Mechanical characterization and cleaning of CVD single-layer h-BN resonators

Hexagonal boron nitride is a 2D material whose single-layer allotrope has not been intensively studied despite being the substrate for graphene electronics. Its transparency and stronger interlayer adhesion with respect to graphene makes it difficult to work with, and few applications have been proposed. We have developed a transfer technique for this extra-adhesive material that does not require its visual localization, and fabricated mechanical resonators made out of chemical vapor-deposited single-layer hexagonal boron nitride. The suspended material was initially contaminated with polymer residues from the transfer, and the devices showed an unexpected tensioning when cooling them to 3 K. After cleaning in harsh environments with air at 450 °C and ozone, the temperature dependence changed with f0Q products reaching 2 × 1010 Hz at room temperature. This work paves the way to the realization of highly sensitive mechanical systems based on hexagonal boron nitride, which could be used as an alternative material to SiN for optomechanics experiments at room temperature. An improved transfer method allows easy placement of highly transparent and strongly adhesive hexagonal boron nitride on target substrates. A team led by Santiago J. Cartamil-Bueno at Delft University of Technology developed a technique that enables the transfer of large-area, single-layer hexagonal boron nitride films grown by chemical vapor deposition onto a substrate of choice, whilst not requiring optical visualization. Following an additional cleaning step, the atomically thin membranes were transferred onto circular microcavities patterned on a silicon oxide substrate, resulting in the formation of suspended drums. Cleaning in harsh environments using a mixture of air and ozone is instrumental to a substantial improvement in the quality factor of the drums, indicating that undesired contamination causes damping of the mechanical motion. These results show promise for the development of sensitive hexagonal boron nitride resonators.


INTRODUCTION
Hexagonal boron nitride (h-BN), also known as 'white graphite', is a layered material consisting of partially-ionic sp 2 -bonded boron and nitrogen atoms in a honeycomb arrangement with a van der Waals and slightly ionic AA′ stacking. 1, 2 h-BN is a III-V indirect wide-bandgap semiconductor (5-6 eV) 3 that has been proved to be the complement insulating material for graphene electronics. 4 The larger chemical and thermal stability of bulk h-BN compared to graphite have enabled applications in harsh environments such as deep ultraviolet photodetectors for space research 5,6 and highperformance lubricants. 7 Among all the 2D materials, few studies on single-layer h-BN exist 8,9 despite the early attempts to exfoliate it 10 and the welldeveloped synthesis techniques. [11][12][13][14][15][16][17] The reason for its difficult exfoliation to a monolayer thickness is unclear, possibly being its AA′ stacking, its slightly-ionic bonds or flatness that causes adhesion between h-BN layers and other surfaces stronger than other 2D materials. [18][19][20][21] Moreover, h-BN surfaces are hydrophillic in nature, but after being exposed to air they get covered by an hydrophobic layer of hydrocarbons as reported in h-BN nanotubes 22 and graphene. 23 Understanding the adhesion, friction and other properties of single-layer h-BN can pave the way to promising applications, such as deep ultraviolet single photon emitters, 24,25 ultrathin piezotronics, 26,27 and to exploit its natural hyperbolic optical dispersion properties, 28 and proton conduction. 29 Fabrication of suspended h-BN devices has been limited to fewlayers. [30][31][32] Mechanical devices made out of 2D materials are often fabricated with a polymer-assisted transfer technique that requires a wet release of the suspended system, 4,8,29 or a dry method that exploits the viscoelastic properties of polydimethylsiloxane (PDMS). 33,34 In the case of drum devices where the membrane seals a cavity of a few hundreds of nanometers in depth, a dry transfer is more convenient as surface forces caused by liquid menisci upon drying tend to break the structure or push it to the bottom of the cavity resulting in collapsed devices.
In this work, we report the fabrication and characterization of mechanical resonators made out of chemical vapor-deposited (CVD) single-layer h-BN. The fabrication includes a novel transfer technique suitable for extra-adhesive 2D materials and highly efficient cleaning process for h-BN resonators. Due to the transfer process, the single-layer h-BN devices were covered with polymer contamination and showed an unexpected tensioning when cooling them to 3 K. After cleaning processes in harsh environments, the temperature dependence of their mechanical behavior changed, and their figure of merit (f 0 Q product) improved by a factor of 30 up to 2 × 10 10 Hz at room temperature. Figure 1a shows the fabrication process of suspended h-BN devices. A continuous layer of CVD h-BN is grown on iron foils and characterized by transmission electron microscopy (TEM) to be single layer as previously reported. 17 Due to the stronger adhesion of h-BN to surfaces, it has to be transported from the growth substrate to the final substrate by using sacrificial layers. Poly (methyl methacrylate) (PMMA) is spin coated on top of the h-BN, and the stack is delaminated with the bubbling technique: 35 a voltage of 3 V is applied between the Fe/h-BN/PMMA stack (cathode) and a Pt wire (anode) of an electrochemical cell with 1 M NaOH (electrolyte). From scanning electron microscopy (SEM) images taken before the delamination, the grain size is estimated to be around 20 µm. The h-BN/PMMA stack is then transferred onto a piece of silicon covered with 300 nm of poly(vinyl alcohol) (PVA), and the PMMA is removed by an acetone bath, resulting in a Si/PVA/h-BN structure.

Fabrication of h-BN microdrums
Using this sandwiched structure, we developed a dry transfer technique that is a hybrid of the methods described in refs. 4, 33 especially designed for the transfer of CVD h-BN to make sealed cavities. Figure 1a (steps vi-xii) describes the dry transfer process. A piece of PDMS (GelPack × 8 PF film) is placed on top of the Si/ PVA/h-BN, and the stack is left floating in deionized water. The water dissolves the PVA layer slowly until the h-BN/PDMS detaches from the silicon substrate which sinks. The floating h-BN/PDMS is picked up with tweezers and dried with nitrogen, and then placed on a clean microscope slide. From this point, we continue with the technique from reference: 33 the h-BN is dry transferred deterministically by using the viscoelastic properties of PDMS on top of the target substrate. A silicon substrate with 300 nm of thermally-grown SiO 2 is patterned via e-beam lithography and reactive ion etching to make circular cavities of different diameters on the oxide layer (fully etched through to the silicon). By using the developed transfer technique, we cover the cavities with CVD single-layer h-BN resulting in drums as shown in Fig. 1b. Some of these suspended membranes have small perforations, wrinkles and polymer contamination (see Supplementary  Information). Figure 2a shows the Raman spectrum with the characteristic peak at 1370 cm −1 , 9,36 which supports the TEM characterization on its single-layer nature. Moreover, the thickness and topographic flatness are measured with an atomic force microscope (AFM, Bruker Fastscan) in tapping mode (FastScan-A probe). As observed in Fig. 2b, the AFM indicates a thickness of 5 nm which is larger than that of a monolayer (0.44 nm), probably due to polymer contamination that disappears in some areas after cleaning processes as seen in Fig. 2c (more details in Supplementary Information).

Mechanical characterization
The static mechanical characterization of suspended CVD singlelayer h-BN is performed by AFM nanoindentation. We use the Peak Force mode (ScanAsyst-Fluid probe) to deflect drums of 5 µm in diameter with a controlled force of 5 nN. Figure 3a shows the AFM image of the suspended drum from Fig. 1b with its corresponding force-deflection curve at its center. The tension and Young's modulus of the material are extracted by fitting the data for positive forces to the following equation valid for a circular structure under a central point load F: 37 where N 0 , E and ν = 0.18 38 are the pre-tension, Young's modulus and Poisson's ratio of the suspended material (chosen as an intermediate value in the range reported in literature), respectively; t = 0.44 nm is its thickness, R is the radius of the drum, and δ is the deflection induced by the AFM tip. The Young's modulus E is extracted by fitting the large-deformation data to the cubic term, although for small applied forces the accuracy is limited. The thickness-dependent term of the linear spring constant is negligible in these atomically-thin devices as the transfer of the material usually introduces stress that dominates the response at small deformations. This membrane behavior (e.g., tension dominated suspended structures) is found experimentally in the h-BN drums as shown below. The extracted tension and Young's modulus values are shown in Table 1 for five different drums, and they are expected to be influenced by the polymer in a second order of importance. Note the large Young's modulus of an undamaged drum (device 5), which is comparable to that of pristine graphene, 39 and the lower values obtained for the other perforated devices.
The mechanical properties are also measured indirectly from the drum dynamics with a laser interferometric technique in a  cryo-station setup (Montana Instruments). A power-modulated diode laser (λ = 405 nm) photothermally actuates the drum, while a continuous-wave He-Ne laser (λ = 633 nm) allows the measurement of its resonance frequency f 0 . 34 Figure 3b displays the frequency spectrum of the same drum shown in Fig. 3a at a room temperature and pressure of~10 −6 mbar, showing the fundamental mode around 13 MHz and the second (splitted) mode around 1.5 times the fundamental one. The factor of 1.5 indicates that the single-layer h-BN acts as a membrane in which pretension dominates over the bending rigidity. 37 The fundamental mode is given by: where ρ = 2100 kg/m 3 is the mass density of the suspended material. By fitting the fundamental peak to the response function of a harmonic oscillator, we extract the resonance frequency that is used to calculate the tension from Eq. 2. The pre-tension values are shown in Table 2.
Temperature dependence of the resonance frequency Figure 4 shows the temperature dependence of the fundamental frequency when cooling the device from 300 to 3 K. Equation 2 has to be modified to take into account the temperature-induced tensioning of the membrane due to the in-plane expansion coefficients of the 2D material and the underlying substrate dominated by the thermal expansion of the Si: (3) where α Si and α 2D are the temperature-dependent expansion coefficient of silicon and the 2D material, respectively, in the basal plane; ΔT = T−300 K is the temperature difference from room temperature. N sidewall is a new tensioning mechanism that originates from the adhesion of the suspended 2D material to the sidewalls of the circular cavity, which happens when the membrane tries to expand laterally inside the cavity. This adhesion could balance out (N sidewall = 0) 40 or increase the tension due to a biaxial in-plane strain (N sidewall (T) = Et6tL(T)/R 2 ∝ α 2D ΔT, where L is the change of adhered length on the sidewall). 41 When cooling down the device made out of single-layer h-BN with negative inplane thermal expansion coefficient (α Si (T) > α h-BN (T)), we would expect a constant or monotonic increase of f 2 , i.e., f 2 0 ðTÞ À f 2 0 300K ð Þ∝T with a zero or negative slope. The data, red circles in Fig. 4, show such a trend suggesting that N adhesion = N sidewall . However, another possible explanation for this could be residual resist remaining on top of the h-BN after transfer.

Impact of cleaning processes
To test this hypothesis, we clean the membranes by annealing the samples at 450°C for 3.5 h in an air environment (78% N 2 , 21% O 2 ), and in a further cleaning step using ultraviolet (UV)-induced ozone environment for 30 min. The oven annealing is expected to cause thermal and chemical degradations of polymers (boiling and evaporation of PVA, oxidation and gasification thermolysis of PMMA), and the UV ozone cleaning should produce photoinduced and chemical decomposition of polymers (photolysis of remaining PVA/PMMA and ozonolysis of PDMS) and decomposition of other organic matter with the UV ozone cleaning. Despite these cleaning processes in harsh environments, the h-BN remains suspended and the material becomes invisible under the optical microscope as expected from its low absorption of visible light, 9 thus indicating that the contaminating polymer layer has been largely removed. After the treatments, the fundamental mode frequency at room temperature increased for all drums, although slightly. We also repeated the measurement as a function of temperature as shown in Fig. 4 by the blue (after oven annealing) and black data points (after ozone cleaning). More details about the temperature dependence of the mechanical properties of the h-BN drums can be found in Supplementary Information together with the complete dataset. The conclusion from the data is that when cooling a slight increase in the resonance frequency remains, although the effect is smaller than before annealing.

DISCUSSION
References 8 and 16 report the mechanical characterization by AFM indentation of similar devices made out of CVD multilayer h-BN (1-2 and 5-15 nm in thickness, respectively). In those works, the material is placed on top of cavities with different transfer techniques that also introduce polymer contamination and wrinkles as we have observed in our devices by SEM and AFM. The Youngs modulus reports (220-250 and 1160 GPa, respectively) are comparable to our measured values at room temperature, and in particular device 5 proves to have a stiffness close to the theoretical value of single-layer h-BN (885-995 GPa). Mechanical characterization and cleaning of CVD SJ Cartamil-Bueno et al.
In another work on graphene, 42 the temperature-dependence of the Youngs modulus was proven to be strongly influenced by the presence of wrinkles in the suspended membranes. More importantly, the previous works on h-BN present membranes that sag into the cavities due to the van der Waals forces between the material and the sidewalls. Considering that the 2D Youngs modulus scales with the thickness, it is expected that single-layer h-BN shows larger sagging than previously reported (20-50 and 15 nm, in the mentioned works). AFM scans of our devices show sagging of 50-60 nm, thus supporting the monolayer nature of the material. Bulk h-BN has a negative in-plane thermal expansion coefficient for temperatures higher than 75 K. 43 We assume that single-layer h-BN exhibits the same tendency as graphene with respect to graphite, i.e., that it has the same negative temperature dependence as in the bulk, but with a larger magnitude. 44 The observed tensioning must therefore originate from (a) an unexpected positive in-plane thermal expansion coefficient for h-BN as a single layer caused by the presence of polymer contamination even after cleaning, 45 (b) other tensioning mechanism such as the proposed sidewall adhesion or the presence of wrinkles, or (c) defect creation during UV ozone cleaning.
When comparing the tension values from the AFM and the laser interferometric studies as shown in Tables 1 and 2 (before  cleaning), we observe a difference of one order of magnitude between the two. The indirect calculation of the tension from Eq. 2 includes the mass of the membrane as a parameter as opposed to the direct measurement, where the mass does not appear in Eq. 1. In fact, the thickness of the material measured by AFM before cleaning (5 nm, about 10 times that of single-layer h-BN) may explain the difference in values obtained from the AFM indentation and the laser interferometric techniques. The spread in values obtained with both techniques can also be justified with residues of inhomogeneous thickness and hence mass. In addition, the quality factors increased significantly with respect to the pre-cleaning values (see Supplementary Information), suggesting that the polymer residues play a significant role in damping the mechanical motion, probably due to their viscoelastic properties.
On the other hand, the increase of the fundamental frequencies at room temperature after cleaning ( Table 2) indicates that either the oven annealing introduced larger pre-tensions (enhanced adhesion to the surface or the sidewall of the cavity) and/or a decrease in the mass (hence obtaining tensions more similar to those obtained from the AFM) according to Eq. 2. The mechanical quality factor of the resonators improved after each of the two cleaning steps (see Supplementary Information). This fact coupled to the disappearance of traces of polymer contaminants and the on-average increase in the resonance frequency of the resonators suggests that polymer contamination is the main damping mechanism encountered in these resonators. It is unclear if there remains some polymer residue, although the tensioning trend when cooling has been reported in graphene and several transition metal dichalcogenide membranes. [46][47][48] However, this trend should be explained differently for 2D materials with a negative expansion coefficient like h-BN or graphene, 49 and for that reason we propose an alternative tensioning effect caused by the sidewall adhesion, which should balance out or increase the tension when cooling certain 2D membranes, and explains our results and those on graphene drumheads. Other stronger cleaning routes for h-BN are available, 50,51 although not all of them might be compatible with suspended single-layer structures.
This study has presented the fabrication, mechanical characterization, and cleaning of CVD single-layer h-BN drumhead resonators. We have developed a transfer technique that allows the transfer of this extra-adhesive material from the growth substrate to a patterned SiO 2 surface with circular microcavities, resulting in suspended h-BN drums. The tension and Young's modulus of the material were characterized by performing AFM indentation on several devices, and the tension values are compared to those measured indirectly from the resonance frequency by means of laser interferometry. The temperature dependence of the dynamics of the drum is realized by cooling the devices to 3 K, and corroborates the assumption of having polymer residues covering the h-BN that dominates the tensioning of the devices. After cleaning in harsh environments, we observe a substantial improvement in the quality factor of the drums, and a change in their temperature-induced tensioning, which demonstrates cleaning of the suspended material. However, the temperature behavior suggests that some residues may still be present or that the structural mechanics model needs to be revisited for 2D materials, thus requiring further research. The strong adhesion of h-BN layers compared to other 2D materials could be exploited to make highly sensitive mechanical systems. Its electrical and optical properties are different than those from graphene, hence being a desirable material for applications where low-optical absorption or electrical insulation is required. Its nearly-zero extinction coefficient above 500 nm and good reflectivity could be used to make large f 0 Q NEMS for optomechanics experiments. Moreover, it could be combined with other 2D materials to fabricate NEMS à la carte with new functionalities for transparent and flexible electronics.

Raman spectroscopy
Raman spectroscopy on the h-BN material over Si/SiO 2 was carried out with a Renishaw in Via Raman microscope. Measurements with a laser light of 514 nm were performed on supported areas. The backscattered light was collected with a ×100 objective (NA = 0.95) and recorded with a 1800 lines/mm grating, which gives a spectral resolution of~1.75 cm −1 . The laser power was about 25 mW.

Atomic force microscopy
A Bruker Fastscan is used in tapping mode for imaging the topography and having a rough measurement of the material thickness (FastScan-A probe). For the mechanical characterization of suspended CVD single-layer h-BN, the AFM is calibrated and used in Peak Force mode for nanoindentation tests (ScanAsyst-Fluid probe, Peak Force setpoint of 5 nN).

Laser interferometry
The frequency response of the resonators is measured using a customized laser interferometer integrated into cryo-station (Montana Instruments). Two lasers are focused on the suspended drumheads with a microscope objective lens (Mitutoyo ×100, NA = 0.75). The suspended material is Fig. 4 Temperature dependence of the mechanical resonance frequency (fundamental mode) averaged for the h-BN drums before cleaning, after oven annealing, and after UV ozone cleaning. By cooling the drums from 300 K down to 3 K, the material is tensioned resulting in higher frequencies. Despite the expected negative inplane thermal coefficient of h-BN, no compression is observed even after cleaning, hence suggesting that other tensioning mechanisms are present Mechanical characterization and cleaning of CVD SJ Cartamil-Bueno et al.
photothermally actuated with a modulated blue diode laser (Thorlabs LP405-SF10, λ = 405 nm) at a power of 260 µW, while a second continuouswave linearly-polarized Helium-Neon laser (λ = 632.8 nm) at 13 mW is used as optical probe for the detection. 34 The reflected red light is captured by a high speed photo-detector (NewFocus 1601) whose signal is crossed with the driving signal of the diode laser connected to a network analyzer (Rohde & Schwartz, ZVB4). The measurements are carried out in vacuum (10 −6 mbar) for room temperature measurements, and higher vacuum when cooling.

Cleaning procedures
After fabrication and first measurements, the membranes are annealed at 450°C for 3.5 h in an air environment (Lindberg/Blue tube furnace). After the immediate new round of measurements, we use a Novascan PSD-UVT decontamination system at room temperature to clean the samples using UV-induced ozone, which precedes another immediate round of measurements.

Data availability
The authors declare that the data supporting the findings of this study are available within the paper and its supplementary information files.