Second flexural and torsional modes of vibration in suspended microfluidic resonator for liquid density measurements

Suspended microfluidic resonators enable detection of fluid density and viscosity with high sensitivity. Here, a two-legged suspended microchannel resonator that probes pico-litres of liquid is presented. The higher resonant modes (flexural and torsional) were explored for increased sensitivity and resolution. Unlike other reported microchannel resonators, this device showed an increase in the quality factor with resonant frequency value. The performance of the resonator was tested by filling the channel with three liquids, one at a time, over a density range of 779−1110 kg m−3 and a viscosity range of 0.89−16.2 mPa s. The highest resolution obtained was 0.011% change in density. Measurements with torsional mode showed an improvement of about six times in sensitivity and about fifteen times in resolution compared to the first flexural mode. When the empty channel was filled with liquids of different viscosity, the quality factor of the first flexural mode remained overall constant with a variation below 3.3% between the fluids, and confirming the inherent property of suspended microchannel resonators. However, it significantly decreased for second flexural and torsional modes. No noticeable difference was observed in the quality factor between different liquid viscosities for all modes.


Introduction
The ability to accurately measure density in small volumes is crucial in variety of fields, e.g. biomedical research [1], 1 Current affiliation: DICEA, Polytechnic University of Marche, Ancona,

Italy
Original Content from this work may be used under the terms of the Creative Commons Attribution 4.0 licence. Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI. chemical industry [2] and petroleum applications [3]. For large volumes (>1 ml), density is typically measured by weighting a known volume of liquid. The viscosity can be obtained by capillary viscometry or by falling sphere inside the liquid [4]. For volumes around 1 ml, resonant based techniques, where change in the resonant frequency as a function of liquid density can be used, like vibrating glass tubes [5], surface acoustic waves [6,7], tuning forks [8], quartz crystal microbalance [9] or thin film bulk acoustic waves [10]. For volumes around 1 µl, again resonant based methods that use microtube resonators [11], microcantilever resonators immersed in liquid [12] [20,21]. Reprinted from [20], Copyright (2017), with permission from Elsevier. or microtube resonators experiencing Coriolis force [13,14] can be used. The vibrating signals are highly damped [15] if the sensor is immersed inside the liquid, significantly effecting the sensitivity [16]. However, by flowing liquid through the sensor, vibration damping can be significantly reduced and the sensitivity can be improved [17]. The energy dissipation in the vibrating structures typically is a function of viscosity, thus enabling to simultaneously obtain density and viscosity of a liquid by monitoring frequency shift and quality factor respectively.
Here, we use a microfluidic cantilever based resonant sensor which uses only picolitre (pl) volumes of liquid. The fluid was flown through the channel inside the cantilever. The cantilever itself was operated in vacuum, thus removing also air damping on the resonant sensor. Such suspended microchannel resonators were used earlier to measure density and viscosity of fluids [18,19]. However, higher resonant modes were not investigated in such devices to monitor fluid properties. We address that aspect in this work.
The device used for the measurement is a two-legged microfluidic cantilever. The unique aspects of the device are: it is made out of transparent silicon dioxide, uses only 12 pl of volume in the sensing region, and the chip size is made to fit in any commercial atomic force microscope (AFM) setup. The first two flexural modes and the first torsional mode of vibration were measured and analyzed.

Device Details
Hollow cantilevers were fabricated using a process originally described elsewhere [22,23]. The fabrication method involved three main steps: i) two silicon wafers with fluid reservoir etched in the top wafer and cantilever channels etched in the bottom wafer were bonded together; ii) the channels and the reservoir were oxidized; iii) the bottom wafer was selectively etched to form on-chip channels and a suspended hollow cantilever, all made out of silicon dioxide. Simultaneously, the perimeter of the chip was etched and made to attach over breaking beams with the wafer. Finally, individual microfluidic cantilever chips were released from the wafer by applying a gentle manual pressure.
The microfluidic chip with fluidic network and the suspended cantilever is shown in figure 1(a) [22]. The cantilever is connected through on-chip microfluidic channel Chip glued to a 3D printed plastic fluidic interface that had two needles glued on either side for the fluid inlet and outlet. The entire assembly was mounted on a Nanosurf cantilever holder and clamped with a spring. network to two reservoirs on the chip, see figure 1(b). These reservoirs were used as interface with the external world. They were used as inlet and outlet or vice-versa. The dotted lines in the figure indicate the microfluidic channel located on the other side of the chip. The cantilever and the on-chip channels were made out of silicon dioxide. The cantilever had two physically separated channels connected near the free end ( figure 1(c)). There was also a hollow pyramidal tip (without any aperture) at the free end, which had no function for the results described in this work. The cross-section of the fluid channel is shown in figure 1(d). The dimensions of the hollow cantilever and the chip are given in table 1.

Fluidic interface
The fluid entered through one of the reservoir (450 µm × 450 µm), flowed through the on-chip fluid channel, suspended cantilever and exited from the other reservoir. The on-chip reservoirs were able to continuously exchange fluids with the macro world through a fluidic interface. The plastic interface was fabricated by a 3D printer with 25 µm resolution (Envi-sionTEC Micro Plus HD) using acrylic resins. The reservoirs on the chip were precisely aligned to the reservoirs on the interface before they were glued together. The chip glued to the fluidic interface is shown in figure 2. Two disposable stainless steel needle tubes (Nordson, 32GA needles) were glued to the plastic fluidic interface. Their free ends were connected to Tygon tubing which had an inner and outer diameter of 0.25 mm and 1 mm, respectively. The fluids were injected from the external world into the hollow cantilever through the tygon tube.

Experimental setup
The resonator chip glued to the plastic fluidic interface was mounted on a standard atomic force microscopy cantilever-holder from Nanosurf AG, where excitation piezo was mounted close to the cantilever chip location. The holder was placed in a small vacuum chamber in order to reduce damping by air and increase the quality factor. The chamber had fluid feed-throughs to inject or aspirate fluids, electrical feedthrough to excite the piezo and gas feed-through to evacuate air. The schematic of the setup with various components is shown in figure 3. The connection between chamber and vacuum pump also accommodated a leak valve to control the pressure inside the chamber and a Pirani gauge for pressure readout. A glass window on the vacuum chamber provided the optical access to the cantilever. A laser Doppler vibrometer (Polytec MSA-400-PM2-D) was used to monitor the resonance frequencies and evaluate quality factors. Using the built-in microscope of the vibrometer, the laser was positioned at the desired location on the cantilever.

Device characterization
To ensure proper reflection of the laser from the transparent surface of silicon dioxide, the chip was coated with 6 nm of Au/Pd. The excitation was provided in the linear cantilever vibration regime by employing wave excitation from the base. The force signal used in the experiments were of pseudo-random type. This excitation eliminated the leakage problem that a random excitation commonly encounters. The signal was a stationary ergodic, with random amplitude and phase distribution. The alignment grooves on the holder ensured reproducible clamping for multiple experiments. The resonance frequency and quality factor were extracted from the frequency-amplitude response in a linear regime. The quality factor was obtained from the frequency-amplitude response by fitting a Lorentzian and using peak frequency and full width at half maxima. The resonant frequencies and corresponding quality factor values in vacuum are given in  The fundamental frequency of this device, in air, was measured to be 134.53 kHz. The second flexural and the first torsional modes have also been obtained and their resonance frequencies were 828.25 kHz and 916.67 kHz, respectively.
The cantilever channels were filled with air, while the surrounding air outside the device was evacuated until 2.8 mbar (with mean free path of 36.75 µm) 4 . As a vacuum developed in the chamber, the resonance frequency for the first flexural mode of the device increased as shown in figure 4(a), 4 The mean free path was calculated considering effective cross-section for collision of air molecules of 2.82743 × 10 −19 m 2 at T = 298 K [24]. 4(b). This increase was more prominent for the second mode (figure 4(c)). A similar trend was observed for the torsional (twisting) mode as shown in figure 4(d). This was consistent with the definition of the damped resonant frequency [25]: where ω n is the natural frequency and τ = 1/(2Q) is the damping ratio. As the pressure decreases, the damping ratio reduces and the resonant frequency increases as shown in figure 4(a). When the effect of air damping was lowered, at 2.8 mbar, the quality factor for the first flexural mode improved to 1700 (≈ 10 times increase) and to 5200 and 8200 for the second flexural and first torsional mode, respectively (≈ 20 times increase from air). Note that the quality factor has not yet saturated indicating that it can be improved by further reducing the pressure level [26,27].

Density measurement using higher modes
After the modal testing of the microfluidic cantilever in air and vacuum, the cantilever was filled with three liquids of different density and viscosity. According to Khan et al [19], the shift in the inverse square of the resonant frequency (ω n ) is directly proportional to fluid density (ρ) changes inside the microfluidic cantilever. Mathematically, this can be expressed as: where A and B are constants that can be determined by conducting experiments with two fluids of known densities. The three fluids along with their viscosities and densities at 20 • C are tabulated in table 2. All experiments were performed at room temperature, 20.25 ± 0.25 • C. The fluids were injected/aspirated through the microfluidic cantilever using a syringe pump (Legato from KD Scientific) with a flow rate of 5 µl/min. Note that the volume inside the resonant probe region is only 12 pl. However, the connecting tube, device interface and the on-chip reservoir together require about 100 µl volume of fluid. To ensure complete exchange of liquids inside the channel, they were injected for about 3-4 min at the rate of 5 µl/min. The measurements were conducted after 10 min to ensure everything was stabilized. For the viscous ethylene glycol (about 16 times more viscous than water), an additional syringe pump withdrawing the fluid from outlet reservoir at a rate of 5 µl/min was used along with injection at the same rate from the inlet reservoir. After every experiment, the channels were rinsed with water.
The fluid flow in microfluidic channels was laminar. The Reynolds' number of the fluid in our device varied from 0.26 (air/first mode) to 31.2 (ethanol/torsional mode). All the values for various liquids and modes of vibration are listed in table A2 in the appendix (B).

Results and discussion
All the results were obtained with cantilever mounted inside the chamber. The chamber was evacuated to a pressure of 2.8 mbar. The channels of the cantilever were either empty (i.e. filled with air) or filled with water, ethanol or ethylene glycol.

Effect of vacuum outside hollow cantilever
The characterization of the empty hollow cantilevers in vacuum showed an improvement in quality factor of about an order of magnitude compared to air (see table 3). The quality factor of the first mode was approximately 15 times lower compared to the quality factor report by Khan et al [19]. However, their devices were operated at a pressure 10 5 times lower than the pressures of our setup. The sensors used by Lee et al [28] had 8 times higher quality factor for the first mode in high vacuum than our device. Their reported value of 8000 for first mode is close to the first torsional mode of our device.
The short term frequency stability of our device was tested by repeated measurements of the resonance peak at time intervals of one minute. The measured resonance frequencies for different modes are plotted in figure 5 as function of time and in terms of probability distribution. From the histogram plots and the fitted probability distribution we document the improved stability for the second flexural and first torsional mode. This confirms that density can be estimated more accurately at higher modes provided sensitivity is also higher at those modes [29]. Note that we did not had any temperature control for our sensors (temperature of the clean room was maintained at 20.25 ± 0.25 • C).

Effect of fluid density on frequency
Upon filling the microfluidic cantilever with a particular fluid, the flow was stopped and the measurements were made. The resonance frequency decreased corresponding to the increase in density of the fluid. The frequency shift obtained for three different liquids for the first flexural mode is shown in figure 6. The normalized amplitude response and the corresponding phase as function of the excitation frequency is plotted. The position of the laser probe on the cantilever is indicated in the inset. Even though this position was not optimal for analyzing the torsional mode, there was a sufficient signal level. The effect of fluid density on frequency for the higher modes is reported in appendix (A).
The resonance frequency shift as function of the fluid density for all modes is shown in figure 7. For each fluid we performed three experiments, the values for all the frequency shifts and the calculated standard deviation are reported in appendix (C).
We did not observe significant difference in the experimental results for repeated filling and emptying of the channels. The results were found perfectly reproducible.
There was a good linear fit (dashed lines) with Pearson's correlation coefficient of −0.99986, −0.99934, −0.99978 for the first, second and torsional mode respectively. In figure 7 solid red lines represent the fit according to equation (2). The coefficients A and B are calibrated by measuring the resonance frequency when filled with glycol and air. The values  for the sensitivity derived from these plots are summarized in table 4. It is evident that higher mode measurements showed an increase in sensitivity with simultaneous improvement in the resolution. The change in density that was detectable using the first torsional mode (0.011%) in pl volume was better than the value of 0.06% in 5 µl volume that was detected using Mode 17 of cantilevers immersed in liquid [16]. The sensitivity levels of 57 Hz (kg m −3 ) −1 in torsional mode and 63.01 Hz (kg m −3 ) −1 for the second flexural mode was 3.6 times and 3.9 times higher respectively compared to 16.3 Hz (kg m −3 ) −1 reported by Khan et al [19], where they monitored only fundamental mode during their measurements. Therefore, monitoring higher modes appear to increase the sensitivity in our device. The measurement system, i.e. laser spot, had no influence in the frequency shift for the filled cantilever. The results were reproducible for next few days. However, a slight change in the frequency was observed if the cantilever was removed and re-positioned. It was due to the irreproducible clamping condition.

Effect of viscosity on the quality factor
The energy dissipation in the microfluidic resonator is a combination of several components of loss that are in general challenging to disentangle [30,31]. To simplify the dissipation in our system, we used the following equation: where Q total is the overall effective quality factor, Q empty is the quality factor of an empty (channels filled with ambient air) hollow cantilever vibrating in vacuum with all the dissipation factors (like clamping, intrinsic losses, etc) and Q fluid is the dissipation factor due to fluid inside the channel. The quality factors obtained experimentally for all the modes when channels are filled with different fluids are given in table 5. The parameter Q total is the experimentally measured value and Q fluid is the value calculated from equation (3). The total quality factor Q total for different liquids measured at different modes is plotted in figure 8. There are several observations from the data. a) Q total increased for second flexural and torsional mode compared to first flexural mode. b) Q fluid decreased for second flexural and torsional mode compared to first flexural mode. c) Q total was almost independent of fluid inside the channel for the first flexural mode. The variation between fluids was less than 3.3%. The small variation of the quality factor between fluids with different viscosities is within the framework of the non-monotonic variation predicted in Burg et al [32] d) Q total for the second flexural and torsional mode significantly decreased when the empty channels were filled with liquids. The decrease from air to water was of 19%, 66% and 75% for the first, second flexural and torsional mode, respectively. e) Once filled with liquids, Q total did not change significantly for different liquids and modes. The maximum variation was when filled with Glycol with a ∆Q total of 360 (see second and  third column of table 5). f) We observed a standard deviation of Q total for ethanol for first flexural mode to be an order of magnitude lower with respect to other liquids. We could not explain this phenomenon and needs additional investigation. The analytical equations to estimate Q fluid for hollow cantilever filled with fluid was derived by Sader et al [33]. The calculated numbers are reported in appendix B. As predicted by the theory, Q fluid of our device decreased at second flexural mode. This means the energy dissipation contribution from fluid is higher at the second flexural mode compared to the first flexural mode. This is attributed to the enhancement of the fluid compressibility at higher modes [33]. Additionally, due to fabrication limitations in producing perfectly symmetric devices along the channel height, the effective center position of the beam could be slightly off-axis. This results in net axial strain being experienced by the fluidic channel as the beam vibrates, leading to pumping of fluid inside the channel. This leads to additional acoustic effects contributing to damping. Interestingly, the Q fluid for first torsional mode also shows higher dissipation than first flexural mode. Here, the legs are vibrating out of phase which can create additional fluid velocity gradients contributing to energy dissipation.
Even though Q fluid seem to predict as estimated by the theory, the interesting part is the Q total for our device. The measured Q total increased for second flexural and torsional mode irrespective of the fluid, however, it was significantly higher for empty cantilever. This indicates that overall energy dissipation is lower at higher modes despite higher damping due to Q fluid at higher resonant modes due to fluid. The energy dissipation due to effective inertia from device is dominating compared to viscous effects at higher modes. This is in contrast to the observations made by others [33,34] and this could be either due to insufficient vacuum level outside the cantilever or the unique device geometry. The mean-free-path of air molecules, 36.75 µm at 2.8 mbar vacuum, is comparable to the dimensions of the cantilever, thus perhaps molecular viscosity of air is still playing a role. It is also known that quality factor increases with mode number in ambient air [29]. The geometry of our device is different from that used by Sader et al [34]. Our device has two channels physically separated (two legged), whereas their channels were physically attached as a single cantilever.
The Q total of fundamental mode was almost independent of the liquid that was filled. This indicates that device inertial dissipation is very dominant over liquid viscous dissipation even over two orders in magnitude change in viscosity of fluid inside the channel. From theory, a non-monotonic behaviour of energy dissipation due to viscosity change was anticipated [32].
Overall, the total energy dissipation in our device is inertial dominated at higher frequency modes including torsional mode. The dissipation due to fluid has increased with higher frequency modes as predicted by theory, but overall still the inertial dissipation dominates. Further experiments would be to increase the vacuum level until the quality factor saturates and investigate the liquid damping behaviour of the device. Table 5. Quality factors for the first two flexural and the first torsional mode when the microfluidic cantilever is empty and filled with ethanol, water and glycol. The standard deviation (SD) is given in the parenthesis. The Q fluid for air is orders of magnitude higher.

Conclusion
In this work, we present a physically separated two-legged hollow cantilever device. It has a quality factor of about 1700 for the first flexural mode of vibration at 2.8 mbar surrounding pressure. These quality factors are found to increase with mode number. For the first time, torsional mode in hollow cantilevers is measured. The first torsional mode exhibits a quality factor of about 8200, which is comparable to the quality factor of devices used in state-of-the-art cantilever based mass sensing applications. When the pl volume channels were filled with different liquids, resonance frequency shift increased with mode number for two flexural modes and the torsional mode. The respective sensitivities in density measurement for the three modes were 10.49 Hz (kg m −3 ) −1 , 63.01 Hz (kg m −3 ) −1 and 57.16 Hz (kg m −3 ) −1 , respectively. There is also an improvement in resolution from 1.58 kg m −3 to 0.11 kg m −3 at higher modes of vibration which make the latter more attractive for sensing. The quality factor decreased when the channels were filled with liquid. However, there was not much change in the quality factor for different liquids at all modes. Furthermore, the quality factor increased for second flexural, which is contrary to the observations of others. The torsional mode had higher quality factor than flexural mode. fluid only is calculated following the formulation given Sader et al in [33]. The results are provided in table A1. The Q due to the fluid strongly depends on the nature of the fluid and on the where ρ and µ are the density and shear viscosity of the fluid with thickness h fluid , respectively. The quality factor due to the fluid is where F(β n ) is the normalized quality factors and expressed by: The channels of our device are considered to have neutral axis at the center, thus the function G(X, Z) consists of only one term depending on the normalized deflection function of the cantilever beam and it does not account for asymmetric distribution of the inner channels (see [33] for further details).

Appendix C. Resonance frequencies for the different liquids
In order to show the accuracy and repeatability, in table A2 we report the values of the resonant frequencies for three separate experiments.