Fabrication and development of novel micromachined parylene-based electroactive membranes with embedded microfluidic architectures

This work describes the design, fabrication, modeling, and testing of monolithic micromachined parylene-based electroactive membranes (µPEMs) with embedded microfluidic channels. The design and modeling employed analytical plate theory to determine the optimal membrane dimensions and structural shapes for various microsystem designs. The µPEMs were fabricated using a combination of surface and bulk micromachining techniques incorporating Parylene C as a biocompatible polymeric structural material combined with patterned electrodes for actuation. Experimental actuation of the electroactive membranes demonstrated reliability with minimal voltage shifts, and theoretical pull-in voltages closely matching experimental results. Different structural parameters of the µPEMs were also tested, such as varying the overall membrane thickness/structural rigidity and actuation chamber depth. Dynamic actuation of the membrane, including, the deflection and system response to various actuation frequencies, was observed and quantified via optical coherence tomography techniques. Microfluidic architectures were monolithically integrated with the membrane actuator and successfully perfused, with no signs of leakage. This compact microsystem has potential applications in microfluidics and Lab/System-On-a-Chip devices, for use in micromixers, particle manipulators, and applying strain to adherent cells cultured on top of the membrane.

microshell structures, dynamic motion measurements, Lab-On-A-Chip, System-On-A-Chip (Some figures may appear in colour only in the online journal) * Author to whom any correspondence should be addressed.
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.

Introduction
Microsystems platforms for System-On-a-Chip (SOC) or Lab-On-a-Chip (LOC) devices often require the integration of multiple microfluidic functions, such as pumping and actuation mechanisms [1], with their respective micromachined structures and components. Currently, many systems are fabricated using soft-lithography techniques employing microreplica molding of Polydimethylsiloxane (PDMS) on SU-8 relief microstructures [2]. For these approaches, it is common to use pneumatic or magnetic actuation due to their compatibility with the deformable PDMS structures [3][4][5]. However, pneumatic actuation is often used for binary actuation states of pumping or valving architectures (open or closed), and do not always allow precise actuation reproducibility and reliability. Furthermore, for magnetic actuation, it is essential to ensure proper positioning between magnetic media and microcoils while also considering the power consumption and heating generated by the micro-coils. In addition, these systems often require manual integration of their components utilizing bonding techniques and connections to/from off-chip components and pneumatic connections [6], resulting in the loss of compact integration capabilities [7]. To overcome these limitations, we have explored the use of electrostatic actuation methods through the design and fabrication of a micromachined parylene-based electroactive membrane (µPEM) device presented herein. The actuation of this platform solely relies on electrostatic forces which eliminates the need for complicated interconnects and manual integration, with the potential to provide greater membrane deflection control, low power consumption, high reproducibility and compact integration capabilities within a single chip [8,9].
Electrostatic actuation dates back to the earliest MicroElectroMechanical system (MEMS) devices, especially surface micromachined polysilicon devices such as micromotors [10] and lateral comb drives [11]. However, as an alternative to using silicon as a structural material, we developed µPEMs using Parylene C, a biocompatible micromachinable polymer, as the structural material [12]. Parylene is much more compliant and ductile compared to silicon and therefore can produce higher actuation stroke lengths without fracture or failure. Parylene is deposited via chemical vapor deposition which can also be masked with photoresist via lithographic patterning and etched using an oxygen plasma [13]. These micromachining approaches are similar to techniques which have been previously reported to create Parylene microactuators for peristaltic pumping [14], microvalves [15], and spring-like structures for digital-toanalog converter MEMS devices [16].
In order to demonstrate novel integration of µPEMs with microfluidics, we present a fabrication method for coupling parylene based electrostatic actuators with embedded microfluidic channel architectures using a combination of surface and bulk micromachining techniques, through the controlled deposition of metal and polymer thin films. This fabrication process can also be used to develop other polymer-based integrated microsystem designs. Following fabrication, the µPEMs are packaged into a chip package or mounted onto a printed circuit board (PCB) and evaluated by electrical characterization, including determining pull-in voltage [14], evaluating sequential and continuous actuation, DC and AC voltage input responses, and observing the dynamic motion of the membrane during actuation, including the total deflection, via optical coherence tomography (OCT) techniques [17,18].

Materials and methods
The design of the µPEM devices consists of a square shaped membrane composite containing an actuation electrode sandwiched between parylene C layers. The membrane composite is fabricated over a micromachined actuation chamber with a ground electrode patterned at the bottom of this chamber. Microfluidic channel architectures are also monolithically integrated on top of the electroactive membrane as a microshell geometry. For this work, the dimension of the square membrane is 1400 µm in length per side. The thickness of the membrane composite (parylene C and gold laminate membrane thickness) and the actuation chamber depth can be varied as needed by changing the deposited film thicknesses and chamber etching depths, respectively. The microsystem design also contains venting holes at the edges of the membrane. These venting holes have two main purposes; the first is to provide solvent access to strip away a sacrificial layer underneath the membrane while performing the final release steps within the microfabrication process. The second purpose is to provide a path for releasing air under the structure during actuation, decreasing the degree of squeeze-film damping [19]. The larger this damping effect, the higher the actuation voltage needed to counter the larger resistance for the membrane to deflect, diminishing the deflection of the microsystem under actuation [20]. Figure 1 shows the top and side view of a µPEM device schematic design and picture of a fabricated device.

Fabrication of µPEM devices
The fabrication of µPEM devices consists of several micromachining steps and a six-mask set microfabrication process. Masks are designed using AutoCAD modeling software (Autodesk, Inc., Version 2022, Mill Valley, CA, USA) and printed as transparency masks for photolithogrpahy processing (CAD/Art Services, Inc., Bandon, OR, USA). A crosssection flow chart of the fabrication process is shown in figure 2.
A silicon dioxide coated silicon wafer (University Wafer, Boston, MA, USA) with a 500 nm thermal oxide film was spin coated with a high resolution positive photoresist (Shipley S1818, Kayaku Advanced Materials, Westborough, MA, USA) on both sides of the wafer as a way to protect the backside from further wet etching steps. The photoresist was spun onto the wafer at 4000 revolutions per minute (RPM) to a thickness of 2.6 µm via spin coating (Headway Research Inc., Garland, TX, USA) and baked on a hot plate for 1 min and 20 s at 115 • C. Using the first photomask, opening windows for the actuation chamber area were patterned using  [21]. Wafers were submerged in etchant for 10 min, rinsed several times in DI water and blow dried by a nitrogen gun. Photoresist from the front and back side of the wafer was removed by submerging in acetone, IPA and DI water baths, and then rinsed several times and blow dried by a nitrogen gun. The actuation chamber was defined using a XeF 2 bulk micromachining technique [22,23] using a Xetch Silicon Etching System (XACTIC Inc., Pittsburgh, PA, USA). Using an etching rate of 0.2 µm per cycle (2.5 Torr XeF 2 and 3 Torr N 2 with a 60 s etching cycle) actuation chamber depths of 7 µm and 10 µm were achieved after either 35 or 50 cycles, respectively. Examples of the characterization and different chambers depths can be seen in the SI figure S-1. Thereafter, wafers were prepared for Parylene C thin film deposition by submerging in a silane solution, A-174 (Specialty Coating Systems, Indianapolis, IN, USA), which promotes the adhesion of Parylene to silicon substrates. Wafers were loaded into a chemical vapor deposition coater (Labcoter ® 2 PDS 2010, Specialty Coating Systems, Indianapolis, IN, USA) to deposit a conformal coating of 2 µm thick Parylene C film. The second mask was used, to define the ground electrode area at the bottom of the actuation chamber. A thick positive photoresist AZ9260 (IMM, Argyle, TX, USA) was spun onto the wafer, to a thickness of 5 µm via spin coating (Headway Research Inc., Garland, TX, USA) and baked on a hot plate for 3 min at 110 • C. After the photolithographic exposure (1400 mJ cm −2 ) and developing (AZ MIF 300 Developer), devices were descummed to further clean any residual debris using an O 2 plasma generator/asher (Nordson March PX-250 Plasma Etch Asher System, Carlsbad, CA, USA) for 60 s at 100 Watts power. A chrome/gold (Cr/Au) metal thin film was deposited onto the wafers via sputtering (PVD75, Kurt J. Lesker, Glassport, PA, USA). A thin layer of Cr (∼30 nm) was first deposited as an adhesion layer, followed, without breaking vacuum, by a thicker (200 nm) layer of Au. The bottom electrode was defined via a lift-off process by dissolving the masking PR in acetone [24]. A second layer of parylene C with a thickness of 1 µm was deposited to insulate the ground electrode. The actuation chamber was planarized using the third mask and a thick photoresist AZ9260 (IMM, Argyle, TX, USA) deposited to the thickness of the actuation chamber depth (7-10 µm) via spin coating at the appropriate RPM recipe, avoiding either under filling or over filling as seen in SI figure S-2. If the photoresist was slightly thicker than expected, a reflow approach was performed, baking the wafer on a hot plate for 4 min at 130 • C, reflowing the photoresist to decrease the thickness. An example of this reflow approach is seen in SI figure S-3. A third layer of parylene C with a thickness of 4 µm was deposited to enclose the actuation chamber. A fourth mask was used to pattern the top actuation electrode on the membrane. Again, a Cr/Au film was sputtered onto the wafer (PVD75, Kurt J. Lesker, Glassport, PA, USA) and lift off was used to define the top electrode. Another film of parylene C with a 3 µm thickness was deposited on the top electrode in order to sandwich the electrode between the parylene C films, forming a laminated membrane composite. A fifth mask was used to define the microfluidic channel patterns on top of the electroactive membrane actuator, with the definition of the microfluidic channels achieved using a thick photoresist (AZ9260, IMM, Argyle, TX, USA). This photoresist was spun onto the wafer twice (double-coating) to achieve a total channel height of 20 µm. After definition of the microfluidic channels, the final layer of parylene C was deposited (5 µm) to create the microshell microchannels while also, defining the final thickness of the actuation membrane. Finally, a sixth mask was used to pattern another layer of thick photoresist (AZ9260, IMM, Argyle, TX, USA) as a protective etching mask [25] for the final etching of the parylene C thin films which define openings for the microfluidic ports/reservoirs, electrical contact pads, and membrane venting holes. This mask was patterned with a PR thickness double the overall thickness of the parylene membrane to be etched. The etching selectivity between parylene C and photoresist is approximately 1:1, so having an etching mask twice the total parylene thickness ensures that the underlying features are protected during the etch. The etching process was carried out via O 2 plasma etching using a Reactive Ion Etcher (Mini-Lab Plasma Pod System, Plasma-Therm LLC., St. Petersburg, FL, USA) at a power of 50 Watts and 80 mTorr pressure, with multiple etching cycles of 10 min to avoid overheating of the wafer.

µPEM dicing, structure release, and packaging
Following fabrication, each of the microsystem devices was diced individually into dies using a precision scriber tool (PELCO FlexScribe, Ted Pella, Redding, CA, USA). Diced devices were released in acetone and IPA baths to remove the sacrificial photoresist from the actuation chamber and microfluidic channels. After the photoresist was completely stripped off from the µPEMs, devices were allowed to air-dry overnight and then placed in a convection oven at 65 • C for 30 min to complete the drying process. µPEMs with and without embedded microfluidic channels were glued with epoxy onto predesigned PCBs (Advanced Circuits, Aurora, CO, USA) using EPO-TEK 301 adhesive (Epoxy Technology, Billerica, MA, USA). Epoxy was carefully spread onto the back of the µPEMs which were then placed on top of the PCB. Devices were left overnight inside a convection oven at 65 • C to bond. Wiring between the devices and PCBs was achieved via soldering of thin wires between the electrical pads and the pins.

Device characterization for static actuation
To determine experimental pull-in voltages of the actuation membranes, as well as repeatability and reliability, a probe station set up was utilized (Micromanipulator Co., Carson City, NV, USA). Packaged devices were placed on the probe station platen and secured via a vacuum chuck. Devices were excited using a 125 V DC power supply (Circuit Specialists, Tempe, AZ, USA) and membrane deflection was observed via a video camera (BFLY-U3-23S6M-C, FLIR Systems Inc., Billerica, MA, USA) mounted on the probe station positioned orthogonal to the device under test.

Device characterization for dynamic actuation
For dynamic motion measurements of the µPEMs, a function generator (Siglent, Shenzhen, China), coupled to a 50× amplifier (TEGAM, Geneva, OH, USA) and digital oscilloscope (TBS 1032B, Tektronix Inc., Beaverton, OR, USA) was utilized, to input and monitor the signals. In addition, a phase-sensitive OCT setup was used. This custom-built spectral-domain OCT system has a center wavelength (λ) of 1325 nm, bandwidth of 100 nm, and an A-line (sampling) rate of 28.3 kHz. OCT has been commonly used to provide highresolution cross-sectional images of biological tissues, such as skin [26] or retina [27]. OCT can also measure dynamic motion and displacement of moving objects in a non-contact manner, making it a very useful technique to measure the small, dynamic motion of µPEMs under AC excitation. The µPEMs were situated on a movable stage and the OCT sample arm beam was focused onto the device surface as seen in figure 3.

Modeling of the µPEMs using analytical plate theory
The physical principles used to model actuation of the µPEM devices are the same as other electrostatically driven MEMS actuators. The actuators are formed by defining two electrodes separated by a defined gap (the distance between the membrane and bottom of the actuation chamber) or spacing, with one fixed electrode (commonly the ground) and the other located on a flexible structure (positive or driven electrode). The latter is free to move due to charge accumulation on the electrode and electrostatic force between the two charged electrodes during electrical bias, equivalent to the force between electrodes in parallel plates capacitors and capacitive microbridges [28]. Due to the electrostatic force between the two electrodes the gap between the plates will decrease as the movable electrode is pulled towards the fixed electrode. As the movable electrode displaces, there is also a mechanical restoring force that opposes the deflection according to Hooke's law [29], where the balance between the electrical and mechanical forces determines the equilibrium plate deflection during actuation or stroke recall when the bias is removed. A force balance between the electrostatic force and mechanical restoring force is shown in equation (1): where the electrostatic force is dependent on the surface area, A = W 2 , of the electrodes, the applied voltage, V, the initial gap distance, d, displacement of the movable plate, x, permittivity of the space between the electrodes, ε, and the restoring mechanical force which is dependent on the flexible rigidity (spring constant), k, of the plate. There is also a critical voltage beyond which a stable equilibrium displacement is no longer sustainable in the system, and the electrostatic force overwhelms the restoring spring force, resulting in collapse (or pull in) of the movable electrode entirely towards the ground electrode. This critical voltage is known as the pull-in voltage (V pi ) [30] and occurs when the plate deflection exceeds one third of the gap distance (x > 1 3 d). The pull-in voltage is a useful parameter since it defines the operational voltage range from different actuator designs which will provide a stable membrane deflection. To calculate the stiffness of µPEMs, classical Kirchhoff-Love plate theory was used to model plate deflection and determine its flexural rigidity. Here, it was assumed that the maximum deflection of the plate occurs at the geometric center of the µPEMs. The µPEMs are treated as a composite laminate, in this case a thin gold electrode metal film embedded within two layers of parylene C films. For analytical modeling purposes, the plate deflection was assumed to be small so that the electrostatic force was assumed to be a uniform load per unit area, where the maximum deflection for a square shaped membrane is: Here q is the uniform load (electrostatic force) per unit area, W is the side length of the square plate, E is the Young's modulus, α is an empirically derived constant from the flexural rigidity, and t represents the total thickness of the plate composite (structural parylene film with embedded gold electrode). The Young's modulus of the parylene-metal composite was estimated using a modulus-based weighting of each composite material fraction as: where E p and ϕ p represents the Young's modulus and volume fraction, respectively, of parylene (P), E M and ϕ M are the Young's modulus and volume fraction, respectively, of the gold metal (M) electrode. This effective Young's modulus was substituted into equation (2). Under the assumptions that plate stiffness follows Hooke's Law and is equal to the applied electrostatic force, F e , divided by the maximum deflection (k = Fe y max ) with the applied force (F e = qW 2 ) equal to the uniform electrostatic load, q, multiplied by the membrane area, W 2 , then the stiffness (K s ) of a square shaped µPEM is given by equation (4): Consequently, we can now express equation (1) in terms of the material properties and membrane geometry for the square plate (0 = − εW 2 V 2 2(d−x) 2 + E eff t 3 αW 2 x) and it is possible to calculate the pull-in voltage for a square geometry by, substituting x in equation (1) with 1 3 d and solving for the pull-in voltage: Using equation (5) it is possible to calculate the µPEM pull-in voltage as a function of shape/area (A = W 2 ), membrane thickness (t), membrane size (W), and gap spacing (d). The solution of this analytical model was used to calculate the theoretical membrane pull-in voltage as a function of membrane thickness (t), width (W), and gap distance (d) (7 µm and 10 µm) for square designs as shown in figure 4. A similar model for circular shaped membrane designs was determined and analytical modeling results were obtained and can be observed in SI figure S-4. Examples of the dimensions of various square membrane designs and calculated pull-in voltages are shown in table 1. To further strengthen the modeling approximations of the µPEMS, Finite Element Analysis modeling using COMSOL Multiphysics was explored and compared as seen in SI figures S-5-S-7.

Fabrication of the µPEMs
The µPEM devices with monolithically integrated microfluidic channels were fabricated using the process described in section 2.1 and shown in figure 5. The microfabrication of µPEM devices on a 4 inch silicon wafer yielded between

Device packaging
An example of a packaged device is shown in figure 6. Once the µPEM devices were fabricated and packaged, they were tested to assess the membrane actuation beha- vior from static and dynamic driving voltages. Membrane deflection behavior is instrumental in developing microfluidic applications for µPEM devices, such as coupling between membrane movement and fluid/particles/cells within the microchannels for our proposed future applications in micromixing, particle manipulation, or applying strain to adherent cells within the microchannels.

µPEMS static actuation
Actuation of µPEMs with and without embedded microfluidic channels was tested with a DC power supply in order to study their static deflection behavior, pull-in voltages, and sequential/continuous actuation. DC voltages ranging from 0 to 125 V were applied to the devices while varying the voltage in 1 volt increments at a time to visualize the deflection of the membrane. When the applied voltage approached the pullin voltage of the structure, the membrane began to deflect downward, and subsequently when the applied voltage was greater than the pull-in voltage, the membrane completely collapsed onto the bottom of the actuation chamber. The deflection following pull-in behavior could be visually detected by the appearance of interference patterns through the transparent portions of the parylene membrane [31,32], which indicate areas where the membrane transitions from free-standing to being pulled down to the bottom of the actuation cavity as seen in figure 7.
In order to compare experimental results to theoretical predictions, devices both with and without embedded microfluidic channels containing varied membrane thicknesses and actuation chamber depths were tested to determine their pullin voltages which were then compared with theory. Tables 2  and 3 show the effect of varying membrane thicknesses on the theoretical and experimental pull-in voltages for 10 µm and 7 µm actuation chamber depths, respectively. In addition, a reliability test was performed on the microsystems to assess actuation hysteresis over 50, 1000, and 10 000 actuation  cycles. The 50 actuation cycle test was performed manually for all devices, observing the deflection and interference pattern under microscopy. For the 1000 and 10 000 cycle tests, a function generator was programmed to generate a square wave every 20 s with similar voltage amplitudes to the previously determined experimental pull-in voltages. The actuation behavior of the microsytems was found to be continuous and repeatable between cycles. Figure 8 shows a comparison of the experimental and theoretical pull-in voltages from the tested µPEM devices as a function of membrane thickness and actuation chamber depth, as well as the variance of pull-in voltages during continuous actuation of the µPEMs under a reliability test for a 50 actuation cycle test. It can be seen that there is a good agreement between the experimental and theoretical pull-in voltages for µPEMs both with and without embedded microfluidic channels. Some differences between theoretical and experimental pull-in voltages were detected for the thicker membranes for the µPEMs without embedded microfluidic channels (table 2 and figure 8 top) which could be due to slight variations in the actual membrane thickness due to variations in parylene deposition rate (since K s ∝ t 3 , a small variation in thickness can have a large influence on stiffness), the stiffness of the membrane may no longer act as a Hookean material at larger deflections, or the assumption that there is an uniform electric field/electrical force per unit area is no longer valid with larger deflections and resulting curvature of the deflecting membrane.

Dynamic motion of the µPEMs
For dynamic actuation of the µPEMs, certain parameters of the input signal needed to be considered, such as the driving voltage peak to peak amplitude (V pp ) and the DC bias (offset) of the driving signal. In order to provide dynamic actuation a function generator was connected to a 50× amplifier with the amplified driving signal connected directly to the µPEM devices. For example, if a peak-to-peak voltage amplitude (V pp ) of 1 V and a 1 V DC bias was output from the function generator, the overall maximum actuation voltage will be 100 V (50 V pp + 50 V DC bias). For this experiment, a µPEM device with embedded microfluidic architecture was tested. The device had a total membrane thickness of 15 µm, and a 7 µm actuation chamber depth. Deflection of the µPEMs was experimentally determined using the OCT system. First, a sine wave with 1.5 V pp and 800 mV DC bias with a 50× amplification for a total of 75 V pp and 40 V DC bias applied to the µPEM was tested with increasing driving frequencies. The membrane deflection measured showed a greater deflection at 10 Hz (deflection of 6 µm) than at 100 Hz (maximum deflection of 2.5 µm) as seen in figure 9(a). At the lower actuation frequency of 10 Hz, we expected the deflection behavior to more closely match that of the static deflection presented in section 3.3. When deflection measurements at higher driving frequencies (200, 450 and 500 Hz) were obtained, the maximum deflection of the membrane decreases (figure 9(b)) due to the unsteady response to the rapidly varying driving signal. Additionally, a driving sine wave of 75 V pp at a frequency of 200 Hz, with two different offsets, 25 V DC and 40 V DC bias was applied in order to determine the influence of the DC bias on the membrane deflection. As seen in figure 9(c) the maximum membrane deflection increases by approximately two when the higher DC offset value was used.
The increase in deflection amplitude is related to the application of stress or voltage applied to a thin membrane. As the DC bias is applied to the µPEM it increases its taughtness  constant electrostatic force on the membrane, with displacement from its resting position increases, enabling a larger deflection amplitude using the same AC drive voltage. Finally, the sine wave V pp bias was increased in order to assess the influence of the amplitude of the driving sine wave on membrane deflection. Voltage amplitudes of 150V pp and 300 V pp at a 600 Hz driving frequency and 40 V DC bias were tested. As seen in figure 9(d), the measured deflection at 300 V pp was approximately zero in comparison to 2.2 µm from the 150 V pp sine wave. The deflection of the membrane at 300 V pp bias was measured at zero deflection because the bias was much greater than the pull-in voltage so the membrane collapsed to the bottom of the actuation chamber and the frequency was higher than the mechanical response timescale so the membrane could not relax back to its undeflected position. It is important to consider, that decreasing the driving frequency will allow the membrane to respond to higher V pp and higher V DC values, avoiding the zero-deflection pull-in behavior. Figure 9(e) compares two driving signals with different frequencies, both of them with the same 40 V DC applied DC bias but one with a 300 V pp AC bias at a frequency of 200 Hz, and the other at 150 V pp at 600 Hz. The maximum deflection for the 200 and 600 Hz signals are 2.3 µm and 2.3 µm, respectively. These results demonstrated the actuation capabilities of the µPEMs at varying driving inputs.

Validation of the integration of microfluidic channels from µPEMs
In order to ensure the microfluidic channels were patent and to validate the integration of the microfluidic architecture on top of the membrane actuators, channels were perfused with 8.91 µm diameter microbeads (Spherotech Inc., Lake Forest, IL, USA). The microbeads were pipetted into the microfluidic reservoir and allowed to fill the microchannels via capillary action. Following perfusion, no leakage of fluid around the microchannel walls was observed. Figure 10 shows images of the particles within the microfluidic structure acquired with the probe station camera.

Discussion
SOC devices represent an opportunity to integrate multiple fluid handling and actuation capabilities within a small footprint in a single chip design [1]. As an alternative to pneumatically driven PDMS based devices, the novel polymer-based MEMS platform described herein, describes the fabrication, modeling, and first experimental characterization of a Parylene-based electrostatic actuator defined over a bulk micromachined actuation chamber with integrated microfluidic channels. Actuation results show that the theoretical and experimental pull-in voltages for the membrane actuators closely match each other. Differences between theoretical and experimental results are likely due to fabrication factors that affect device mechanics, such as variation in the actuation chamber depth and total thickness of the actuation membrane due to variations in deposition parameters, as well as large deflections invalidating the uniform electric force assumption used in the theoretical model (SI figures S-6 and S-7). Dynamic actuation with OCT measurements demonstrated that the system response depends on the driving signal parameters. For example, larger deflections of the µPEM can be achieved when it is driven at lower frequencies or smaller deflections when driven at higher frequencies. In addition, increasing the DC bias allows greater deflection of the electroactive membrane. The integrated microfluidic channels within the µPEM devices were perfused with fluid and microbeads, showing no leakage, and mechanical rigidity in the structure. Moreover, the system packaging capabilities make this an ideal microsystem with ease of use and handling, avoiding robust set up and off-chip connections. Future work will focus on exploiting the electroactive deflection behavior and coupling the membrane movement to drive fluid circulation within the microfluidic channels similar to using piezoelectric actuators [33]. This may lead to further microfluidic applications such as micromixing [34] or particle manipulation for cell sorting [35] as well as stimulating strain sensitive cells cultured within the microsystem.

Conclusions
This work presents a compact parylene-based MEMS platform with embedded microfluidic channels which has applications for LOC and SOC devices. We successfully developed a fabrication protocol and analytical modeling techniques for these devices. The analytical model informs the design process for different membrane geometries and configurations, and can be used to determine appropriate pull-in voltages for the µPEMs. Fabricated µPEMs were characterized and tested for their actuation performance using two different actuation conditions, static actuation and dynamic motion. Pull-in voltages, continuous actuation, and repeatability of devices were tested and experimentally demonstrated and analyzed. For all microsystem configurations (with and without microfluidic channels), the actuation and experimental pullin voltages remained close to the theoretical pull-in voltages, making it a precise and reliable platform. In addition, this fabrication approach can be translated into the development of other microsystems that rely on electrostatic actuation mechanisms or polymeric-suspended microstructures for applications beyond the biomedical field, such as pumps [36], pressure sensors [37] and resonant gas sensors [38].

Data availability statement
Additional details on fabrication and modeling of the µPEMs can be found in the supplementary information.
The data cannot be made publicly available upon publication because they are not available in a format that is sufficiently accessible or reusable by other researchers. The data that support the findings of this study are available upon reasonable request from the authors.