Highly integrable silicon micropumps using lateral electrostatic bending actuators

The trend towards microfluidics and the related development of microfluidic components and integrated systems started in the early 1990s and has been boosted by a growing demand for portable devices during the last years (Mohith et al. 2019). As active drive elements, pumps are indispensable for microfluidic systems. Micropumps are nowadays used across numerous and diverse applications. In the area of medical technology and life sciences, micropumps find their application in the synthesis of biopharmaceutical proteins and in protein engineering as well as in drug screening and lab-on-a-chip (LoC) systems (Abgrall and Gué 2007). In addition, micropumps are a key component in point-of-care (PoC) diagnostics and drug delivery systems. Compared to conventional pumps, micropumps have a significantly reduced installation space. At the same time, new drive solutions are required to achieve the required pumping performance while keeping small sizes and low energy consumption.

In many application areas such as cooling (Singhal et al. 2004; Kandlikar 2014) miniaturized chemical analysis (Manz et al. 1990; Schoot et al. 1992), gas chromatography (Terry et al. 1979; Agah et al. 2006), and mobile on-chip applications (Lötters et al. 2013; Richter 2017) there is a need for micro-scale pumping featuring a low power consumption as well as a reduced size in addition to a certain flow rate and pressure difference between inlet and outlet. Piezomembrane-based and piezoelectric pumps can be considered as state-of the art. They can create high pump pressures and thus generate large forces. Furthermore, pump frequencies of several hundreds of Hz are possible. However, the deflections and thus flow rates are usually small and high operating voltages (typ. several 100 V) are required, entailing a barrier especially for portable applications. In addition, the active element usually consists of a piezoelectric alloy such as lead–zirconate–titanate (PZT), which is toxic due to its lead content and does not comply with the “Restriction of Hazardous Substances in electrical and electronic Equipment” (RoHS) directives. It is worth to mention that various lead-free materials have been developed but do not fully match the excellent physical properties of PZT materials.

An alternative to piezo-based systems are electrostatically driven silicon micropumps (Kim et al. 2015; Machauf et al. 2005; Zengerle et al. 1995). Their advantages are: scalability of the drive mechanism, low power consumption as well as fast reaction times due to the short reaction paths, and low volumes. Those make electrostatically driven silicon micropumps attractive for manifold applications. Most of the systems are based on a membrane attached to the outer edge, which oscillates mechanically out-of-plane (Richter et al. 2006; Han et al. 2012). The RoHS compatibility of a silicon micropump in combination with high achievable deflections has a great potential to replace PZT-based micro-membrane pumps. In membrane-based micropumps, the challenge is to design the membrane and the actuator precisely in order to achieve a stroke volume, i.e. compression ratio, sufficient for the pumping effect and to ensure a reliable operation (Richter et al. 1998; Laser and Santiago 2004). The small deflections of microactuators make this task even more difficult. For common electrostatic actuation mechanisms, this is challenging because of the dependence of the maximum deflection on the electrostatic gap, especially because of the limiting "pull-in" effect (Zhang et al. 2014). The concept and geometry of the electrostatic bending actuators developed by Fraunhofer IPMS bypasses this pull-in limitation (Conrad et al. 2015). Pull-in related effects such as adhesion, electrical discharge, and dielectric charge have been identified as the main failure mechanisms for a large variety of electrostatic micro devices (Zhang et al. 2014). The deflection of in-plane moving electrostatic bending actuators exceeds the natural pull-in limit by far (Conrad et al. 2015; Schenk et al. 2017). The benefit of the functional principle is that very small electrode distances in the micron range can be used and the resulting extremely high electrostatic forces can be fully utilized. Thus, the principle can be considered as a serious competitor to the existing thermomechanical, electromagnetic, piezomagnetic, and magnetostrictive drive principles, as well as to classical electrostatic microactuators.

The driving principle of the electrostatic micropump relies on the so-called Nanoscopic Electrostatic Drive (NED) developed at Fraunhofer IPMS (Conrad et al. 2015). The basics of the NED-based pump was described elsewhere (Borcia et al. 2018; Uhlig et al. 2018) and shall thus not be explained here into detail. Briefly, NED actuators consist of an array of electrodes stacked in-plane with an electrostatic gap in the range of several microns or even less. Each electrode pair is considered as a beam with the option to be bent in a double S-shaped bending line in a clamped–clamped arrangement. Electrically isolating spacers between an electrode pair are made of alumina. Upon applying an electrical voltage to the electrodes of the NED, the beams are deflected. The NED-actuators with electrodes are fully immersed in the pumping medium, with the permittivity of the medium acting as a scalar enhancer of the electrostatic force between the electrodes. Thus, to achieve the same electrostatic (abbrev.: es) forces when changing the pumping medium, the voltage needs to be adapted accordingly (Gaudet et al. 2017).

The beams are arranged in a way that adjacent beams are deflected oppositely. This in turn creates a chamber for the fluid. This scenario is depicted in Fig. 1(a), where the NED-actuators are visualized as beams. Reciprocating movement of the actuators under alternating driving voltage creates a two-step pump cycle with supply and pump phase, resulting in a rectified flow of a pumping medium thanks to the passive flap valves.

Since the movement of the actuators takes place in the plane of wafer, a clearance of 1 µm height is present above and below the actuators as well as the valves, depicted in Fig. 1(b). The height of the clearance is currently determined by silicon-on-insulator wafer used for device processing. During device operation a parasitic flow of the pumping medium takes place through these clearances. This flow has the effect, that part of the stroke volume \({V}_{a}\) of the actuators is transferred in and out of the chamber resulting in an increased dead volume of the pump. Optimizing the ratio of hydraulic resistances of the flow path in the chamber to the ones of the clearances will perspectively reduce the amount of parasitic flow and thus the dead volume associated with it.

The entity shown in Fig. 1(a) can be understood as a base unit. The arrangement of base units across the chip gives rise to a modularity and thus more complex designs. Each base unit is made up of one pump chamber, which is comprised of an even number of actuator stacks (containing a certain number of parallelly stacked NED-actuator beams named \({N}_{par}\)) as well as two rectifying valves. In the design presented here a single base unit features a chip volume of (4 × 1 × 1) mm³ (L × W × H). Arranging the base units in series or in parallel across the chip results in pumps with different specifications as already described in Uhlig et al. (2018). The small construction volumes in combination with low operating voltages are of high relevance especially regarding portable applications. In contrast to piezoelectric materials with ferroelectric properties (under purely elastic loading of all materials involved), electrostatic bending actuators do not exhibit a hysteresis in their deflection characteristic curve. Hence, defined deflections with a high reproduction rate are possible.

With the working principle of a mechanical reciprocating micropump, a known model from literature described in Zengerle and Richter (1994) can be used to estimate the performance of the NED-micropumps. The model divides the pump in its functional elements such as actuator and valves. The behavior of the functional elements is described by characteristic curves, which in rare cases can be estimated analytically or for more complex cases are determined via numerical simulation. The governing differential equation of the model is given in Eq. (1), which describes the temporal change of the pressure \(p\) generated in the pumping chamber.

The model is based on the underlying assumption that inertia effect is neglected, which means, that the input can be transient functions, but the response is steady- or pseudo steady state. Furthermore, it is assumed that the pressure distribution is spatially uniform across the pump chamber and the pumping medium is incompressible. For more insight into the model the reader is forwarded to Tay (2002), Nguyen (2004) and Zengerle and Richter (1994).

In Eq. (1), \({Q}_{iv}, {Q}_{ov}\) are the characteristic parameters of the flap valves for inlet and outlet-valve, respectively, and \({p}_{c}\) is the counter pressure at the outlet valve. For convenience, the pressure at the inlet valve is set to zero, thus the pressure difference across the pump is described by \({p}_{c}\). The term \(\left({\left.\frac{\partial {V}_{a}}{\partial U}\right|}_{p}\frac{\partial U}{\partial t}\right)\) describes the change of the characteristic volume displacement curve of the NED-actuator unit under alternating supply voltage, cf. Supplementary Figure S1(b). The terms \(\left(\frac{\partial {V}_{a}}{\partial p},\frac{d{V}_{iv}}{dp},\frac{d{V}_{ov}}{dp}\right)\) present the pressure dependent volume displacements of the elastic components of the NED-actuator as well as the inlet- and outlet-valve, respectively. The parameters \({N}_{hub}\) and \({N}_{par}\), are introduced to Eq. (1) to account for the modularity of the NED-micropump. The number of actuator beams in a pump base unit is described by \({N}_{hub}\), which here is the same for all micropump sub chips with \({N}_{hub}=4\). Subsequently, \({N}_{par}\) describes the number of parallelly stacked actuators in one beam stack, cf. Fig. 3.

The characteristic curves, cf. Supplementary Figure S1, of the NED-actuators and flap valves are determined via finite element method (FEM) including 2-way fluid structure–interaction (FSI), respectively, using ANSYS Workbench R19.2. For the actuator, a Multiphysics electrostatic-structural simulation approach is used to determine the stroke volume–pressure generation curve of the NED-actuator. The characteristic flow curves of the flap valves are determined via FSI using the transient structural and fluent packages combined with system coupling toolbox in ANSYS. The simulation method is already explained in detail in Uhlig et al. (2018). For the hydraulic elasticities of the flap valves constant values \(\frac{d{V}_{iv}}{dp}=\frac{d{V}_{ov}}{dp}=0.02\, \text{nl/}{\text{kPa}}\text{,}\) were used which are mean values of the volume displacements under pressure change arising from the valve simulations. The characteristic curves used as input for the model are displayed in the supplementary part. For the supply voltage \(U\), a modified square wave signal is used given in Eq. (2).where \(f\) is the excitation frequency and \(\delta\) a parameter responsible for the rounding of the square edges of the square wave, which in turn protects the actuator beams from mechanical defects due to sudden, harsh movement. The parameter \(\delta =0.07\) to match the function used in the experimental characterization.

Equation (1) yields the transient change of chamber pressure in the pump. Together with the characteristic flow curve of the outlet valve \({Q}_{ov}\left(p-{p}_{c}\right)\), the transient flow through the valve can be determined. Integration of the transient flow subsequently yields the transferred volume \(\Delta V\). Eventually, the mean flow rate can be calculated by dividing transferred volume \(\Delta V\) by the corresponding period \(\Delta t\). For small excitation frequencies \({Q}_{P}\) equals the transferred volume times the excitation frequency.

The parasitic “leakage” flow \({Q}_{L}\) through the clearances above and underneath the actuators as well as the valves is considered in Eq. (1) by subtracting \({Q}_{i}-{Q}_{L}\) the leakage flow from the flow curves of the corresponding cycle phase, expressed by Eq. (4)

Thereby the factor \(\left(\frac{{V}_{L}}{{V}_{a}}\right)\) is the ratio of the leakage volume associated with leakage flow to the total stroke volume \({V}_{a}\) of the chamber. The factor depends mainly on the geometry of the chamber (flow path to clearance path) and can be estimated analytically. For the here presented cases of the NED micropumps the length and width of the flow path in the chamber is constant through all the designs. Only the number of stacked actuator beams changes, which increases the length of the clearance paths. Thus, for \({N}_{act}=2, 4, 8, 16\) NED-actuators stacked in parallel \(\left(\frac{{V}_{L}}{{V}_{a}}\right)=0.61, 0.44, 0.28, 0.16\), respectively. One can see that volume associated with the leakage flow through the clearances decreases, and thus the additional dead volume, by increasing the hydraulic resistance of the clearances, which in this case is achieved by stacking more actuators in parallel. The term \({Q}_{iv}\left(-p\right)\) depicts the valve leakage in closing direction, cf. Fig. 10(d). The comparison of the model prediction and experimental data is presented in Sect. 6.

The micropump consists of two main components: device- and cover wafers. The device wafer contains the NED-actuator, whereas the cover wafer is bonded on top to obtain a closed micropump chamber. Well-known materials such as crystalline silicon, silicon dioxide, alumina and parylene-C are used to fabricate the micropumps. Each of these materials are RoHS compliant. The basic fabrication processes comprise lithography, thin-film layer deposition and etching. Etching is a fundamental technology during micropump fabrication since the pump structures are created by bulk micromachining. For the creation of high aspect ratio trenches, deep reactive ion etching (DRIE) is used.

Material of choice for device wafers 200 mm in diameter, (100) oriented, bonded silicon-on-insulator (BSOI) wafers. In a BSOI-Wafer a layer of monocrystalline silicon, termed “device layer” is bonded to a substrate wafer referred to as “handle layer”. The bond interface between device and handle is thermally grown silicon dioxide SiO₂. The sandwiched oxide layer is commonly called buried oxide (BOX) and serves in MEMS technologies primarily as an etch stop layer. The commercial BSOI wafers used throughout this work had thicknesses of 75 µm, 1 µm, and 650 µm of device, BOX, and handle layer, respectively. For the cover wafers a double sided polished, 400 µm thick, (100) oriented, 200 mm in diameter Si wafers are used.

Processing of the device wafer starts with the fabrication of insulating spacer islands of the NED-actuators, shown in Fig. 2 (1a and 1b). A photoresist layer is deposited serving as etching mask for the subsequent DRIE process. The isolation spacer islands are 2.4 µm wide, more than 15 µm long and 75 µm deep trenches filled by Al₂O₃ using atomic layer deposition (ALD) (Fig. 1(b)). The alumina will serve mechanical spacer between the electrodes of the NED-actuator. Since ALD layers are deposited conformal in the trenches and across the wafer surface, the redundant Al₂O₃ of the wafer surface is removed by chemical–mechanical polishing (CMP) from the device and handle-layers. This process step is the reason why the process flow is setup to begin with the spacer island processing, since no delicate or moveable structures are present at this point, which could be damaged during CMP planarization.

In the following steps Fig. 2(1c), electrostatic gaps, fluidic channels, and passive flap valves located at the inlets and outlets are etched in the device layer. To do so, a hard mask made of PECVD-deposited silica is used. The device layer is patterned by DRIE with the BOX serving as etch stop layer. The holes for the electric contacts of the device are created in the handle layer in a separate step, also depicted in Fig. 2(1c). The ultimate process step of the device wafer is the removal of the silica mask and Box layer to release the NED-actuators and passive moveable elements, shown in Fig. 2(1d). This is accomplished by HF vapor etching, which does not attack silicon and alumina.

To create the 1 µm clearance above the actuators, silica is grown uniformly across the cover wafer and subsequently patterned. At the location of the fluidic contacts, the oxide is patterned as well, shown in Fig. 2(2a). Using DRIE, the fluidic contacts are created in the top cover wafer with a previously deposited layer of aluminum at the back side of the wafer, which serves as an etching stop and is subsequently removed, see Fig. 2(2c).

In the last step of the wafer processing, a 500 nm parylene-C layer is deposited on bond side of the cover wafer serving as an adhesive layer for the bonding process of the device and cover wafer. After bonding, the wafer is diced, which completes the fabrication process of the electrostatic micropump. A schematic cross-sectional view of the micropump is shown in Fig. 2, step (3), with the fluidic contacts on top and the electrical contacts on the bottom side of the chip.

Figure 3 shows the top view of a single pump test structure with one pump chamber (\({N}_{ser}=1\)) comprised of 4 NED actuator stacks each containing \({N}_{act}=4\) stacked actuator beams. Four micropump designs were fabricated in the cleanroom. Figure 4 shows a micrograph of the layout from the 4 chip designs. The four chip designs contain four types of the micropump: intended for gas and liquids as well as parallel and in series for both types of fluids, respectively.

This section describes the measurement setup and procedure for the characterization of single micropump chips. The main parameters detected were flow rate and counter pressure dependent on the applied drive frequency.

The main results comprise flow rate and counter pressure graphs dependent on the applied frequency at fixed driving voltage. Figure 7 shows some characteristic curves. In Fig. 7(a), the flow rate for various sub chips with the number of \({N}_{act}=2, 4, 8 \,{\text{or}}\, 16\) parallel stacked actuators at 130 V and variable frequency are depicted. The flow rate as well as the optimal frequency of the excitation signal increase with the number of actuators \({N}_{act}\). At lower frequencies, the mechanical movement of the actuators follow the electrical signal. At higher frequencies this is not possible due to the inertia of the mechanical system of actuators and valves. Thus, the flow rate rises to a peak value and subsequently decreases until at higher frequencies the flow vanishes. Since it is a complex combination of mechanic and electric effects, it is possible to see more than one peak.

Figure 7(b) gives the flow rate for different sub chips with \({N}_{act}=2, 4, 8 \,{\text{or}}\, 16\) parallel actuators at 130 V and variable counter pressure. The frequency is fixed to the value at which the maximum flow rate was observed. The flow rate decreases linearly with the counter pressure. More actuators allow for higher counter pressures, as the total pressure applied to the system is divided evenly over all actuators. In Fig. 7(c), the flow rate of the sub chip with \({N}_{act}=4\) actuators in parallel at different voltages and frequencies of the excitation signal are presented. The flow rate does not increase linearly with the voltage since the electrostatic force scales with the square voltage. Additionally, at lower voltage the stiffness of the actuators is the dominant force.

While Fig. 7(a–c) display the flow rate dependent on frequency, driving voltage, and counter pressure, respectively, (Fig. 7d) illustrates the detected fluidic power at 130 V driving voltage and resonance frequency determined from earlier measurements. The fluidic power is derived from the counter pressure and flow rate. It is an indicator for the efficiency of the pump with a maximum at half the maximum of the counter pressure (Schomburg 2011). For battery-driven mobile applications, where power efficiency is an important issue, the working pressure should be chosen at maximum fluidic power.

The measurement results of the micropumps with a serial arrangement of pump chambers are presented in Fig. 8. Four different arrangements were tested with \({N}_{ser}=1, 2, 4 \, or \, 6\) chambers in series. Figure 8(a) shows the flow rate curves at variables excitation frequencies with a maximum flow peaking at 1 or 2 kHz. A serial arrangement of chambers does not increase the flow rate since every chamber transfers the stroke volume only to the next chamber. However as stated above each base pumping unit, e.g., pumping chamber, works against a portion of the pressure difference over the pump. Therefore, with increasing number of chambers in series, higher counter pressures can be reached. The experimental proof of this principle is shown in Fig. 8(b), where roughly a doubling of the achieved counter pressure takes place when doubling the number of chambers in series. The chambers are identical with 4 NED actuator stacks generating the stroke volume, each containing \({N}_{par}=2\) stacked actuator beams in parallel, cf. Fig. 2. It is highlighted here, that each curve displayed is the mean of a minimum of 3 measured pumps from different chips.

For the sub chip with \({N}_{ser}=8\) only \(6\) chambers were active during the measurements, due to electrical shortages at the contact regions. However, the measurements still follow the trend of reaching higher counter pressures when increasing the number of chambers in series. The corresponding flow rate curve and maximum flow rate is slightly decrease due to extra hydraulic resistance that the \(2\) inactive chambers add to the system.

The micropump approach presented in this paper features a high modularity, scalability as well as design freedom since single systems (e.g., base pumping units) can be stacked in series or parallel. Thus, not only the dimensions of a pump unit can be adjusted to the respective application, but also the required power density. The results presented here feature a serial arrangement of pumping units as well as parallel stack of NED-actuators. Modular concepts intend to increase the reachable back pressure of the system. A parallel stack of base units is possible as well and would result in a linear increase of flow rate with the number of base units, since each base unit pumps the stroke volume \({V}_{a}\).

The comparison of the predictions of the fluidic model (Eq. (1)) with the measured micropump data is given in Fig. 9. For low flow rates up to about \(f=500\) Hz the model follows the measured data well. In this region a quasi-static behavior of the system can be assumed. Here the flow rate also increases linearly with the frequency. The decrease of flow rate after reaching the maximum is not captured by the model due to the neglected inertia. In tendency the model predicts higher flow rates compared to the measurements. This can mainly be accounted for by the model assumptions, where the peripheral fluidic system is not considered and a uniform pressure distribution in the pump chamber is assumed. Thus, additional pressure drops caused by the hydraulic resistance are not accounted for, which are decreasing the flow rate as seen in the experimental data. Especially for the pump with \({N}_{act}=16\) the deviation of the model to the experimental data is slightly higher compared to the other pump. Here the internal fluidic feed channels in the pump are longer, due to the space the actuator stacks take up, thus increasing the internal hydraulic resistance of the pump.

In general, a good agreement between simulation and experimental validation was found confirming the usefulness of the applied model (Eq. (1)). In addition, the functionality of the developed S-shaped design of the NED-actuators could be validated. During the experimental investigations it was shown that a controlled micropump behavior could be achieved and mainly adjusted by parameters like the applied driving-voltage and -frequency as well as the counter pressure. Typical driving frequencies are in the range of some kHz referring to the high level of miniaturization and the related mechanical stiffness of the NED-actuators. Comparing the characteristic curves resulting from the experimental investigations, the electromechanical actuation principle of the pump using the lateral NED bending actuators could be affirmed. Furthermore, a good agreement was found between predictions of the model and the real behavior of the pump.

A summary of the maximum observed flow rates and counter pressures is given in Fig. 10. It shows the values for different configurations with \({N}_{par}=2, 4, 8 \, or\, 16\) actuators in parallel or \({N}_{ser}=2, 4 \, or\, (6)\) chambers in series. As expected, the flow rate for the pumps with multiple chambers in series is roughly equal to the pump with one chamber but same number of parallel actuators. The flow rate of the serial configuration is comparable to the one of 2 actuators in parallel since there are also 2 parallel actuators per actuator pair in the serial configuration. The results give a good overview of the modularity that can be achieved with the new micropump concept in terms of targeting the counter pressure.

This paper presents a micropump based on a novel electrostatic driving principle which can be considered as one of the smallest oscillating displacement pumps published so far (Mohith et al. 2019). The proof-of-principle was successfully accomplished. A test set-up on a microfluidic bench was designed and implemented allowing for a complete fluidic characterization of the micropump chips. The results comprise flow rate investigations dependent on the number of actuators in parallel or in series. An increase of the number of parallel actuators increases the flow rate as well as the counter pressure, while the number of pump chambers in series enhances counter pressure. These results meet the simulations.

At this point in time, the main challenge is that the driving electrodes are not physically separated from the pumping liquid. This means, electrical breakdown prevents the micropump for use of conducting liquids including any water-based fluid. To cope with this challenge, some concepts have been developed, but not set-up so far. The pump thus might be useful for pumping silicone and predominantly for gas pumping. The first micropump chips were tested successfully a dielectric liquid called Novec^TM7100 (3M Novec7100 Data Sheet 2020), confirming the functional demonstration of the new approach. Future systems should avoid contact between electrodes and liquid to allow pumping of conducting liquids. This would open the door to a wide range of applications ranging from medical and biological sciences to fine chemicals production and microreaction engineering.

The first generation of micropump chips are making use of passive flap valves based on a cantilever check valve design. A next step regarding the valves would be the replacement of the passive NED-valves by active NED-valves, which have been developed in parallel to the pump. Active valves would allow to implement a closed-loop-control enabling to control pump frequency (and thus flow rate) and the counter pressure.

Furthermore, a reduction of the fluidic short-cuts through the gaps above and below the electrodes is assumed to lead to a significant enhancement of the pump capabilities along with an increase of the transported volume and the achievable counter pressure.