonator measurements establish its effect on Q.

The resonators were single-side clamped plates with a fixed length of 500 μm. Widths ranged from 190 to 1000 μm. ICP reactive ion etching used aluminum hard masks, followed by Bosch release and photoresist removal with TechniStrip P1316.

The main diamond etch used oxygen at 49 SCCM, CHF₃ at 0 to 1 SCCM, and 25 °C. Chamber pressure was 15 mTorr, with 1000 W ICP power and 100 W RF power. Back thinning used the same dry etch approach at 500 W ICP power to reduce surface damage.

An assumed 30 nm/min rate predicted about 120 nm removal during the first 4 min. The authors state that the rate was not constant because grains, grain boundaries, and amorphous carbon etch at different speeds.

Photothermal actuation used a 635 nm laser diode, and laser Doppler vibrometry recorded resonance peaks. Measurements were made below 10⁻⁵ mbar, where the authors treated fluidic damping as negligible. The measured set included Euler-Bernoulli, torsional, roof-tile-shaped, and higher-order plate modes. SEM tracked cross sections and morphology, while AFM measured surface topography.

Plasma preferentially removed amorphous carbon and small grains because their bonds were weaker. The mean Q increased by more than 11,000 after the first treatment. A second large gain of nearly 14,000 occurred between the second and third treatments.

By about 12 min, the amorphous clusters were removed and the exposed surface showed more uniform grain size. Continued etching removed intermediate grains and rounded the largest grains, which had a lower etch rate. The mean etch rate fell as etch time increased because carbon in diamond bonds resisted the plasma more strongly.

These morphology changes support selective removal of the growth-side defect region. The authors note that the first jump could also include surface loss released by the Bosch process.

AFM found RMS roughness rising from 20 nm to 120 nm as backside material was removed. Height differences increased from roughly 100 nm to as much as 767 nm after 28 min. Despite that rougher surface, Q did not correlate with roughness under the measurement conditions.

The authors therefore identify friction in amorphous carbon and dense grain boundaries around small grains as the dominant initial losses. They describe these as structural defects within the substrate-near growth region.

A surface-friction model assumed a 600 nm defect layer and its removal by 500 nm. For a 200 μm wide plate, the model predicted a 4.6-fold Q increase as thickness fell from 2.2 to 1.7 μm. The measured threefold gain is close to that model, although the treatment produced a rough, nonuniform cross section.

After the defect-rich region was reduced, the Q plateau was attributed mainly to thermoelastic and material damping. At 32 min, reduced resonator volume and increased surface-to-volume ratio were cited as reasons for the decline. Water adsorption, oxide formation, and dangling bonds were listed as possible surface-related losses that become more important with higher surface-to-volume ratio. At 20 to 500 kHz, the authors did not treat roughness-induced acoustic scattering as a relevant loss for these devices.

The treated mean Q of 73,924 reached the range reported for single-crystalline silicon resonators of similar thickness. The authors still place polycrystalline diamond below single-crystal diamond. Prior single-crystal devices exceeded Q of 1 × 10⁶ under vacuum at room temperature. Individual polycrystalline modes reached 315,000, which the authors describe as the same order of magnitude as single-crystal diamond. A previous study found a similar Q trend after thermal treatment at 800 to 850 °C in nitrogen.

The result applies to one HFCVD film, one clamped-plate geometry, and modes from 20 to 500 kHz. Mode-to-mode loss variation remains large, and the first Euler-Bernoulli mode did not benefit. Backside treatment cannot be applied to some complex structures because it requires access to the film underside. The authors call for improved nanoscale seeding and suggest tailored stacked diamond layers as possible future approaches.

For device engineering, the growth interface is a controllable loss region. The measured process window favors stopping near the plateau before the Q decline at 32 min. Post-deposition plasma treatment trades added surface roughness for lower structural loss in this device family.

For polycrystalline diamond MEMS resonators, the diamond materials catalog from 6CCVD lists MPCVD-grown plates and wafers up to 125 mm. Catalog thicknesses span 0.1 to 500 micrometers; the paper needs a 2.2 micrometer PCD device layer and backside plasma access. The paper used HFCVD on a 4 in. silicon wafer, so engineering support can assess growth and etch compatibility for the resonator structure.

View Original Abstract

Polycrystalline diamond is a promising material for MEMS resonators due to its remarkable mechanical and electrical properties and compatibility with standard semiconductor manufacturing processes. However, the growth on non-diamond substrates is challenging, and a seeding process is needed to grow closed thin films. Hot filament chemical vapor deposition and reactive ion etching are powerful tools to deposit and micromachine diamond thin films on silicon substrates. In this paper, a polycrystalline diamond thin film with micrometer-sized grains is used to fabricate MEMS resonator devices, and quality factors are measured using laser Doppler vibrometry. The resonator’s bottom and top sides, as well as the cross sections, are investigated, and a substrate-near region with an elevated amount of non-diamond carbon bonds is identified with TEM in the energy-filtered transmission electron microscopy mode. By monitoring the quality factors of several resonance modes of 142 MEMS resonator devices, while a reactive ion etching treatment is performed at the bottom side of the resonators, we find an initial increase in the mean quality factor of all out-of-plane modes in the frequency spectrum from 20 to 500 kHz by nearly a factor of 3 after 28 min of back thinning of an initially 2.2 μm thick polycrystalline diamond thin film due to the reduction of the defect-rich substrate-near region. Even more, we find no correlation between surface roughness and quality factor, indicating losses at structural defects, such as grain boundaries and non-diamond carbon clusters, as dominant for energy dissipation.

  1. 2007 - MEMS technology for timing and frequency control [Crossref]
  2. 2012 - A review of MEMS oscillators for frequency reference and timing applications [Crossref]
  3. 2015 - CMOS-MEMS resonators: From devices to applications [Crossref]
  4. 2020 - MEMS resonators for frequency reference and timing applications [Crossref]
  5. 2014 - Single-crystal diamond nanomechanical resonators with quality factors exceeding one million [Crossref]
  6. 2018 - Reducing intrinsic energy dissipation in diamond-on-diamond mechanical resonators toward one million quality factor [Crossref]
  7. 2008 - The CVD of nanodiamond materials [Crossref]
  8. 2010 - Diamond growth by chemical vapour deposition [Crossref]
  9. 1999 - CVD diamond films: Nucleation and growth [Crossref]
  10. 2008 - High-performance polycrystalline diamond micro- and nanoresonators [Crossref]