Q-optimised nanoelectromechanical diamond resonators
Paper details
Section titled “Paper details”| Field | Value |
|---|---|
| Journal | Microsystems & Nanoengineering |
| Published | 2026-03-03 |
| Authors | Evan L. H. Thomas, S. Mandal, William G.S. Leigh, Oliver A. Williams |
| DOI | 10.1038/s41378-026-01189-1 |
| Citations | 1 |
Free-free nanocrystalline diamond (NCD) resonators reduced dissipation by up to 8.8× at about 100 MHz. One device reached an unloaded Q factor of 9176 at 12 K. The best measured product of resonant frequency and Q was 1.26 × 10¹² Hz for a roughly 60 MHz as-grown device. The data associate lower loss with support geometry and show no clear difference in length-independent loss between rough and polished films.
Free-free geometry suppresses clamping loss
Section titled “Free-free geometry suppresses clamping loss”Doubly clamped beams anchor both ends, allowing flexural strain to transfer energy into the supports. Free-free resonators replace those anchors with second-order flexural supports attached at zero-displacement nodes. These supports resonate with the main beam’s second-order mode and restrict energy transfer.
Nominal widths were 500 nm and lengths were 8.8, 9.85, 11.4, and 13.95 µm. The set covered target frequencies of 40, 60, 80, and 100 MHz. The two resonances within each pair were separated by 4 to 5 MHz.
For the as-grown 8.8 µm pair near 100 MHz, doubly clamped dissipation was 9.7 × 10⁻⁴. The free-free value was 1.1 × 10⁻⁴, giving the measured 8.8× reduction. The polished pair gave 2.2 × 10⁻³ for the doubly clamped beam and 2.6 × 10⁻⁴ for the free-free design. The 8.5× reduction shows that the geometry change reduced loss for both stocks.
Nanocrystalline diamond growth and resonator fabrication
Section titled “Nanocrystalline diamond growth and resonator fabrication”The devices used 2-inch silicon (100) float-zone wafers with 500 µm thickness and 500 nm thermal oxide on both sides. Oxygen plasma at 30 W and 30 SCCM O₂ for 1 min cleaned the surfaces. Electrostatic attachment during nanodiamond colloid seeding produced densities of about 10¹¹ cm⁻² after 10 min of ultrasonication.
NCD growth used a Seki AX6500 MWCVD reactor at 4.1 kW, 40 Torr, and about 790 °C. The gas was 5% CH₄/H₂ during incubation until about 50 nm. The methane fraction then fell to 1% CH₄/H₂ for the remaining growth.
One wafer supplied rough stock, while the other was chemically mechanically polished to an indicated 413 nm. AFM measured centre roughness of 34 nm RMS for rough stock and 3 nm RMS for polished stock over 25 µm². Mean thicknesses before release, including chromium and gold, were 442.0 ± 0.4 nm and 444.9 ± 0.1 nm, respectively.
Central pieces from each wafer received 5 min of 1:3 H₂O₂:H₂SO₄ treatment to standardise oxygen termination and improve metal adhesion. Thermal evaporation deposited a 105 nm chromium, gold, and nickel stack. Chromium provided adhesion, gold provided conduction, and nickel served as an etch mask. Electron-beam lithography used a 410 nm MMA/PMMA bilayer. The ICP etch ran at 10 mTorr, 40 SCCM O₂, 1500 W ICP power, and 100 W RIE power. Its carrier wafer stayed at 25 °C. HF vapour release of the SiO₂ buffer used a 50 °C carrier wafer.
Cryogenic measurement isolates unloaded dissipation
Section titled “Cryogenic measurement isolates unloaded dissipation”At 12 K and about 10⁻⁴ mbar, the resonators were driven in plane by Lorentz force in fields up to 8 T. Measurements covered fields from 0 to 8 T in 0.5 T steps, with three S21 values recorded at each field. The authors observed 15 resonances attributed to mechanical motion: nine from rough stock and six from smooth stock. Five rough devices were doubly clamped and four were free-free. The smooth set contained three of each geometry.
Each trace was fitted with a Lorentzian response, background terms, and a phase factor. The symmetric lineshapes indicated a linear drive regime. Peak heights scaling with B² confirmed that the signals came from mechanical motion. Magnetomotive damping follows Q⁻¹ proportional to B²L⁴, so the authors extrapolated dissipation to zero field. For a 13.95 µm smooth-stock beam near 40 MHz, unloaded dissipation was 1.19 × 10⁻⁴, corresponding to Q = 8370.
Length scaling identifies clamping loss
Section titled “Length scaling identifies clamping loss”For rough-stock doubly clamped devices, the aL⁻⁵ + b model fitted better than the aL⁻³ + b model. The limited dataset makes this scaling tentative. It agrees with earlier 10 to 19 µm NCD resonators and indicates more rigid clamping than the conventional α(w/L)³ model.
Free-free supports lowered the rough-stock length-dependent coefficient from 29.3 ± 3.5 µm⁵ to 3.8 ± 0.8 µm⁵. The length-independent offsets were 5.2 ± 2.2 × 10⁻⁵ for doubly clamped beams and 2.2 ± 0.9 × 10⁻⁵ for free-free resonators. The smooth-stock fit gave length-dependent coefficients of 41.6 ± 16.1 µm⁵ and 4.9 ± 1.9 µm⁵ for the two geometries. Its offsets were 0.9 ± 5.3 × 10⁻⁵ and 2.1 ± 1.3 × 10⁻⁵, respectively. These fits were indicative because only three smooth-stock devices used each geometry.
Surface loss showed no dominant effect at 12 K
Section titled “Surface loss showed no dominant effect at 12 K”No statistically significant difference appeared in length-independent loss between rough and polished devices. Surface loss should vary little with beam length. The matching offsets therefore suggest that upper-surface dissipation was not dominant at 12 K. A silicon carbide study at 4.2 K reported significantly degraded Q for stock with roughness up to 7.1 nm, unlike the present NCD comparison.
The authors could not exclude structural or chemical defects in the bulk or on the O₂-etched and nucleation undersides. Residual clamping loss remains a possible contributor. Finite tether widths, imperfect nodal anchoring, fabrication variation, and material-property uncertainty can preserve length dependence. Undercut depth and HF vapour etch uniformity can also vary support loss. Silicon carbide formation during diamond deposition and etch temperature nonuniformity can change the oxide removal rate.
At 12 K, thermoelastic damping was expected below 10⁻¹². Molecular damping was negligible at about 10⁻⁴ mbar. Internal loss in the gold and chromium was estimated below 7%. Its largest share occurred in the smooth-stock free-free beam near 40 MHz. Temperature-dependent tests with pressure decoupled from temperature would likely separate the remaining mechanisms.
Performance against prior resonators
Section titled “Performance against prior resonators”The highest doubly clamped fQ was 3.63 × 10¹¹ Hz for the smooth-stock device near 40 MHz. At 12 K, reference fQ values were about 7 × 10¹⁰ Hz for GaAs and 1.2 × 10¹¹ Hz for silicon. Free-free resonators produced fQ values of order 10¹² Hz for all but the devices near 100 MHz. The maximum was 1.26 × 10¹² Hz for the rough-stock resonator near 60 MHz.
The result is consistent with other NCD doubly clamped beams measured at sub-kelvin temperatures. It is comparable with bulk-mode NCD disk resonators measured at 300 K. The reported fQ exceeds cryogenic SCD cantilever values by an order of magnitude or more.
Direct comparison with coupled NCD beams at 400 to 1440 MHz is difficult. Those studies used 0.04 to 1.1 K and reported fQ values in the low 10¹³ Hz range. Prior NCD beams improved Q by about 4× when cooled from 10 K to millikelvin temperatures. Further cooling could therefore improve the present devices.
Implications for diamond device engineering
Section titled “Implications for diamond device engineering”NCD retains high Young’s modulus in a thin film that can be deposited on sacrificial layers. This reduces the integration problems associated with single-crystal diamond (SCD). Prior estimates for similarly dimensioned devices give diamond frequencies 1.6 times those of silicon carbide and 2.4 times those of silicon.
The measurements support prioritising support geometry over top-surface polishing for low-loss devices at 12 K. For rough stock, free-free supports reduced the length-dependent coefficient by close to an order of magnitude. Device design still requires control of nodal placement, support width, undercut depth, and release uniformity.
Fitted Young’s moduli were 910 ± 94 GPa for rough stock and 1072 ± 99 GPa for smooth stock. Undercuts likely bias these values downward. The authors attribute their difference mainly to sample size and device variation. The moduli indicate high stiffness but do not establish a material advantage from polishing.
Materials and support
Section titled “Materials and support”For magnetomotive resonator devices, the paper requires released nanocrystalline diamond with a Cr/Au/Ni stack. The matching 6CCVD material is MPCVD-grown polycrystalline diamond with Au metallization, as listed in the diamond materials catalog. The linked custom fabrication and engineering support page can be used to discuss the custom stack and patterned geometry.
View Original Abstract
Nanomechanical resonators are increasingly becoming of interest across a range of applied and fundamental physics applications. Within many of these, the retention of bulk diamond’s high Young’s modulus, coupled with the compatibility with standard substrate materials, makes nanocrystalline diamond (NCD) particularly well suited for fabricating high-frequency devices. As device dimensions shrink in pursuit of ever-higher frequencies, however, dissipation from sources such as clamping and surface loss often becomes increasingly significant. To address this, a series of doubly clamped beams and clamping-loss-suppressing free-free resonator geometries were fabricated from both as-grown and chemically mechanically polished NCD. At 12 K, the free-free geometries curtailed the pronounced length-dependent loss seen in doubly clamped beams, reducing dissipation by up to 8.8× and achieving Q factors of the order of 10,000 from ~40 MHz to ~100 MHz. Minor differences in dissipation between devices fabricated from the as-grown and polished stock, meanwhile, suggest that surface-related loss is likely a minor contributor to dissipation at this temperature, contrasting with trends in alternative material counterparts. As such, the combination of NCD’s apparent low surface-related loss and the loss-scaling suppression offered by free-free geometries provides a promising route to minimising dissipation in high-frequency nanomechanical resonators.
Electronic Grade Intrinsic Diamond Epitaxial Wafer H Terminated Superior 100 Mosfet Ready Substrate
View Specs →Cvd Diamond Coating On Silicon Wafer
View Specs →