**Low-Power, Multimodal Laser Micromachining of Materials for Applications in sub-5** μ**m Shadow Masks and sub-10** μ**m Interdigitated Electrodes (IDEs) Fabrication**

#### **Cacie Hart 1,2 and Swaminathan Rajaraman 1,2,3,4,\***


Received: 16 January 2020; Accepted: 7 February 2020; Published: 8 February 2020

**Abstract:** Laser micromachining is a direct write microfabrication technology that has several advantages over traditional micro/nanofabrication techniques. In this paper, we present a comprehensive characterization of a QuikLaze 50ST2 multimodal laser micromachining tool by determining the ablation characteristics of six (6) different materials and demonstrating two applications. Both the thermodynamic theoretical and experimental ablation characteristics of stainless steel (SS) and aluminum are examined at 1064 nm, silicon and polydimethylsiloxane (PDMS) at 532 nm, and Kapton® and polyethylene terephthalate at 355 nm. We found that the experimental data aligned well with the theoretical analysis. Additionally, two applications of this multimodal laser micromachining technology are demonstrated: shadow masking down to approximately 1.5 μm feature sizes and interdigitated electrode (IDE) fabrication down to 7 μm electrode gap width.

**Keywords:** multimodal laser micromachining; ablation characteristics; shadow mask; interdigitated electrodes

#### **1. Introduction**

In recent years, laser technology has been examined as an exciting material processing method for both industrial and academic researchers [1–4]. The high quality of laser beams allows for improved micromachining precision for a variety of materials [5,6]. Additionally, the compact size, high efficiency, cost-effectiveness, direct machining, 3D fabrication, and ease of integration are appealing to academic researchers with limited benchtop area and lean budgets.

Laser micromachining, specifically, involves the ablation of materials where the features produced by the laser are in the micro-scale [6,7]. Laser micromachining techniques are currently employed in the automobile [8], medical [2,9], semiconductor [10,11], and solar cell industries [12]. Lasers used in micromachining are available in a wide range of wavelengths (ultraviolet to infrared), pulse durations (micro- to femtosecond), and repetition rates (single pulse to megahertz) [6,8]. Because of this flexibility, laser micromachining allows for the processing of a variety of materials.

Figure 1 compares laser micromachining to three commonly used direct-write microfabrication methods [13–19]. All these methods involve a mechanism for removing material directly from a substrate in a desired pattern using computer-controlled machinery. While other patterning methods, such as photolithography are often used, they are not direct-write techniques and involve several steps to pattern materials; thus, a comparison to these other methods is not provided in this figure.

Photolithography is, of course, required for one of the methods depicted (reactive ion etching or RIE) for performing feature definition followed by RIE to "engrave' the feature in the substrate beneath [15–17,19]. Since RIE requires photolithography, the process involves more steps when compared to the other methods depicted in Figure 1 [17]. Micromilling is a very technically simple process; however, this simplicity comes at the expense of frequent drill bit brakeage and the inability to produce features in the sub-100 μm range [14]. As with some RIE technologies, focused ion beam (FIB) milling can define nanoscale features; however, the write process can take days depending on the pattern complexity. Additionally, in FIB-based milling processes, the material surface may become amorphous due to the implantation of gallium ions [13,16–18]. The comparison depicted in Figure 1 is by no means comprehensive and has been included to show the characteristics of laser-based micromachining with some other commonly used technologies for patterning materials in micromachining and micro-electro-mechanical systems (MEMS). Several other techniques for direct subtractive fabrication of micromachined features are available such as electrical discharge machining (EDM) [20], ultrasonic drilling [21], water jet cutting [22] etc.

**Figure 1.** Comparison of three widely used direct-write fabrication methods (reactive ion etching (RIE) [15,17], micromilling [14], and focused ion beam (FIB) [13,16,18]) to laser micromachining. Each of these methods allows for high aspect ratio, serial, single-substrate fabrication. In the case of RIE multi-wafer/multi-substrate parallel processing is also possible as with some laser micromachining examples as well for higher throughput micro/nanofabrication.

Laser micromachining has occasionally been used for shadow mask fabrication; however, the state of the art in laser micromachining (minimum feature size of 10 μm) [23] (to date) has relied entirely on single-wavelength excimer, CO2, and Nd:YAG (neodymium-doped yttrium aluminum garnet) lasers [1,3,12,23–36]. To the best of our knowledge, multimodal laser micromachining has not been used for shadow mask fabrication. While the term "multimodal" in the laser field typically means that the laser contains multiple transverse electromagnetic modes, the laser micromachining tool used in this work has the ability to operate at several wavelengths in the same tool, a rare capability not afforded by many lasers. Specifically, with respect to shadow mask microfabrication with lasers, several impressive research efforts are reported in literature: Klank, et. al. fabricated 85 μm channels in polymethyl methacrylate (PMMA) with a CO2 laser [1]. Fan, et. al. also used a CO2 laser to micromachine 250 μm lines in wax for use as a shadow mask [3]. Tahir et. al. used a 1064 nm Nd:YAG laser to fabricate 250 μm channels in wood, glass, plastic, and rubber [12]. Chung, et. al. fabricated shadow masks with minimum feature sizes of 200 μm with a 785 nm Ti: sapphire laser [26]. Shiu, et. al. machined 140 μm channels in low-carbon steel for use as a shadow mask using an excimer laser [34].

The lasers used in the aforementioned research can produce shadow masks for MEMS applications, but they are unable to micromachine highly precise features on the scale of a single micron, unless they operate in the femtosecond regime [3,7,23,24,30,37]. The higher power of these lasers allows for the processing of thicker materials in a more reasonable time scale; however, this capability comes at the price of benchtop machining, high costs, space, high power usage (GW), and absence of multimodality to micromachine several materials with the same tool. Characterization of lasers used for materials processing is necessary for the proper use of such tools. By understanding the necessary energy, repetition rate/frequency, spot size, and depths that can be micromachined with various processing conditions in specific materials, users can subsequently optimize their experiments for successful results in different applications.

In this paper, we report the full characterization of a multimodal laser micromachining tool and two applications of the usage of microstructures fabricated using the tool. Multimodal laser micromachining allows for a wide range of materials processing capabilities; however, this comes at the price of limited power and nanosecond ablation. These tools allow users to operate at specific laser wavelengths (1064 nm, 532 nm, and 355 nm, in our case) and reduced powers (2.6 mJ maximum power). These conditions, however, provide greater control over the material selectivity for laser micromachining and ablation depths. Specifically, one can remove a material from a device while leaving the remainder of the constituent materials unaffected [38].

Typically, such laser micromachining tools are employed in liquid crystal display (LCD) repair, semiconductor failure analysis, and removing shorts and passivation layers in integrated circuits [6, 39]. In this paper, we characterize the multimodal tool for the laser micromachining of six (6) different materials: Stainless steel (SS) and aluminum, to be machined with infrared (IR) mode, polyethylene terephthalate (PET) and Kapton®, to be machined with ultraviolet (UV) mode, and polydimethylsiloxane (PDMS) and silicon, to be machined with green mode. Subsequent shadow mask patterning allows for the definition of organic and inorganic layers and the development of a fully functional interdigitated electrode (IDE) devices. These devices are further characterized in this paper.

#### **2. Microfabrication Method Overview**

In laser micromachining, the laser beam is collimated into a small spot and patterning is achieved by either moving the substrate within a fixed beam or rastering the laser across a surface [6]. The desired machining patterns simply need to be drawn in a CAD program (such as AutoCAD, Autodesk, San Lafayette, CA, USA) and imported as a drawing exchange format (DXF) file into the control program of the laser micromachining tool. Once the program is executed with the laser, substrate material removal can be a result of photochemical, photothermal, or photophysical ablation [40], as shown in Figure 2. Commonly used processes include laser cutting, scribing, drilling, or etching to produce relief structures or holes on a substrate in ambient temperatures [3,8,23,27,40–42]. The power of this technique lies in the ability to construct desired patterns on arbitrarily shaped surfaces, with the only limitation being the degrees of freedom and the resolution of the motion controller. Laser micromachining is considered a rapid prototyping technique because designs can be changed immediately without the need to fabricate new molds or masks.

**Figure 2.** Mechanisms of ablation in laser micromachining. Substrate removal can be a result of thermal ablation, physical ablation, chemical ablation, or a combination of these mechanisms.

Every laser micromachining system is comprised of three parts, as shown in Figure 3a for the multimodal laser micromachining tool: (1) the source laser, (2) the beam delivery system, and (3) the substrate mounting stage. Obviously, the laser source is at the heart of the system, as it determines which substrates and feature sizes can be micromachined [8]. The system used in this work has a multimodal laser source that allows for switching between three wavelengths of light: 1064 nm infrared (IR mode), 532 nm visible green (Visible mode), and 355 nm ultra-violet (UV mode). This wavelength switching allows for greater substrate compatibility and a wide array of feature sizes in an extremely compact benchtop system.

**Figure 3.** (**a**) External schematic of a QuikLaze 50ST2 Multimodal Laser Micromachining System. The three components of every laser micromachining system are shown: (1) the laser source box; (2) the beam delivery system; and (3) the motorized substrate mount. (**b**) Internal schematics of the laser source box. This shows the three crystals and a switch that create the different wavelengths of light the system produces with the ability to switch between three wavelengths.

Beam delivery involves optical components, including fixed focusing objects and mirrors, galvanometric scanners, optical fibers, wave-guides, apertures, and q-switches, that are used to generate the laser spot. The selection of these optical components depends on the working distance, desired spot size, and required energy [6,8,41,43]. The combination of the laser source and optical components of the beam delivery system determines the ultimate properties of the laser beam. The beam delivery system used in this work is depicted in Figure 3b with greater detail.

Lastly, the substrate mounting system depends on how the rastering occurs on the tool. If the laser beam is to be rastered over the substrate surface, then a stationary substrate mount may be used. However, in most cases, including the system used in this work, it is more desirable to raster the substrate itself within a laser beam. In this instance, the substrate mount is manipulatable in the x- and y-directions, and even in the z-direction in some cases.

In the multimodal laser system used in this work, the laser beam is initially generated as a 1064 nm IR mode, shown in Figure 3b. The IR laser beam is subsequently passed through two crystals in the beam delivery system that produce the two additional wavelengths. The beam delivery system employs filters prior to the microscope optics to filter out the unnecessary wavelengths, allowing only the desired wavelength to interact with the substrate surface.

#### **3. Theoretical Background**

A thermodynamic approach to calculating the theoretical "depth of cut" for the materials tested in this work was adapted from Schütz, et. al. [10]. This formula is based solely on material properties and laser energy. Examining the depth of cut in this manner allows for a more fundamental understanding of the interaction between the laser and the material with fewer assumptions when compared to a molecular approach used for photochemical ablation [10,42,44].

$$a\_p = \frac{\phi}{\rho \int\_{T\_{\text{row}}}^{T\_v} c\_p(T)dT + \sum\_{i=1}^n H\_{ph}^v} - d \tag{1}$$

Equation (1) shows the relationship between the depth of cut per pulse (*ap*), the laser fluence (Φ), and involved material properties, including density (ρ), vaporization temperature (*Tv*), heat capacity (*cp*), and the phase change enthalpy (*H<sup>v</sup> ph*). The d term (right-hand side of the equation) is a correction term that includes optical and thermal losses. This equation represented schematically (Figure 4) balances intrinsic energy and enthalpy, which are state variables and omits fundamental process parameters that are not state variables (i.e., particle dynamics) [45,46]. While the inclusion of these fundamental process parameters could lead to a more accurate theoretical solution, the complexity would be vastly increased due to the use of non-linear partial differential equations and the need for sophisticated simulation software to solve such equations [5,6,10,40,44,47]. Additionally, since this laser micromachining setup does not operate in the femtosecond regime, ablation occurs through melt expulsion and redeposition driven by the vapor pressure and the recoil pressure of light [44,47,48]. As a result, a simple thermodynamic analysis was performed to extract the relationship between the "depth of cut" and laser fluence which can be compared to the results obtained through experiments.

**Figure 4.** Schematic showing the variables in the thermodynamic theory for evaluation of the depth of cut.

#### **4. Materials and Methods**

#### *4.1. Multimodal Laser*

A QuikLaze 50ST2 multimodal laser (New Wave Research Inc., Fremont, CA, USA) was used for all the laser micromachining performed in this paper. A selection of three wavelengths as mentioned in the previous sections allows the laser to be tailored to a specific application. The microscope of the laser system is equipped with 10×, 50×, and 100× lenses, each with specific wavelength limitations. Because of additional filters in the microscope lenses, the green wavelength can be used through any of the lenses, while the UV and the IR can only be used through the 50× and 100× lenses, respectively.

The laser outputs a 5 mm diameter Gaussian beam, which is then shaped into a rectangle by the XY aperture. The size of this rectangle is determined by user inputs into the control software. The maximum pulse duration of the laser is 5 ns for all wavelengths; however, it can be adjusted by the user in the program. Additionally, the laser fluence depends on the user specifications in the control software and the wavelength of light used, as the laser output energy can be adjusted by the user. The fluence ranges from a maximum of 27,000 J/cm2 to a minimum of 1.08 J/cm2.

#### *4.2. Materials Used*

To ensure a comprehensive study of multimodal laser micromachining, several materials were machined. Success in laser micromachining for a given material is typically determined only by the choice of wavelength because each material reacts differently to a specific wavelength. In general, metals absorb shorter wavelengths more effectively than longer wavelengths [6,8]; however, there are limitations to how effectively material is removed at shorter wavelengths determined by the microscope optics used. Because the amount of light transmitted through the microscope objectives required for each wavelength varies, the "effective" absorption of the material at a given wavelength is changed.

Stainless steel 12.5 μm thick (type 304) (Trinity Brand Industries, Countryside, IL, USA) and 16 μm-thick aluminum foil (Reynolds Group Holdings, Auckland, NZ, USA) were machined using the 1064 nm IR wavelength through the 100× microscope lens. The absorbances of these materials at this wavelength is 37% and 5%, respectively [42]. Kapton® of thickness 12.5 μm (DuPont, Wilmington, DE, USA) and 25 μm-thick polyethylene terephthalate (PET) (McMaster-Carr, Elmhurst, IL, USA) were machined using the 355 nm UV wavelength through the 50× microscope lens. The absorbances of these materials at this wavelength are 22.5% and 12%, respectively [40,43]. Finally, Silicon (University Wafer, Boston, MA, USA) and polydimethylsiloxane (PDMS) (Dow Corning, Midland, MI, USA) were micromachined using the 532 nm green wavelength through the 10× microscope lens. The absorbances for these materials at this wavelength are 25% and 72%, respectively [10,12,49].

#### *4.3. Laser Characterization*

To establish protocols for processing each material, characterization grids were machined for all six (6) materials. These grids consisted of 100 spots, as shown schematically in Figure 5, and were designed in SolidWorks (Dassault Systems, Waltham, MA, USA).

DXF pattern files were subsequently uploaded into the New Wave Laser program, and each spot was assigned a specific frequency (range of 5 Hz to 50 Hz) increasing by 5 Hz increments along the x-axis of the grid, and a specific energy from 10% (0.27 mJ) to 100% (2.7 mJ) increasing along the negative y-axis of the grid by 10%. The laser spot size was varied for each grid in order to provide full characterization of the laser's capabilities over a wide range of fluence. Grids were subsequently patterned in all 6 materials using the multimodal laser.

All grids were imaged using scanning electron microscopy (JEOL JSM-6480, Tokyo, Japan). Full images of the grids were obtained, as well as images of the individual spots in both flat and 45◦ angled orientations. ImageJ (NIH, Bethesda, MD, USA) was used to characterize the depth of the laser cut, as well as the resultant spot size.


**Figure 5.** Layout of the grid of spots laser cut into the materials. The grid was designed to have 100 μm distances between the spots.

#### *4.4. Shadow Masks*

Shadow masks for the patterning of materials were fabricated from Kapton®, SS, PET, and aluminum substrates. These materials were chosen because they are all commonly available in most microfabrication laboratories and it was determined that they could be ablated all the way through the material using the multimodal laser, a necessity for shadow masks. Full coverage, circle-on-line IDE shadow masks were designed using SolidWorks and machined using the multimodal laser and the appropriate wavelength for the substrate material. The shadow masks were then imaged using the scanning electron microscope (SEM, JEOL JSM-6480, Tokyo, Japan) to characterize the design (CAD dimensions) to device (fabricated shadow mask) translation. These shadow masks were subsequently used to pattern both metal and gelatin for design to device studies, as described in Section 4.5.

Additionally, traditional interwoven comb IDE shadow masks were fabricated to test the lowest feature size limits of the multimodal laser. These structures allow for more rapid shadow mask fabrication at the lowest possible widths ensuring higher sensitivity in IDE assays; thus, allowing for any necessary adjustments to achieve the desired electrode gap width. Due to the size of these structures, atomic force microscopy (AFM) (Anasys Instruments, Santa Barbara, CA, USA) and SEM (JEOL, Tokyo, Japan) were used to characterize the electrode gap widths of these shadow masks.

Lastly, because of the widespread use of PDMS in microfluidics, several lines of 2 mm length were laser micromachined into this material in order to determine the depth of cut for a given number of laser passes. All lines were micromachined through the IR microscope lens with the green laser at 2.7 mJ with an X-Y aperture area of 50 μm2. The number of laser passes was varied from 5 to 40 passes. In this mode of operation, the laser beam is continuously scanned across the surface of the material in the desired pattern.

The process flow for the shadow mask fabrication and subsequent patterning of organic and inorganic layers is depicted in Figure 6.

**Figure 6.** Schematic of laser micromachined shadow mask fabrication and subsequent materials patterning demonstration.

#### *4.5. Patterning of Organic and Inorganic Layers*

In order to access the accuracy of low power, multimodal laser micromachining, material patterning through multimodal laser micromachined shadow masks was performed.

#### 4.5.1. Metal Patterning

The Kapton®, SS, aluminum, and PET multimodal laser micromachined shadow masks were affixed onto glass microscope slides (Fisher Scientific, Hampton, NH) with Kapton® tape (DuPont, Wilmington, DE, USA). Titanium-Gold (Ted Pella, Redding, CA, USA) electrodes, traces, and contact pads of 5nm-30 nm thickness respectively were subsequently deposited (Deposition rate: 1 Å/s at <sup>1</sup> <sup>×</sup> <sup>10</sup>−<sup>6</sup> Torr) through these shadow masks onto glass microscope slides via electron beam evaporation (Thermionics, Port Townsend, WA, USA). After metal patterning, the shadow masks were removed by carefully detaching the Kapton® tape from the glass slide with tweezers. Transmitted light microscopy (TS2 Inverted Microscope, Nikon, Tokyo, Japan) was used to image the patterned metal structures. ImageJ (NIH, Bethesda, MD) was used for further optical analysis of the captured images and measurement of the electrode structures.

#### 4.5.2. Gelatin Patterning

Gelatin (Millipore Sigma, St. Louis, MO, USA) and deionized (DI) water were mixed to make a 5 wt % gelatin solution to act as a sample bioink. Kapton® IDE shadow masks were dipped in DI water to allow for better adherence to the glass microscope slide substrates. The 5 wt % gelatin solution was subsequently pipetted onto the glass slides through the shadow masks. The gelatin solution was allowed to set, then the shadow masks were carefully removed to expose the gelatin IDE patterns. The resultant gelatin structures were imaged using a Nikon TS2 inverted microscope.

#### *4.6. Impedance Characterization of Metal Patterns*

The full spectrum (10 Hz to 100 kHz) impedance of each of the resultant patterned metal IDE structures was measured with a BODE 100 impedance measurement system (Omicron Labs, Klaus, Austria). Dulbecco's phosphate buffered saline (DPBS) (Gibco, Waltham, MA, USA) acted as the electrolyte solution and platinum-titanium wire (eDAQ, Deniston East, NSW, Australia) was used as a counter electrode used during the impedance measurements. Impedance data was extracted and plotted with Origin 2016 (OriginLab, North Hampton, MA, USA).

#### **5. Results and Discussion**

#### *5.1. Laser Characterization*

Laser characterization grids were fabricated for each of the six materials for a range of spot sizes as described in the Materials and Methods section. The spot sizes for the materials depended on the microscope objective that was used for the ablation. The spot sizes ranged from 50 μm down to 2 μm for IR ablation through the 100× lens for SS and aluminum, 60 μm down to 4 μm for UV ablation through the 50<sup>×</sup> lens for Kapton® and PET, and 250 <sup>μ</sup>m down to 20 <sup>μ</sup>m for green ablation though the 10× lens for silicon and PDMS. SEM imaging of these grid structures, and subsequent processing in ImageJ was used to calculate the depth of cut for each spot micromachined by the laser. Figure 7 shows examples of the characterization grids for each of the six materials at the maximum spot size for each wavelength. For PET, Kapton®, SS, and aluminum, full ablation (through vias, all the way through the substrate) is achieved for the full range of power and frequency combinations shown in Figure 5. Since these materials were fully ablated within the 10 pulses used, the ablation depth was averaged over the number of spots. Some deviation from the theoretical calculation for these materials was observed. Ablation rates of 46.3 μm/mJ for Kapton® and SS, 92.6 μm/mJ for PET, and 59.3 μm/mJ for aluminum were achieved. Neither silicon nor PDMS were able to be fully ablated through for any power and frequency combination. Silicon began measurable ablation at a minimum of 35 Hz and 2.43 mJ and had an ablation rate of 1.4 μm/mJ. PDMS began measurable ablation at a minimum of 35 Hz and 2.16 mJ, which gives an ablation rate of 1.5 μm/mJ.

**Figure 7.** Scanning electron microscope (SEM) images of the laser characterization grids for each of the six materials. Kapton®, SS, polyethylene terephthalate (PET), and aluminum are all at the maximum spot size for their respective wavelengths. Silicon and polydimethylsiloxane (PDMS) are both at 50% spot size to show the grid in its entirety. All scale bars are 200 μm.

Traces (lines) were scribed in PDMS using the green laser through the IR lens with between 5 and 40 passes and a 50 μm laser spot size. Cross-sectioning the scribed lines carefully using a razor blade, followed by SEM imaging of the cross section, resulted in the measurement of the depth of cut for each number of laser passes. Figure 8 shows the linear relationship that was found between the depth of

cut and the number of laser micromachining passes. Characterizing this relationship allows for the fabrication of microchannels of specific depths in PDMS by varying the number of laser passes.

**Figure 8.** Results from scribing lines in PDMS with a varied number of laser passes. The depth of cut increases linearly with the number of laser passes as expected. This allows for the direct tuning of microchannel depth by selecting the number of passes the laser makes in order to cut PDMS.

#### Comparison to Theoretical Values

Figure 9 and Table 1 show the comparison between the theoretical ablation depth model and our experimental results. For silicon and PDMS, the thermodynamic model fits well. The comparison in ablation depth per pulse for both materials between theory and practice is in the same order of magnitude. Furthermore, the ablation depth per pulse/ depth of cut plots for both materials show a clear correlation between the experimental and theoretical data. Neither one of these materials are ablated all the way through by the laser.

**Figure 9.** Comparison of theoretical and experimental ablation depth for a range of fluences in silicon.


**Table 1.** Comparison of theoretical and experimental ablation depths.

\* Limited due to full ablation of materials in less than 10 pulses.

Conversely, deviation from theory is observed in the experimental ablation depth per pulse comparison for Kapton®, SS, PET, and aluminum. Kapton® and SS perform similar to silicon and PDMS as observed in Table 1 with both the theoretical and experimental values in the same order of magnitude with experimental values being smaller than theoretical values. The discrepancy between PET and aluminum is roughly an order of magnitude higher predicted theoretical values. The laser ablated through these materials at some point (difficult to measure experimentally in our current setup), so deviation is expected in all these materials. For example, both Kapton® and SS could have shown an average ablation depth of ~3 μm per pulse, and both PET and aluminum could have shown an average ablation depth of approximately ~16 μm per pulse. Because of the thickness of these materials, when the normalization of the ablation depth to the number of pulses is performed, the experimental values are determined to be substantially lower. The key results from the experimental data in this case is that the laser is capable of fully ablating these thicknesses and that a reasonable number of pulses can be used to ablate even thicker samples of these materials. These results also show the limitations of a thermodynamic approach, as the material thickness is not considered in the calculations.

#### *5.2. Applications*

Two applications of this multimodal laser micromachining technique were additionally demonstrated in this work namely the microfabrication of shadow masks and IDEs.

#### 5.2.1. Shadow Masks

Shadow masking technology is an integral part of fabricating micro/nanostructures for prototyping in microelectronics, optical, microfluidic, MEMS, packaging, and biomedical lab-on-a-chip applications [11,23]. Typical methods for producing shadow masks, such as photolithography and deep reactive ion etching (DRIE) or ion beam milling, are expensive, require cleanroom-based fabrication, expensive vacuum equipment, ultra-pure air filtration, and advanced know-how [3,50]. Multimodal laser micromachining, on the other hand, is simple, cost effective, and makerspace-compatible, all vital attributes for cell-based assays and microfluidics applications.

Fabrication of the shadow masks down to ~1.5 μm was successfully demonstrated, as shown in Figure 10. To the best of our knowledge, this is the lowest feature size demonstrated for laser defined shadow masks [3,23]. Previous work with laser micromachining has produced feature sizes down to ~10 μm. As a result, patterns that are an almost an order of magnitude better than the state of the art are reported in this paper.

**Figure 10.** Images of sub-5 μm trace widths in Kapton®. (**a**) Atomic force microscopy (AFM) image of 3.5 μm trace width (white area). (**b**) AFM image of 1.5 μm trace width (white area). (**c**) SEM image of full comb finger electrode structure (approx. 5 μm trace width). (**d**) SEM image of shadow mask feature of approximately 2.43 μm trace width.

#### 5.2.2. Patterning Through Shadow Masks

The laser micromachined shadow masks were further used to pattern both metal and gelatin/bio-ink as described in Section 4.5. The organic and inorganic layer patterning can be utilized for applications such as accurate cell placement in single cell and culture assays, precision confinement and growth of cellular constructs, tissue engineering, metal micro/nanoelectrodes, definition of organic insulation layers, and other lab-on-a-chip and diagnostic applications [3,23,41,49,51].

Design of a shadow mask to microfabricated device translation for the four materials that the multimodal laser was completely able to micromachine in its entirety are shown in Figure 11. Aluminum, PET and SS shadow masks were used to fabricate metal IDEs while Kapton shadow masks were used for fabricating a gelatin/bio-ink IDE. It was observed that Kapton® and SS demonstrated the best design to device translation for 125 μm to 7 μm. Both SS and Kapton® showed minimal thermal damage from laser microfabrication at their respective laser wavelengths (1064 nm and 355 nm). Additionally, both materials have coefficients of thermal expansion (both approximately <sup>20</sup> <sup>×</sup> <sup>10</sup>−<sup>6</sup> <sup>K</sup><sup>−</sup>1) which are 2<sup>×</sup> larger than the thermal expansion of glass (9 <sup>×</sup> 10−<sup>6</sup> K−1) [46] theoretically suggesting better translation results and experimentally verified in Figure 11 (for N = 3 measurements at the various design values). The best design to device translation was observed to be ~98% for the 7 μm gelatin features due to the rapid deposition and curing of the gelatin. The worst design to device translation for SS was observed to be 95% for 125 μm metal IDE on glass due to the higher run-off possibilities during the electron beam evaporation process [5]. PET and aluminum demonstrated a deviation from the designed IDE pitch by 11.2% and 25.8% at maximum pitch, respectively. PET has a coefficient of thermal expansion (80 <sup>×</sup> 10−<sup>6</sup> K<sup>−</sup>1) [46] that is nearly an order of magnitude larger than that of glass, so some deviation from the design dimensions is expected due to thermal mismatch effects during e-beam evaporation. The aluminum shadow mask exhibited physical melting during the laser micromachining process, which could explain the large deviation from design dimensions.

**Figure 11.** Design to shadow mask to device (interdigitated electrode (IDE) on glass) for the four materials that cut all the way through the substrate: (**a**) stainless steel to Ti-Au metal IDE; (**b**) Kapton® to gelatin IDE pattern; (**c**) aluminum to Ti-Au metal IDE, and (**d**) PET to Ti-Au metal IDE.

Figure 12 shows the design schematic, SEM images of the shadow masks, and transmitted light microscope images of both Ti-Au metal patterning and gelatin bioink patterning for circular fill IDE patterns with electrode gaps from 125 μm to 7 μm. Metal patterning appeared to work best with Kapton® or SS shadow masks because their coefficients of thermal expansion are closer to that of glass. Kapton® and PET shadow masks worked best for gelatin patterning. Both shadow mask materials allowed for simple adhesion to the glass substrate utilizing surface tension effects by simply dipping the mask in DI water prior to attachment on glass substrates. Allowing the gelatin to fully set prior to shadow mask removal was key in preventing the patterns from bleeding.

**Figure 12.** Results of laser micromachining of the shadow masks, as well as inorganic and organic layer patterning. The design of an interdigitated electrode (IDE) was translated into metal (titanium-gold) and bioink (gelatin).

The impedance of the metallized IDEs (Figure 13) was found to decrease with decreasing electrode gap width as is expected [52] with the 1 kHz impedance decreasing by 33.47% between the extremities of the designs tested. Table 2 further illustrates impedances at key frequencies, clearly depicting resistive behavior at the lower and upper ends of the spectra and capacitive behavior in the mid-band with increasing impedance values as expected [52].

**Figure 13.** Full spectrum impedance measurements of several IDEs of varying pitch. The electrophysiologically significant frequency of 1 kHz (green line) reports impedances from approximately 39 kΩ (7 μm) to 56 kΩ (62 μm).


**Table 2.** Values of the real part of the impedance at significant frequencies.

#### **6. Conclusions**

Complete characterization of the laser micromachining processes for six (6) commonly used microfabrication materials was developed in this work using a multimodal laser micromachining tool. Characterization of the QuikLaze 50ST2 multimodal laser for the laser micromachining of six (6) different materials demonstrated that the ablation depths that were experimentally obtained fit relatively well with a simple thermodynamic theory for most of the materials. While more complex theories or analysis [47] could improve the discrepancy, this technique is accurate enough to allow one to readily calculate possible ablation depths of a new material using such a laser micromachining tool. Additionally, two applications of multimodal laser micromachining were demonstrated: shadow mask fabrication and patterning of organic and inorganic materials in the sub-5 μm range and IDE fabrication in the sub-10 μm range. The ability of such a technique to allow for rapid prototyping of shadow masks and devices, combined with the compact, benchtop-friendly design gives multimodal laser micromachining tremendous promise as an efficient fabrication method in academic and industrial research settings.

**Author Contributions:** C.H. and S.R. conceived and designed the experiments. C.H. developed the theoretical background for laser micromachining. She additionally performed all the laser micromachining experiments with the six (6) different materials. C.H. imaged all the samples and plotted the data. C.H. and S.R. analyzed the data and interpreted it. C.H. and S.R. wrote and edited the paper. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding. It was supported by Rajaraman's startup funding at UCF.

**Acknowledgments:** The authors would like to thank Charles Didier for helping design the schematics. We would also like to thank the UCF Materials Characterization Facility for use of the SEM. Additionally, we would like to thank the Rajaraman's startup funding at UCF that funded this work.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Article* **A Low Contact Impedance Medical Flexible Electrode Based on a Pyramid Array Micro-Structure**

#### **Song Wang 1, Jin Yan 2, Canlin Zhu 1, Jialin Yao 2, Qiusheng Liu <sup>1</sup> and Xing Yang 1,\***


Received: 30 October 2019; Accepted: 27 December 2019; Published: 1 January 2020

**Abstract:** Flexible electrodes are extensively used to detect signals in electrocardiography, electroencephalography, electro-ophthalmography, and electromyography, among others. These electrodes can also be used in wearable and implantable medical systems. The collected signals directly affect doctors' diagnoses of patient etiology and are closely associated with patients' life safety. Electrodes with low contact impedance can acquire good quality signals. Herein, we established a method of arraying pyramidal microstructures on polydimethylsiloxane (PDMS) substrates to increase the contact area of electrodes, and a parylene transitional layer is coated between PDMS substrates and metal membranes to enhance the bonding force, finally reducing the impedance of flexible electrodes. Experimental results demonstrated that the proposed methods were effective. The contact area of the fabricated electrode increased by 18.15% per unit area, and the contact impedance at 20 Hz to 1 kHz scanning frequency ranged from 23 to 8 kΩ, which was always smaller than that of a commercial electrode. Overall, these results indicated the excellent performance of the fabricated electrode given its low contact impedance and good biocompatibility. This study can also serve as a reference for further electrode research and application in wearable and implantable medical systems.

**Keywords:** flexible electrode; polydimethylsiloxane (PDMS); pyramid array micro-structures; low contact impedance

#### **1. Introduction**

Electrodes are important components of implantable or wearable medical monitoring systems. They are used to collect electrocardiography (ECG) and electroencephalogram (EEG) signals. If the collected signals are inaccurate, patient treatment may be delayed, their conditions worsened, or their lives endangered. Therefore, high-quality electrodes are significant for precisely monitoring the physiological parameters of the human body. At present, ECG, EEG, electromyogram (EMG), or electro-ophthalmogram (EOG) signals are measured on a standard silver/silver chloride (Ag/AgCl) wet electrode for medical diagnosis [1]. Such electrodes usually reduce the contact impedance between the skin and electrode by using conductive gels [1]. Standard Ag/AgCl wet electrodes enable low electrode–skin contact impedance and the acquisition of signals with high stability and repeatability. However, the conductive gel may cause problems such as becoming dry with time, which increases the contact impedance and aggravates measurement errors [2]. Conductive gels can also sometimes cause skin irritations or allergic rashes [3,4].

The above problems have prompted researchers to explore dry electrodes with other conductive materials [4,5] and the use of microelectromechanical system technology [6]. For example, Fei Yu et al. (2012) prepared a flexible dry electrode based on parylene substrate to collect ECG signals in zebrafish; their electrode exhibits a good signal-to-noise ratio [7]. Electrode impedance is a criterion for evaluating electrode performance. A smaller impedance corresponds to better collected-signal quality [8]. Some progress has been made in the study of dry electrodes, but some challenges remain, such as those related to reducing the impedance of flexible dry electrodes. To achieve such reduction and obtain high-quality signals, this study theoretically analyzed the impedance of electrodes and the electrode–skin contact impedance. We then proposed a method of reducing electrode impedance by fabricating pyramidal array microstructures on polydimethylsiloxane (PDMS) substrate and adding a parylene transition layer between PDMS and metal film. Finally, the impedance of the flexible electrode was experimentally tested and analyzed.

#### **2. Theoretical Analysis of Flexible Electrode Based on Micro-structure**

#### *2.1. Theoretical Analysis of Selecting Material of Flexible Electrode Substrate*

To reduce electrode impedance and enhance the bonding force between substrate and metal film, we selected substrate materials for flexible electrodes on the basis of two aspects: one is the bonding degree between substrate and biological tissues, and the other is flexibility that should be similar to that of biological tissues. The first basis for substrate selection considers the degree of adherence between the substrate and biological tissues. A material should have flexibility similar to that of biological tissues, i.e., they should be easily bent and stretched. The second basis for substrate selection considers the flexibility, i.e., polymer materials with good elasticity should be selected. Young's modulus of PDMS in polymer (0.4–1.0 MPa) is much smaller than that of other polymer materials, such as parylene (2400–3200 MPa). Thus, PDMS is easier to bend and deform and is more suitable for the fabrication and bonding of flexible substrates with large thickness. At the same time, PDMS is inexpensive and simple to manufacture; it can be prepared by simple spin coating. These characteristics indicate the suitability of PDMS materials as the substrate material for flexible electrodes. However, considering the impedance stability and thermal expansion between substrate and metal, two major problems arise in selecting PDMS material as substrate to directly bond with metal. On one hand, a porous PDMS material has strong permeability, and an electrode easily absorbs water owing to capillary action, which can lead to changes in electrode impedance. On the other hand, the weak adhesion forces between PDMS and metal confer difficulty in the deposition and transfer of a metal electrode onto PDMS [9]. Considering that the thermal-expansion coefficient of PDMS (αPDMS <sup>≈</sup> 20 <sup>×</sup> 10−<sup>5</sup> K) relatively differs from that of gold (αAu <sup>≈</sup> 1.42 <sup>×</sup> <sup>10</sup>−<sup>5</sup> K) [10], the metal film on PDMS substrate during electrode fabrication and storage is prone to cracking because of changes in temperature. Consequently, electrode impedance increases.

To address the two problems, a layer of parylene can be introduced between PDMS substrate and gold electrode. On one hand, the water permeability of parylene is very low, so water cannot easily penetrate parylene. The combination of parylene and PDMS can offset the shortcomings of the water absorption of PDMS and overcome the change in electrode impedance. On the other hand, parylene is also a common material for preparing flexible substrates. Its thermal-expansion coefficient (αparylene <sup>≈</sup> 3.5 <sup>×</sup> 10−<sup>5</sup> K) is similar to that of gold [10], and it is hardly affected by cracks caused by temperature factors after bonding with gold. The resulting PDMS–parylene layer possesses a more favorable combination of stiffness and flexibility that can improve film manageability, and the fabrication process is straightforward [11]. This process overcomes the disadvantages of the commonly used PDMS flexible electrode, which induces impedance changes owing to its easy water absorption and crack and wrinkle formation.

#### *2.2. Theoretical Analysis of Contact Impedance of Flexible Electrode–skin*

To analyze electrode–skin contact impedance, selecting a suitable impedance circuit model is very important. We utilize the H. Feriberger equivalent circuit model, which is extensively used in medicine [12], as is shown in the following Figure 1:

**Figure 1.** Skin–electrode contact impedance equivalent circuit diagram.

Where *Rc*, *Rs*, *RN*, and *Ze* are the impedance of capacitance, impedance of electrode - skin, internal fixed impedance of human body (a constant value) [12] and the total impedance(a series–parallel structure of skin impedance, skin capacitance, and internal impedance of human body). *us* is the given constant voltage, *Cs* is the skin capacitance between two electrodes. ε<sup>0</sup> and ε*<sup>r</sup>* are the vacuum dielectric constant and skin dielectric constant, respectively. *A* is the contact area of flexible electrodes with the skin, *d* is the distance between flexible electrodes, ρ is the skin impedance coefficient, ω is the angular frequency. The related formulas are as follows:

$$C\_s = \frac{\varepsilon\_0 \times \varepsilon\_r \times A}{d} \tag{1}$$

$$R\_s = \rho \times \frac{d}{A} \tag{2}$$

$$R\_c = \frac{1}{j\omega \times \mathbb{C}\_s} = \frac{d}{j\omega \times \varepsilon\_0 \times \varepsilon\_r \times A} \tag{3}$$

$$R\_c // R\_s = \frac{\frac{\rho d}{A} \times \frac{d}{j\omega \varepsilon\_0 \varepsilon\_r A}}{\frac{\rho d}{A} + \frac{d}{j\omega \varepsilon\_0 \varepsilon\_r A}} = \frac{\rho d}{A(1 + j\omega \varepsilon\_0 \varepsilon\_r \rho)}\tag{4}$$

$$R\_{\varepsilon} = R\_{\varepsilon} / / R\_{\varepsilon} + R\_{N} = \frac{R\_{\varepsilon}}{1 + j\omega R\_{\varepsilon} \mathbb{C}\_{\varepsilon}} + R\_{N} = \frac{\rho d}{A(1 + j\omega \varepsilon\_{0} \varepsilon\_{I} \rho)} + R\_{N} \tag{5}$$

As previously mentioned and analyzed, a smaller electrode–skin contact impedance corresponds to a higher quality of collected signals. Thus, the contact impedance directly affects electrode performance. According to Equations (2) and (3), a larger contact area of electrodes with the skin means a smaller impedance of the electrode–skin contact capacitance. Moreover, the total electrode–skin contact impedance decreases according to Equations (4) and (5), so increasing the contact area between the electrode and skin can produce high-quality signals. To increase the contact area with skin per unit area, some special microstructures were designed and fabricated on the electrode contact surface. These microstructures were micro-pyramidal arrays with a base of 8 μm × 8 μm and an 8 μm interval between each pyramidal microstructure. This dense pyramidal array effectively increased the contact area with skin per unit area without increasing the electrode size. The images of the pyramidal microstructures are shown in Figure 2.

**Figure 2.** The SEM photographs of pyramidal microstructure arrays.

In this study, 781,250 microstructural pyramids were fabricated on the surface of PDMS flexible dry electrode with dimensions of 2 cm (L) × 1 cm (W). The total surface area of the flexible dry electrode with micro-pyramidal array was 2,363,072 μm2, which was 363,072 μm2 larger than that of the planar flexible dry electrode. Therefore, the contact area of the flexible dry electrode can be increased by 18.15% through the fabrication of pyramidal microstructures on PDMS substrates, thereby effectively reducing the contact impedance between electrode and skin.

#### **3. Fabrication of Flexible Electrodes**

The fabrication of flexible electrodes based on PDMS pyramidal arrays was divided into many steps. Firstly, a mold of pyramidal array pits was fabricated by etching on silicon wafers. The etching method was as follows. Apply reactive ion etching (RIE) technology to etch silicon nitride with a thickness of 300 nm on the upper surface of silicon wafer for 30 min. Then use the anisotropic wet etching method of silicon: the silicon wafer was put into 33.3% potassium hydroxide solution for etching, while heating in water bath at 80 ◦C for 10 min.

Secondly, fabricate PDMS flexible substrate with pyramid microstructure. The method was as follows. Put the silicon template with pyramid microstructure into acetone, ethanol, deionized water and mold release agent in turn and perform ultrasonic treatment for 10 min respectively. Then take out the silicon wafer and dry it to obtain the pretreated silicon template. At the same time, we mixed the PDMS liquid elastomer and curing agent uniformly with the mass ratio of 10:1 and placed the mixture in a vacuum chamber to be pumped for 15 min so that bubbles in the mixture could be fully discharged. After that, pour the PDMS mixture on the pretreated silicon template with pyramid microstructure. At the same time, spin-coat the PDMS mixture uniformly with a spin coater. Next, place the silicon wafer with PDMS mixture in an oven for baking at 70 degree for 2 h to solidify the PDMS mixture.

After solidification, strip the solidified PDMS from the silicon wafer to obtain a 310 μm thick PDMS flexible substrate with pyramid microstructure, and then a 3.5-μm-thick parylene-C layer was fabricated on the prepared PDMS substrates by chemical vapor deposition. Finally, a metal film was magnetron sputtered onto parylene-C. In this study, gold was selected as the metal layer material with 40 nm thickness. This fabrication was simple and inexpensive as the fabrication of PDMS film substrate did not require any special deposition equipment. The fabricated PDMS film was also very flexible. The detailed fabrication process of flexible electrodes is shown in Figure 3.

**Figure 3.** Flow chart of flexible electrode fabrication.

#### **4. Testing Performance of Electrode**

#### *4.1. Electrode Self-Impedance Testing*

In this experiment, two types of PDMS flexible electrodes with and without a parylene transition layer were compared and analyzed. The impedance of the electrode was tested at 20 Hz. The impedance of PDMS pyramidal array electrode without the parylene transition layer was 74.94 MΩ, and that of PDMS pyramidal array electrode with the parylene transition layer was 49.52 Ω. After measuring the electrode impedance, we also observed the morphology of the two electrodes under a microscope, and the results were as follows.

According to Figure 4, several cracks were present on the metal film of the electrode without the parylene transition layer, and some of the micro-pyramid structures collapsed and cracked. Meanwhile, the contact surface morphology of the PDMS electrode with the parylene transition layer was intact, barely producing micro-cracks. This finding was due to the thermal-expansion coefficient of PDMS greatly differing from that of gold, which caused the destruction of gold on the pyramid structure when the metal was directly sputtered. The thermal-expansion coefficient of parylene is close to that of gold, but pyramid deformation can hardly be transmitted and did not affect the gold film during magnetron sputtering after the parylene transition layer was added. The experiments demonstrated that adding the parylene transition layer between PDMS and gold can effectively prevent the breakage of the metal film, thereby ensuring the small and stable impedance of PDMS electrode with pyramidal array microstructure.

#### *4.2. Characterization and Testing Performance of Electrode–Skin Contact Impedance*

To verify whether the fabricated flexible electrode based on PDMS pyramidal array microstructure had low electrode–skin contact impedance, we used a commercial electrode (Tianrun Sunshine (Beijing) Commercial Products Co., Ltd. Beijing, China) with the same size to perform comparative experiments. For impedance analysis, a Precision Impedance Analyzer (Wayne Kerr 6500B, made in UK, Wayne Kerr Electronics, West Sussex, UK) was used. We obtained the spectrum of the electrode–skin contact impedance for flexible and commercial electrodes separately by scanning the frequency, whose interval was set from 20 Hz to 1 kHz. The driving voltage was set at 1 V. The two electrodes were attached to the left arm of a test person, and their center distance was kept at 10 cm. The experimental data were analyzed using MATLAB (R2017b, The MathWorks, Inc., Natick, MA, USA).

**Figure 4.** Scanning electron microscope (SEM) photographs of metal films on electrodes (**a**) Electrodes without parylene transition layer (**b**) Electrodes containing the parylene transition layer.

Figure 5 shows that the difference between the fabricated flexible electrode and commercial electrode was obvious at low frequencies; both their contact impedance and phase gradually decreased with increased scanning frequency. We found that although the scanning frequency was affected by power-frequency interference at around 50 Hz, the impedance value of the fabricated flexible electrode at any frequency was always smaller than that of the commercial electrode, indicating its superiority over the commercial electrode.

**Figure 5.** Comparison of skin contact impedance between the commercial electrodes and the flexible electrodes with pyramidal array micro-structure.

#### **5. Conclusions**

This study proposed a simple and efficient method of fabricating a low-contact-impedance and flexible electrode with pyramid array microstructure. By fabricating the micropyramid array on the electrode surface, the contact area of the fabricated electrode with the skin increased, which indirectly reduced the impedance of flexible electrodes. It is shown that the contact area of fabricated electrodes is increased by 18.15% per unit area with adding pyramid array micro-structure on the surface of the electrode through calculations.

Upon analysis of the contact impedance of the flexible electrode–skin circuit equivalent model, we studied the simple process of fabricating the flexible electrode with pyramid array microstructure. Experimental results demonstrated that the contact impedance ranged from 23 to 8 kΩ at 20 Hz to 1 kHz scanning frequency, which was always smaller than that of the commercial electrode. Therefore, the proposed method is effective and the fabricated flexible dry electrode has good performance in terms of low impedance and good biocompatibility, indicating the potential applications of the electrode in EOG, ECG, and EMG, among others.

**Author Contributions:** X.Y. and Q.L. conceived and designed the experiments and provided experimental instruments. S.W. and C.Z. performed the experiments. S.W. analyzed the data. X.Y., S.W., J.Y. (Jin Yan) and C.Z. wrote the paper. J.Y. (Jialin Yao) took part in some experiments. All authors have read and agreed to the published version of the manuscript.

**Funding:** This work was financially supported by the Beijing Natural Science Foundation (Grant No. L172040) and National Natural Science Foundation of China Project (Grant No. 61671271, 51705283 and 51735007).

**Acknowledgments:** I would like to acknowledge Professor Yonggui Dong in Department of Precision Instruments, Tsinghua University, for his detailed guidance and many valuable opinions on scientific research of this paper.

**Conflicts of Interest:** The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Article* **Rapid and Sensitive Detection of Bisphenol A Based on Self-Assembly**

#### **Haoyue Luo, Xiaogang Lin \*, Zhijia Peng, Min Song and Lifeng Jin**

Key Laboratory of Optoelectronic Technology and Systems of Ministry of Education of China, Chongqing University, Chongqing 400044, China; 20133029@cqu.edu.cn (H.L.); 201808021026@cqu.edu.cn (Z.P.); 201908131078@cqu.edu.cn (M.S.); 20152394@cqu.edu.cn (L.J.)

**\*** Correspondence: xglin@cqu.edu.cn

Received: 27 November 2019; Accepted: 27 December 2019; Published: 30 December 2019 -

**Abstract:** Bisphenol A (BPA) is an endocrine disruptor that may lead to reproductive disorder, heart disease, and diabetes. Infants and young children are likely to be vulnerable to the effects of BPA. At present, the detection methods of BPA are complicated to operate and require expensive instruments. Therefore, it is quite vital to develop a simple, rapid, and highly sensitive method to detect BPA in different samples. In this study, we have designed a rapid and highly sensitive biosensor based on an effective self-assembled monolayer (SAM) and alternating current (AC) electrokinetics capacitive sensing method, which successfully detected BPA at nanomolar levels with only one minute. The developed biosensor demonstrates a detection of BPA ranging from 0.028 μg/mL to 280 μg/mL with a limit of detection (LOD) down to 0.028 μg/mL in the samples. The developed biosensor exhibited great potential as a portable BPA biosensor, and further development of this biosensor may also be useful in the detection of other small biochemical molecules.

**Keywords:** alternating current (AC) electrokinetics; bisphenol A; self-assembly; biosensor

#### **1. Introduction**

Bisphenol A (BPA) is an important organic chemical raw material, which is widely used in the production of plastic products and fire retardant. BPA is an endocrine disruptor, which can mimic human hormones and maybe lead to negative health effects [1,2]. Studies have shown that BPA can contact humans through the skin, respiratory tract, digestive tract, and other channels. After BPA enters the body, it combines with intracellular estrogen receptors and produces estrogenic or anti-estrogenic effects through a variety of reaction mechanisms, thereby affecting endocrine, reproductive, and nervous systems, as well as causing cancer and other adverse effects [3–5]. Therefore, a rapid and sensitive detection method of BPA is of great significance.

At present, the detection methods of BPA mainly include liquid chromatography with an electrochemical method [6], chromatography-mass spectrometry [7,8], surface-enhanced Raman scattering [9], enzyme-linked immunosorbent assay (ELISA) [10], etc. These methods are complicated and require expensive equipment with professional personnel. The movement of biomolecular molecules in the detection process relies on natural diffusion, which is time-consuming and cannot meet the requirements for the rapid detection of BPA. For example, Pasquale et al. [11] have proposed a method to determine BPA levels in fruit juices by liquid chromatography coupled to tandem mass spectrometry. However, it cost about 15 min to accomplish the detection of BPA. Sheng et al. [12] have developed an optical biosensor based on fluorescence, but they required the addition of labels to generate the sensor response. Xue et al. [13] have reported a novel SPR biosensor that combines a binding inhibition assay with functionalized gold nanoparticles to allow for the detection of trace concentrations of BPA. However, the biosensor detected BPA in 60 min, which was too long for rapid detection. Recently, various types of electrochemical sensors have attracted more attention. These electrochemical sensors are always modified via different sensing materials, such as molecularly imprinted polymers [14], carbon nanotube [15,16], graphene [17,18], nanocomposites [19,20], and metal composites [21]. For example, Maryam et al. [22] have used an electroactive label-based aptamer to detect bisphenol A in serum samples with the linear range of 0.2–2 nM. Lee et al. [23] have reported a simple and label-free colorimetric aptasensor for BPA detection. They have used the spectroscopic methods, which was time-consuming and could not achieve point-of-care testing. Inroga et al. [24] have used gold leaf-like microstructures to develop a tyrosinase-based biosensor for bisphenol A detection with the limit of detect of 0.077 μmol/L. Liu et al. [25] have developed an electrochemical enzyme biosensor bearing biochar nanoparticles as signal enhancers for bisphenol A detection in water. Peng et al. [26] have developed a signal-enhanced lateral flow strip biosensor for the visual and quantitative detection of BPA based on the poly amidoamine (PAMAM) binding with antibody. The detection would cost 10 min. Liu et al. [27] have developed a solution-gated graphene transistor (SGGT) modified with DNA molecules in a microfluidic system for a recycling detection of BPA. Compared with most of the recent publications, which have reported the electrochemical method to achieve the detection of BPA, the results of this work may not be better than them. However, in this work, we have demonstrated a novel method to detect BPA based on self-assembly technology and alternating current (AC) electrokinetics effect. To our knowledge, there has not been any report that detects BPA with self-assembly technology and an AC electrokinetics effect. So, it has the tremendous potential to detect BPA more sensitively with a lower limit of detection in the near future.

In this work, a rapid, highly sensitive BPA biosensor based on self-assembly technology and AC electrokinetics (ACEK) method was developed, which could accomplish the rapid and sensitive detection of BPA. The surface of the interdigital electrode was functionalized with the antibody of BPA via self-assembly and used for the specific capture of antigen of BPA. ACEK is used to realize the enrichment of biomolecules through the manipulation of microfluidic and nanoparticles by an AC electric field [28,29]. Compared with traditional detection methods, the ACEK method has the advantages of less sample consumption, faster detection, and a simpler procedure with an appropriate AC signal applied to microelectrode sensors in sample fluids. Then, ACEK microflows accelerate molecules of BPA binding [30,31]. With the enrichment of the target molecules, the interfacial capacitance of the interdigital electrode will change. Therefore, the change of the interfacial capacitance can be used to characterize the concentration of sample solution. Compared with the previous work [32], there are some differences in this work. Firstly, in this work, we used the BPA antibody to achieve the detection of BPA, while an aptamer probe was used to detect BPA in the previous work. Secondly, we used the self-assembly technology to form a self-assembled monolayer with the mercaptan containing alkyl chains immobilized on the gold electrodes via Au–S bonds. Then, N-hydroxysuccinimide (NHS) and 1-(3-dimethylaminopropyl)-3-ethyl carbon diimide hydrochloride (EDC) were used as the crosslinking agent to assist in the formation of amide bonds between the carboxyl group of the self-assembled monolayer and the amino group of the BPA antibody. With these processes, the BPA antibody was immobilized onto the electrodes stably, while a BPA aptamer specific to BPA with the 5 thiol modifier C6 SH (Fisher Scientific, PA) was directly immobilized onto the electrodes in the previous work. Thirdly, in this work, the interdigital microelectrodes had interdigitated arrays with widths of 10 μm separated by 10-μm gaps. However, the interdigital microelectrodes had interdigitated arrays with widths of 6 μm separated by 6-μm gaps in the previous work. Lastly, in this work, a 10 kHz AC signal was applied to the microelectrodes, but a 20 kHz AC signal was applied to the microelectrodes in the previous work.

With the improved biosensor design and use of an antigen and antibody, we successfully detected BPA at the nanomolar level within only one minute. Moreover, the developed BPA biosensor is expected to be used in the detection of biological fluids. It is of great significance to promote the detection of other small biochemical molecules.

#### **2. Materials and Methods**

#### *2.1. Materials and Reagents*

The buffer solution 10 × PBS was purchased from Solarbio (Beijing, China). The 10 × PBS was diluted in deionized water to make the working solution at 0.05 × PBS, 0.01 × PBS for diluting other substances. At the same time, the 0.01 × PBS was used as the background solution. 11-mercaptoundecanoic acid (MUA) was purchased from Yuanye Bio-Technology Co., Ltd., (Shanghai, China) The MUA was dissolved in anhydrous ethanol to prepare 5 mmol/L of MUA solution for forming the gold–sulfur bond. 1-(3-dimethylaminopropyl)-3-ethyl carbon diimide hydrochloride (EDC) was obained from Sigma (St. Louis, MI, USA); EDC was dissolved in 0.05 × PBS to make 0.4 mol/L EDC solution. N-hydroxysuccinimide (NHS) was acquired from biotopped (Beijing, China), and NHS was dissolved in 0.05 × PBS to prepare 0.1 mol/L NHS solution. Then, we mixed the two solution 1 to 4. Ethanolamine was purchased from MACKLIN (Shanghai Macklin Biochemical, Ltd., Shanghai, China). Ethanolamine was diluted with 0.05 × PBS to make 1 mol/L solution for closing. The BPA antibody and antigen were purchased from QF Biosciences Co., Ltd., (Shanghai, China). The antibody was diluted with 0.05 × PBS to make two concentrations of 5.3 μg/L and 5.3 μg/mL, and the antigen was diluted in 0.01 × PBS to make the solution that included 0.028 μg/mL, 0.28 μg/mL, 2.8 μg/mL, 28 μg/mL, and 280 μg/mL for testing. Antigen and antibody ware stored at −20 ◦C. In order to guarantee the accuracy of detection, preparation of the solution was carried out on a clean bench.

#### *2.2. Preparation of Interdigital Microelectrodes*

In this work, the interdigital microelectrodes were fabricated on silicon wafers, which had interdigitated arrays with widths of 10 μm separated by 10-μm gaps [32]. Before the detection, the interdigital microelectrodes should be modified with the following steps, as shown in Figure 1. Firstly, they are immersed in acetone for 4 min with ultrasonic cleaning, rinsed in absolute ethyl alcohol for 3 min with ultrasonic cleaning, rinsed in deionized water for 3 min with ultrasonic cleaning, and dried with a drying oven. Secondly, MUA was added to the surface of the gold electrodes for forming the Au–S bonds and realizing the self-assembled monolayer [33,34]. Then, sensors were placed in the incubator overnight with the temperature at 25 ◦C. Thirdly, before the EDC and NHS solution were added to the electrodes, the electrodes surface should be cleaned with absolute ethyl alcohol and blow dried with nitrogen; then, the sensors should be placed in an incubator for 2 h. After activation of the carboxyl group, the electrodes' surface should be cleaned with deionized water and blow dried with nitrogen; then, the chambers were pasted on the sensors. After that, 10 μL of antibody should be added to the surface of the sensors, which are then placed in an incubator for 3 h with the temperature at 37 ◦C. EDC and NHS were used as the crosslinking agent to assist in the formation of amide bonds between the carboxyl group of the self-assembled monolayer and the amino group of the BPA antibody. With these processes, the BPA antibody was immobilized onto the electrodes stably. Finally, in order to enhance the specificity of detection, ethanolamine was utilized to close the unbound active sites of the electrodes' surface. Then, the sensors were placed in an incubator for 1 h with the temperature at 25 ◦C. After the above steps, the modification of the electrodes surface was accomplished. The sensor needs to be cleaned after each sensing procedure; it is possible to wash off the antigen only and reuse the sensor, which indicates that the sensor has an expected reusability of five times before the performance of the sensor is believed to be dissatisfactory.

**Figure 1.** Representation of surface modification techniques on the interdigital electrodes' surface for the detection of bisphenol A (BPA).

#### *2.3. Apparatus and Methods*

An impedance analyzer of model IM3536 (HIOKI, Ueda, Japan) was a high-precision apparatus to detect the impedance, capacitance, and resistance of the modified interdigital electrode after being dropped different concentrations of antigens solution. Firstly, different frequencies (1 kHz, 10 kHz, 20 kHz) and different voltages (100 mV, 600 mV, 1.1 V) were applied to the experiments, respectively. Here, Figure 2 showed the relationship between the normalized capacitance and different concentrations of BPA with a 10-kHz AC signal of different voltages (100 mV, 600 mV, and 1.1 V) applied to the interdigital electrodes. Finally, a 10-kHz AC signal of 600 mV was selected to be applied to the interdigital electrodes via the impedance analyzer of model IM3536 as the measuring signal. At the same time, the voltage used in the experiments is rms. In this work, before detecting antigen concentrations, each sensor should detect the background solution. However, the measurements of the background solution are made in this work only and do not need to be used in commercial application. We detected the background solution as the blank control group to highlight the antibody–antigen binding response to the change in capacitance of electrodes. Then, 10 μL of different concentrations of antigens solution were added to interdigital electrodes. The impedance, capacitance, and resistance of the modified interdigital electrodes were detected within one minute. The normalized capacitance change rate of the sensor was computed to demonstrate antigen–antibody binding, which was shown with the slope of normalized capacitance versus time (%/min). Then, the slope was linearly fitted by the least square method. The normalized capacitance was computed as *Ct*/*C*0, where *Ct* is the capacitance value at time *t* and *C*<sup>0</sup> is the capacitance value at time zero [32].

**Figure 2.** The relationship between the normalized capacitance and different concentrations of BPA with a 10-kHz AC signal of different voltages ((**a**) 100 mV, (**b**) 600 mV, and (**c**) 1.1 V) applied to the interdigital electrodes.

#### *2.4. Electrical Double Layers*

What happens when a solution comes into contact with a solid metal surface? Helmholtz built a model to attempt to explore this question [35]. He called it model H as the left half of the dotted line in Figure 3. As is shown in Figure 3, this model can be equivalent to a flat plate capacitor, and the relationship between the charge density (σ) on one side and the potential (*V*) difference between the two layers is described by the following equations [35,36], where *d* is the distance of center of the positive and negative charge.

$$
\sigma = \frac{\varepsilon\_{\mathcal{E}0}}{d} V \tag{1}
$$

$$\frac{\partial \sigma}{\partial V} = \mathbb{C}\_{H} = \frac{\varepsilon \varepsilon\_{0}}{d} \tag{2}$$

We can conclude the capacitance (*CH*) of the flat plate capacitor from Equation (1). Hence, the H model can successfully describe the common electrochemical phenomenon with two basic equations. However, the Helmholtz layer shows an obvious defect, as shown in Figure 3. In the corollary of the equation, *CH* is a constant, but in the experiment, there are spreading layers, leading to *CH* not accurately describing the surface change. We describe the diffuse layer as *CD* and call the electrical double layer (EDL) as *Cd* [36], which will be affected by the relative potential and electrolyte concentration. From this, we can deduce that *Cd* is equal to *CH* in series with *CD*. The value of *Cd* can be determined by the following equation.

$$\mathbb{C}\_{d} = \frac{\mathbb{C}\_{H}\mathbb{C}\_{D}}{\mathbb{C}\_{H} + \mathbb{C}\_{D}} \tag{3}$$

**Figure 3.** The double layers show the opposite charges equally distribute on both sides of the interface.

As mentioned above, both the charging and discharging processes of the electrical double layer (EDL) are similar to those of the parallel plate. When the electrolyte is added to the surface of interdigitated microelectrodes, the electrolyte will contact with the surface of the microelectrode in Figure 3. Thus, it can be equivalent to the parallel plate, as shown in Figure 4 [37]. When the electrode is bare, the interface capacitance of it can be described by the following equation.

$$C\_{int,0} = \frac{A\_{int}}{\frac{A\_d}{\varepsilon\_s}}\tag{4}$$

where ε*<sup>s</sup>* is the permittivity of the solution, *Aint* is the electrode area, and λ*<sup>d</sup>* is the electrical double layer (EDL) thickness.

When antibodies are immobilized to the surface of microelectrodes via self-assembly, the interfacial capacitance *Cint* is expected to change to

$$C\_{int,ab} = \frac{A\_b}{\frac{\lambda\_d}{\varepsilon\_s} + \frac{d\_{ab}}{\varepsilon\_t}}\tag{5}$$

where ε*<sup>t</sup>* is the permittivity of the antibody, *Ab* is the effective area after antibodies are immobilized to the surface of microelectrodes, and *dab* is the antibody thickness.

In the experiments, the solution including the antigens was added to the surface of interdigitated microelectrodes, on which the antibodies were immobilized. When the antigens bind to antibodies, the molecular deposition on the sensor surface become thicker, and the interfacial capacitance *Cint***,***ab* will be expressed with

$$\mathcal{C}\_{int,ag} = \frac{A\_g}{\frac{\lambda\_d}{\varepsilon\_d} + \frac{d\_{ab}}{\varepsilon\_t} + \frac{d\_{ag}}{\varepsilon\_p}}\tag{6}$$

where ε*<sup>p</sup>* is the permittivity of the antigen, *dag* is the antigen thickness, and *Ag* is the effective area of the interfacial capacitor after the binding of an antigen to an antibody. Assume that the area of interfacial capacitance is equal before and after the binding. From Equation (6), we can see that the diminution of interfacial capacitance can be lead by the thickness increase of the dielectric layer. In this work, the biosensing utilizes the change of interfacial capacitance *Cint***,***ab*. Thus, the relative changes of interfacial capacitance are used to detect the specific binding of antigens to antibodies. The relative changes of interfacial capacitance are

$$\frac{\Delta C}{C\_{\text{int},ab}} = \frac{C\_{\text{int},ab} - C\_{\text{int},ag}}{C\_{\text{int},ab}} = -d\_{ag}l(\frac{\varepsilon\_p}{\varepsilon\_s}\lambda\_d + \frac{\varepsilon\_p}{\varepsilon\_t}d\_{ab} + d\_{ag}).\tag{7}$$

Consequently, the value of **Δ***C*/*Cint***,***ab* can be used to detect the biomolecular interactions. Beyond that, measuring the value of **Δ***C*/*Cint***,***ab* can overcome the experimental difference caused by the different surface roughness of each electrode and the difference of experimental treatment in each group, and improve the accuracy of experimental results.

**Figure 4.** Changes on the electrode surface due to the specific binding of antigens to antibodies.

#### *2.5. The Binding Mechanism of Interdigital Electrode Surface*

In this work, we adopt the interdigital electrodes as the sensors. The interdigital microelectrodes are finger-shaped in its surface. At the same time, this shape can be utilized to achieve the effect of ACEK. On the surface of the self-assembled electrode, there are two forms of binding antigens to antibodies.

In Figure 5, when the AC signal is not applied to the interdigital electrodes, the antigens binding to antibodies only depend on the deposition. In this case, only a small fraction of antigens have the chance to bind to the antibodies. This process takes a long time and does not guarantee the activity of antigens and antibodies. However, when an AC signal is applied to the interdigital electrodes, the ACEK effect is generated on the surface of the electrodes. Under the action of the ACEK effect, more and more antigens are rapidly enriched near the antibodies to promote the binding [38], thus improving the accuracy, rapidity, and sensitivity of detection. Hence, in this work, the ACEK effect was used on the electrodes to accelerate the binding.

**Figure 5.** The performances of different detection environments on electrodes.

#### **3. Results and Discussions**

#### *3.1. Detection of Antigen with 5.3* μ*g*/*L Antibody*

In this study, different concentrations of BPA standers (1.2 <sup>×</sup>10−<sup>7</sup> mol/L to 1.2 <sup>×</sup>10−<sup>4</sup> mol/L) were tested to evaluate the performance of the method. A 10-kHz AC signal of 600 mV was applied to measure capacitance of the biosensor for 60 s.

Figure 6a shows the relationship between the normalized capacitance and different concentrations of BPA. Obviously, the change of capacitance was linear, and the rate of change of the curve increased as the concentrations of BPA increased, corresponding to the degree of binding between antibodies and antigens. The change rate of normalized capacitance curves was found by least square linear fitting, which provided a quantitative index of antibody–antigen binding. In Figure 6a, the slope of these capacitance curves was found to be −10.6%/min, −14.6%/min, −24.7%/min, and −37.7%/min for BPA levels at 1.2 <sup>×</sup>10−<sup>7</sup> mol/L, 1.2 <sup>×</sup>10−<sup>6</sup> mol/L, 1.2 <sup>×</sup>10−<sup>5</sup> mol/L, and 1.2 <sup>×</sup>10−<sup>4</sup> mol/L, respectively.

**Figure 6.** (**a**) Detection of different concentrations of antigen with 5.3 μg/L antibody. (**b**) The change rate of capacitance as a function of BPA concentrations in 0.01 × PBS.

In Figure 6b, we calculated the averages and standard deviations (SDs) of the biosensor response and demonstrated the correlation between the concentration of BPA and change rate of the capacitance. The range of 1.2 <sup>×</sup>10−<sup>7</sup> mol/L to 1.2 <sup>×</sup>10−<sup>4</sup> mol/L BPA samples showed change rates of −11.62%/min ± 2.08%/min, −18.53%/min ± 3.93%/min, −25.07%/min ± 2.73%/min, and <sup>−</sup>30.43%/min <sup>±</sup> 2.56%/min, respectively. In the range of 1.2 <sup>×</sup> <sup>10</sup>−<sup>7</sup> mol/L to 1.2 <sup>×</sup> <sup>10</sup>−<sup>4</sup> mol/<sup>L</sup> BPA, *dC*/*dt* was logarithmically dependent on the concentration of BPA. A negative linear correlation between *dC*/*dt* and the concentration of BPA was observed. The dependence is expressed as *<sup>y</sup>*(%/min) = <sup>−</sup>6.912x <sup>+</sup> 3.045 with a Pearson correlation coefficient *<sup>R</sup>*<sup>2</sup> <sup>=</sup> 0.985.

In the experiments, we used the impedance analyzer IM3536 to detect the change of the capacitance of the sensor when antigen–antibody binding. However, the mechanism of measurement of the impedance analyzer IM3536 is to measure the impedance and the phase angle of the device, and the capacitance is calculated from the impedance and phase angle. Considering that the developed sensor is not a pure resistor or pure capacitor, the impedance and phase angle of the sensor will change at any moment when the sensor is immersed into solution, so the capacitance has oscillation. However, the oscillation is in the range of error measurement of the impedance analyzer IM3536. In order to mitigate these oscillations in the capacitance used in the sensor transduction, it's an effective way to slow down the speed of measurement or take the average of multiple measurements.

#### *3.2. Detection of Antigen with 5.3* μ*g*/*mL Antibody*

To verify the difference in detection between electrodes modified with different antibody concentrations, 5.3 μg/mL of antibody was immobilized on the interdigital electrodes. The experimental conditions are consistent with the experiment above.

Figure 7a displays the change rate of the normalized capacitance, which was found to be <sup>−</sup>11.5%/min, <sup>−</sup>20.0%/min, <sup>−</sup>25.5%/min, and <sup>−</sup>30.8%/min for BPA concentrations at 1.2 <sup>×</sup>10−<sup>7</sup> mol/L, 1.2 <sup>×</sup>10−<sup>6</sup> mol/L, 1.2 <sup>×</sup>10−<sup>5</sup> mol/L, and 1.2 <sup>×</sup>10−<sup>4</sup> mol/L, respectively. The averages and standard deviations (SDs) of the biosensor response were exhibited in Figure 7b. From it, we can evaluate that the concentration of BPA was negatively correlated with *dC*/*dt*, and the linear correlation is expressed as *<sup>y</sup>*(%/min) = <sup>−</sup>6.93x + 1.47 with a Pearson correlation coefficient of *<sup>R</sup>*<sup>2</sup> = 0.989. It follows that the electrode with a concentration of 5.3 <sup>μ</sup>g/mL can also detect the BPA levels range from 1.2 <sup>×</sup>10−<sup>7</sup> mol/<sup>L</sup> to 1.2 <sup>×</sup>10−<sup>4</sup> mol/L, and the linearity and correlation are better.

**Figure 7.** (**a**) Detection of different concentrations of antigen with 5.3 μg/mL of antibody. (**b**) The change rate of capacitance as a function of BPA concentrations in 0.01 × PBS.

Table 1 lists the quantitative results of the recent publications with different methods to detect BPA. Compared with most of the recent publications, which have reported the electrochemical method to achieve the detection of BPA, the results of this work may not be better than them. However, in this work, we have demonstrated a novel method to detect BPA based on self-assembly technology and the AC electrokinetics effect. There is nearly no report to detect BPA with self-assembly technology and the AC electrokinetics effect. So, it has the tremendous potential to detect BPA more sensitively with a lower limit of detection in the near future.



**Table 1.** *Cont.*


#### **4. Conclusions**

In this work, a rapid, highly sensitive BPA biosensor based on self-assembly technology and the AC electrokinetics (ACEK) effect has been proposed. A higher concentration of BPA solution was dropped on the self-assembly interdigital electrode sensor, a larger number of antibody–antigen binding occurred, and a larger normalized capacitance change rate was detected for the sensor. In this work, we used antigen–antibody-specific binding to detect BPA. At the same time, the limit of the biosensor is 1.2 <sup>×</sup> <sup>10</sup>−<sup>7</sup> mol/L, which is better than some existing detection methods. For example, Sun et al. [39] have used the oscillopolarogaphic method to detect BPA in food packaging materials with a limit of detection of 4.4 <sup>×</sup> 10−<sup>6</sup> mol/L. Yan et al. [40] have developed a simple and renewable nanoporous gold-based electrochemical sensor for BPA detection with a limit of detection of 4.3 <sup>×</sup> <sup>10</sup>−<sup>7</sup> mol/L. The present work could successfully detect BPA at nanomolar (nM) levels, which was higher than the results in the previous work [32], which could successfully detect BPA at femto molar (fM) levels. Although the results of the present work were not better than the results in the previous work, we have used a novel method (self-assembly technology) to achieve the detection of BPA. At the same time, in order to acquire better results, we are improving the conditions, processes, and materials of the experiments. The novel method has the tremendous potential to detect BPA more sensitively with a lower limit of detection. Further development of the method may provide a more convenient, highly sensitive, and effective detection for BPA in complex samples.

**Author Contributions:** Methodology, formal analysis, investigation, data curation, writing—original draft, H.L., Z.P.; conceptualization, validation, writing—review and editing, supervision, project administration, funding acquisition, X.L.; resources, writing—original draft, M.S., L.J. All authors have read and agreed to the published version of the manuscript.

**Funding:** This work was supported by the Fundamental Research Funds for the Central Universities (Project No.2019CDYGZD006), National Natural Science Foundation of China (Project no. 6137 7001), the Postgraduate education and teaching reform research project of Chongqing University (Project No. cquyjg18323), Venture & Innovation Support Program for Chongqing Overseas Returnees and Fundamental Research Funds for the Central Universities (No. 10611CDJXZ238826).

**Acknowledgments:** The authors thank Xiaodong Zheng for his support in validation and writing—review and editing.

**Conflicts of Interest:** The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.

#### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Review* **Advances in Liquid Metal-Enabled Flexible and Wearable Sensors**

**Yi Ren 1, Xuyang Sun <sup>2</sup> and Jing Liu 1,2,\***


Received: 8 January 2020; Accepted: 13 February 2020; Published: 15 February 2020

**Abstract:** Sensors are core elements to directly obtain information from surrounding objects for further detecting, judging and controlling purposes. With the rapid development of soft electronics, flexible sensors have made considerable progress, and can better fit the objects to detect and, thus respond to changes more sensitively. Recently, as a newly emerging electronic ink, liquid metal is being increasingly investigated to realize various electronic elements, especially soft ones. Compared to conventional soft sensors, the introduction of liquid metal shows rather unique advantages. Due to excellent flexibility and conductivity, liquid-metal soft sensors present high enhancement in sensitivity and precision, thus producing many profound applications. So far, a series of flexible and wearable sensors based on liquid metal have been designed and tested. Their applications have also witnessed a growing exploration in biomedical areas, including health-monitoring, electronic skin, wearable devices and intelligent robots etc. This article presents a systematic review of the typical progress of liquid metal-enabled soft sensors, including material innovations, fabrication strategies, fundamental principles, representative application examples, and so on. The perspectives of liquid-metal soft sensors is finally interpreted to conclude the future challenges and opportunities.

**Keywords:** soft sensors; liquid metal; fabrication; principle; arrays; application

#### **1. Introduction**

Sensors are common and core devices for information acquisition, which can detect external stimuli and transform them into measurable signals according to specific rules. In particular, biomedical sensors are targeted at human beings and other biological organisms to monitor physiological signals of life activity for health monitoring and disease diagnosis [1].

A typical sensor mainly consists of three components, including sensing element, conversion element and electrical circuits. The sensing element directly perceives the measured quantity and converts the signal into other values following a definite relationship, while the conversion element converts the non-electric quantity sensed by the sensitive element into electrical quantity [2]. Generally, the sensitive element and conversion unit are integrated as a whole. Diverse sensing materials are utilized to sense environmental change and produce a useful signal, such as piezoelectric effect, thermoelectric effect, Hall Effect etc. [2]. According to various detecting principles, existing sensors mainly include pressure sensors, force sensors, liquid sensors, speed sensors, acceleration sensors, radiation sensors, thermal sensors and so on [3–9].

With the rapid development of soft electronics [10–20], flexible and wearable sensors have shown evident advantages in biomedical applications, such as health-monitoring, wearable devices, artificial skin and intelligent robotics [10,21–31]. For instance, strain sensors can be attached to human body to monitor skin strain and muscle movement [32]. Introduction of soft sensors has greatly propelled the development of intelligent artificial skin, which provides not only aesthetic functions, but also perception of tactile sensation and temperature measurement [33,34]. They can also be applied to realize intelligent perception and control for robots [35–37]. Moreover, flexible sensors can tightly fit the object and maintain performance even while being stretched. In this way, they can respond to changes more sensitively and detect position and posture alteration in real time.

Up to now, materials and techniques to achieve flexible sensors have been widely explored. To obtain better application output for the human body, both flexibility and biocompatibility of the substrates are required. Soft thermoplastic polymers and silicone elastomers are the most common substrates used to fabricate flexible sensors, such as polyethylene terephthalate (PET), poly urethane (PU), polydimethylsiloxane (PDMS) and Eco Flex [38]. Moreover, the sensing elements are other important parts, which are mainly made up of conductive materials, including organic and inorganic nanomaterials [39,40]. Carbon-based materials, such as carbon nanotube and graphene, have been investigated in various soft circuits [39,41–46]. Although circuits fabricated by carbon-based materials are highly stretchable, they are not as conductive as those fabricated by metals. Nanoparticles of rigid metals are applied to design soft circuits with high conductivity, among which silver nanoparticles have been extensively studied [47–53]. Moreover, liquid metal, mainly refers to metal whose melting point is around the room temperature, is increasingly popular in the area of soft circuits because it possesses both good deformability and conductivity [54]. The room-temperature melting alloys can be fabricated by stirring in a liquid state, and sometimes heating is needed. For instance, the conductivity of gallium is 2.2 <sup>×</sup> 106 *<sup>S</sup>*/*m*, which is a little lower than that of silver but is much higher than that of organic materials [54]. Furthermore, liquid metal is more flexible and cheaper than silver. Thus, progress in liquid metal has been expected to propel the development of soft circuits. Typical liquid metal includes gallium, bismuth, lead, tin, cadmium, indium and their alloys. These low-melting alloys can be produced by stirring in a liquid state, and some raw materials with high melting points need to be heated. When exposed to the air, liquid metal will be oxidized immediately and form a layer of protective film. Unlike conventional liquid metal like mercury, gallium and its alloys is proven to be low-toxic and biocompatible, and has been introduced to biomedical area [55].

Due to the unique properties of liquid metal, it has been vigorously studied in soft electronics, such as flexible conductors, antennas, transistors, electrodes, energy-harvesting devices, sensors, liquid robots and liquid computational systems [56–68]. In particular, liquid-metal sensors are designed on the basis of variable resistance and capacitance during stretching and bending. Compared to nanoparticles of rigid metals, liquid metal appears more fluidic and deformable. Therefore, liquid-metal based sensors would fit well to the targeted area, further reducing discomfort and displaying better sensitivity and flexibility. The main problem that must be paid attention to liquid metal during application is the risk of leaking from soft encapsulation.

In this review, the progress of soft sensors based on liquid metal is reviewed. First, we introduce the composition of liquid metal, as well as its distinct properties thus enabled. Next, fabrication methods of liquid-metal sensors and general comparisons among them are discussed. Subsequently, we move to interpret the basic principles of how liquid-metal sensors work. Following that, several typical sensors or sensor arrays based on liquid metal and their representative applications are illustrated. Finally, we point out the existing issues that would hinder further development in this area, and propose possible strategies for future research.

#### **2. Materials**

#### *2.1. Composition*

At this stage, liquid metal applied to soft electronics mainly refers to gallium (Figure 1A) and its alloys, among which EGaIn (gallium and indium) and GaIinstan (gallium, indium and tin) are most commonly available [55]. EGaIn contains 75.5% gallium and 24.5 % indium by weight and its melting point is 15.5 ◦C, while GaIinstan consists of 68.5% gallium, 21.5% indium and 10.0% tin by weight and its melting point is 10.5 ◦C. Those alloys can be prepared by stirring in liquid state and sometimes heating is required. Similar to gallium, bismuth can be mixed up with other metals to form low-melting alloys as well. Their physical properties like melting point and electric conductivity can be regulated according to the addition of different metals [69].

In particular, the properties of liquid metal can be optimized by mixing liquid metal with other metal nanoparticles. The doping ratio decides the fluidity of the mixture. When the ratio of metal nanoparticles is high enough, the mixture will be in the form of a paste. For instance, magnetism, ferromagnetic nano powders could be dispersed in liquid metal [55]. Moreover, nickel particles could be mixed with EGaIn for fast fabrication of soft electronics by direct writing or printing [70–72]. As shown in Figure 1B, Ni particles were wrapped by a liquid metal oxide layer after stirring for a moment. The introduction of Ni particles could help enhance the adhesion force of liquid metal to substrate. Furthermore, to improve its conductivity, one could load copper particles into liquid metal to make ever superior composite [73]. Figure 1C shows the mixture of Cu-GaIn. The doping ratio and particle diameter could be adjusted according to different application requirements. Thermal and mechanical properties were tunable as well.

**Figure 1.** Liquid metal and its mixture. (**A**) Photo of gallium droplet, reproduced with permission from [69]. (**B**) Schematic diagram of Ni-GaIn particle, reproduced with permission from [70]. (**C**) Photo of Cu-GaIn mixure, reproduced with permission from [73].

#### *2.2. Property*

It is well known that liquid metal exhibits unique physical and chemical properties, which have been extensively studied. The most outstanding feature of liquid metal is that it has both fluidity and metallicity, making it advantageous to soft electronics. Previous experiments have testified that although liquid metal is fluidic, its dynamic performance is different from that of water, such as droplet shape and splashing morphology [74]. Figure 2A shows the splashing process of water and liquid metal under 25 ◦C. Clearly, it is hard for liquid metal to form secondary droplets due to high dynamic viscosity, which is really easy for splashing water. Moreover, gallium and its alloys could be oxidized quickly when exposed to air, which would lead to an even higher viscosity, increasing the difficulty to form secondary droplets. But on the other hand, formation of the oxide skin can increase the adhesive force, thus providing convenience for fabrication of liquid metal circuits. Research has found that liquid metal possesses a high surface tension and is hardly permeable on various substrate materials [75]. Metal particles can be mixed to regulate the adhesion and wettability of liquid metal, as mentioned above.

Liquid metal shows good electrical and thermal conductivity. The electrical conductivity of EGaIn is around 10<sup>6</sup> *S*/*m* and the thermal conductivity is much higher than that of water. Moreover, when applying an electric or magnetic field, liquid metal would respond [76]. Coated with ferromagnetic materials, liquid-metal manipulation could be induced by an external magnetic field. As shown in Figure 2B, liquid-metal droplets could move following the magnet. An electrical field could induce the movement of liquid metal as well [77]. Figure 2C presents the movement of an EGaIn droplet under the control of an electrical filed. Furthermore, Tan et al. studied the movement of liquid-metal droplets with and without aluminum, respectively, and found that when there existed aluminum, liquid-metal droplets could run faster under the control of an electric field [78]. This is because that EGaIn and aluminum would constitute galvanic battery in electrolyte solution and the gas generated could propel the movement of the liquid metal. This redox can be represented by the chemical formula below:

$$2Al + 6H^{+} \rightarrow 2Al^{3+} + 3H\_{2}.\tag{1}$$

Since liquid metal shows these distinctive properties, the application of liquid metal in soft circuits is attracting increasing attention.

**Figure 2.** Typical properties of liquid metal. (**A**) Comparison of splashing process of water (**a**) and liquid metal (**b**), reproduced with permission from [74]; (**B**) Schematic of the magnet-controlled moving of an EGaIn droplet coated by ferromagnetic materials, reproduced with permission from [76]; (**C**) Sequential snapshots of a EGaIn droplet's motion, reproduced with permission from [76];.

#### **3. Fabrication**

Owing to the unique physical properties of liquid metal, different methods to fabricate liquid-metal soft circuits have been designed and explored. Up to now, printing and microfluidic technologies are most commonly used [79]. Other methods including selective wetting, laser engraving, wiping etc., are investigated as well.

#### *3.1. Printing Technology*

Based on the fluidity of liquid metal, the concept of liquid-metal printed electronics was proposed [79,80]. Researches have proven that liquid metal can be used as ink and printed directly by a brush or a roller ball pen [81–83]. Compared to a traditional printed circuit board, this made electronics fabrication more convenient and rapid. Figure 3A shows how to paint a circuit with a roller ball pen and the light-emitting diodes (LEDs) are successfully lighted. To standardize the fabrication process of soft circuits by liquid metal ink and avoid artificial errors, automatic printing of liquid metal by a versatile desktop printer was further developed [84,85]. The concrete structure of a desktop printer with liquid metal ink is presented in Figure 3B. Circuits could be precisely designed by the desktop printer and then transferred to soft substrate, such as PDMS. Those circuits would be further encapsulated for practical application [86]. Votzke et al. designed a highly stretchable strain sensor using printed liquid metal, which has been used to measure elbow flexion angle [87]. Furthermore, Gannarapu et al. proposed the method of freeze-printing to fabricate 3D curvilinear paths. Liquid metal was continuously dispensed and frozen. The printed solid network could be encapsulated in

elastomers [88]. A shortage of direct printing lies in that the resolution of circuits is sometimes limited by the resolution of the printer and can hardly be improved. To design liquid metal with smaller size, Liang et al. wrote liquid metal on a prestretched elastomeric substrate surface [89]. Then the prestretched substrate was released to recover the original size with electronic patterns shrinking on it.

Recently, mask-based printing has been designed as an alternative to promote resolution. For instance, a rigid mask could be fabricated by micromachining technology in advance. Areas covered by the mask would not be available to liquid metal and specific patterns could be deposited through the mask with an airbrush (Figure 3C) [90]. Moreover, a photoresist mask was commonly used as well, which can be removed finally, as well as the liquid metal falling on it [91,92]. However, unlike a rigid mask, a photoresist mask cannot be recycled and the process of photoetching is complex and time-consuming. Guo et al. introduced toner as the mask and liquid metal could selectively adhere to area without toner [72]. A roller was applied to deposit liquid metal and the circuit could further be transferred to elastic substrate. Figure 3D shows the whole process of Guo et al.'s method.

In general, compared to other methods, printing technology may be more convenient, but the fabrication precision is relatively limited.

#### *3.2. Microfluidic Technology*

Meanwhile, microfluidic technology is advantageous for the fabrication of stable, uniform and sealed liquid-metal circuits [93–99]. Figure 3E is the schematic of the microfluidic technique. Firstly, photoetching was applied to produce a raised pattern model. Then, the mixture of PDMS prepolymer and curing agent with the quantity ratio of 10:1 was poured onto the model to form a PDMS layer with specific grooves. Plasma bonding was utilized to form sealed micro-channels. To fill the channels with liquid metal and exhaust air, a pair of inlet and outlet holes must be produced before plasma bonding. After that, liquid metal could be injected into the channels. Circuits fabricated by microfluidic technology showed excellent stability and uniformity. Microfluidic technology has already been applied widely and commercially.

Several types of soft force sensors have, thus, been designed by this method. As shown in Figure 3F, the soft pressure sensor was fabricated by microfluidic technology. Research has proven that the soft force sensors could generate a response to external pressure through the change of capacitance and exhibit good flexibility.

**Figure 3.** Printing and microfluidic technology. (**A**) Printing circuits with liquid metal ink by a roller ball pen, reproduced with permission from [82]. (**B**) Mechanical printing of liquid metal circuits with desktop printer, reproduced with permission from [84]. (**C**) Schematic of patterning liquid metal with a rigid mask, reproduced with permission from [90]. (**D**) Schematic of patterning liquid metal with toner mask, reproduced with permission from [72]. (**E**) Schematic of microfluidic technology: (**a**) The silicon chip; (**b**) Silicon chip coated by photoresist; (**c**) Lithography model; (**d**) Pouring PDMS; (**e**) PDMS with micro-grooves; (**f**) Punching; (**g**) Plasma bonding; (**h**) Injecting liquid metal. (**F**) A stretchable pressure sensor fabricated by the microfluidic technology: (**a**) Vertical view of the pressure sensor; (**b**) Original state of the pressure sensor; (**c**) A pressure sensor under bending; Reproduced with permission from [96].

#### *3.3. Selective Wetting*

Not limited to printing and microfluidic technology, fabrication methods of flexible sensors based on liquid metal are diverse due to the unique properties of liquid metal [79]. For instance, considering that the wetting behavior of liquid metal varies with the surface properties of the substrate, selective wetting of liquid metal is studied.

Kramer et al. had fabricated elastomer conductors by deposition, which was based on the different wetting behavior between liquid metal and thin metal films [100]. Figure 4A shows the whole process of fabricating circuits by this method. The surface was sputtered with indium, which prevented liquid metal from adhering to this area. As shown in Figure 4B, the circuits thus fabricated on a soft substrate were proven to be highly flexible. Similarly, Li et al. presented a stretchable pulse sensor fabricated by selective plating as well, in which a thin film of copper would be electroplated in advance to enhance the wetting of liquid metal [101]. Clearly, surface modification could be utilized by either enhancing or

weakening the wettability of liquid metal so as to form designed patterns. The limitation of this method may be that the process of surface modification is complex and the edge of the pattern produced by selective wetting is rough.

**Figure 4.** Diverse techniques to fabricate liquid-metal soft electronics. (**A**) Fabrication process of selective liquid metal deposition: (**a**) Photoetching; (**b**) Photoresist developing; (**c**) Spinning poly acrylic acid (PAA); (**d**) Sputtering the surface with indium; (**e**) Removing photoresist; (**f**,**g**) Flooding the surface with liquid metal; (**h**,**i**) Removing excess liquid metal; (**j**) Removing indium mask; (**k**) Encapsulating the circuit. Reproduced with permission from [100]. (**B**) A circuit embedded in elastomer patterned by selective deposition of liquid metal, reproduced with permission from [100]. (**C**) Illustration of CO2 laser engraving technology, reproduced with permission from [102]. (**D**) Flexible circuits fabricated by CO2 laser, reproduced with permission from [102].

#### *3.4. Laser Engraving*

The laser engraving technique is commonly applied in the field of micromachining. CO2 laser engraving technology was introduced for rapid prototyping of soft electronics as well [102]. Figure 4C shows the direct patterning of liquid metal by a CO2 laser cutter. Firstly, a thin layer of liquid metal was sealed by two layers of PDMS and the top layer mainly functioned as protection during the laser-cutting process. PDMS could absorb the energy from the CO2 laser and be vaporized. Then liquid metal would escape together with PDMS vapor, thus forming desired patterns. After that, a new layer of PDMS would be cast to encase the circuit. Lu et al. have designed several systems through this technology, including sensors and resistive circuits (Figure 4D). CO2 laser-engraving technology is advantageous for the fabrication of any 2D structure. However, it causes a higher wastage of liquid metal since a large proportion of liquid metal escaped with vaporized PDMS, which could be avoided by collecting the liquid metal that escaped. Recently, NdYAG lasers were applied for simpler patterning of liquid metal [103]. Pan et al. produced a film composed of a liquid metal grid through direct laser writing, and proved that the circuit showed high optical transmittance, high stretchability, low resistivity and low trace width.

#### *3.5. Dewetting and Wiping*

Meanwhile, methods of dewetting [104] and wiping [105] utilized the property that liquid metal would be oxidized immediately and form a thin oxide layer when exposed to the air or oxygen. The oxide layer could increase the adhesion between liquid metal and micro-channels. Therefore, while liquid metal outside the channels was removed, liquid metal in the channels would be left owing to the higher adhesion force. The difference is that dewetting uses NaOH solution to remove excess liquid metal and wiping uses absolute ethyl alcohol (Figure 5A). Compared to dewetting by NaOH solution, wiping by absolute ethyl alcohol is more controllable since the reaction between NaOH solution and liquid metal is more intense.

As shown in Figure 5A, the first few steps of wiping were similar to microfluidic technology: photoetching was used to form the desired pattern model and micro-grooves. But inlet and outlet holes were unnecessary, which significantly simplified the fabrication process and improved the resolution. After spraying liquid metal onto PDMS by a nozzle, paper soaked with absolute alcohol was applied to remove excess liquid metal. The circuits could be encapsulated by casting another layer of PDMS. Figure 5B (a) is an electrode array with high resolution fabricated by spraying and wiping. Figure 5B (b),(c) show the detailed micrograph of the electrode and lines. It could be concluded that the channels were fully filled by liquid metal. A higher failure rate may be the main limitation of the wiping technique, but this can be made up by increasing the depth of micro-channels and repeating the operations of spraying and wiping.

**Figure 5.** Technologies of spraying and wiping. (**A**) Fabrication process: (**a**) Photolithography model; (**b**) Pouring PDMS onto the photolithography model; (**c**) PDMS substrate with micro-channels; (**d**,**e**) Spraying liquid metal with an airbrush; (**f**) Removing needless liquid metal with the help of paper soaked with absolute ethyl alcohol; (**g**) The final circuit; (**h**) Encapsulating the circuit. (**B**) An electrode array fabricated by spraying and wiping: (**a**) The electrode array; (**b**) The micrograph of the electrode plate; (**c**) The micrograph of lines. Reproduced with permission from [105].

#### **4. Basic Principle of Liquid-Metal Sensors**

#### *4.1. Liquid Metal as Soft Connection*

The simplest liquid-metal sensors are perhaps those that use liquid metal as soft connections (Figure 6). The detecting mechanism is the same as rigid ones, but the stretchability is greatly improved. Liquid metal possesses both good conductivity and fluidity, ensuring the reliability of the connections during stretching and bending. For instance, based on photoplethysmography (PPG) technology, Li et al. designed a stretchable pulse biosensor with liquid-metal interconnections and soft substrate [101]. This system was proven to be convenient and comfortable for heart-rate monitoring. In addition, Hong et al. designed a soft polyaniline nanofiber temperature sensor array, whose interconnections were made up of Galinstan [106]. It could be used as a portable and wearable device to measure body temperature. Combining liquid-metal wiring with graphene sensors, Jiao et al. developed several strain sensors, which could be used for health-monitoring and motion-sensing [107]. Meanwhile, Hu et al. put forward the idea of using liquid metal to connect the sensing element and successfully produced a sensor array, which had the ability to sense temperature and force [108]. All these research works prove that a liquid-metal connection provides advantages in fabricating flexible sensors.

**Figure 6.** Illustration of circuits using liquid metal as soft connection.

#### *4.2. Resistive Sensors*

Resistive sensors based on liquid metal have been widely researched. External change can be reflected through the fluctuations in resistance. The resistance *R* follows the formula:

$$R = \rho L / (S)\_\prime \tag{2}$$

where ρ is the resistivity of liquid metal; *L* is the length and *S* is the cross-sectional area. External heat and force can lead to the deformation of liquid metal, including thermal expansion, stretching and pressing, which further causes a change in resistance. As shown in Figure 7A, stretching can influence the length and cross-sectional area of liquid-metal lines, while pressing can change the cross-sectional area. When being bent, the shape of liquid metal does not change significantly, as well as the resistance.

Generally, liquid-metal resistive sensors include temperature sensors and force sensors. For instance, Li et al. has found the matching metal of liquid metal and designed a printable tiny thermocouple base on them. This thermocouple showed an excellent linear property when the temperature ranged from 0 to 200 ◦C [109]. Likewise, some force sensors were fabricated on the basis of resistive response, including pressure sensors [95,110], strain sensors [97,111] and shear force sensors [111]. To accurately measure the change of resistance, several specific circuit organizations were applied. Figure 7B is the schematic of a Wheatstone bridge with a Galinstan based sensor. The application of a Wheatstone bridge helps detect resistance of the sensor with high precision. When the voltage *Va* is equal to *Vb*, which means that the output sensor *Vout* is zero, the resistance of the Galinstan based sensor can be calculated by the following equation:

$$R\_x = \frac{R\_2}{R1} R\_0 \tag{3}$$

*Vout* is the amplification of *Vb* − *Va*. This circuit can sensitively reflect the change of voltage and thus the resistance of the sensor. For instance, when applying external force to the Galinstan-based pressure sensor, deformation of the soft circuit would cause the change of resistance, which further influences the output voltage *Vout*. The resistance of *R*<sup>0</sup> should be adjusted until *Vout* is zero again. Then the new resistance of the pressure sensor can be calculated, as well as the external force.

**Figure 7.** Working principle of resistive sensor. (**A**) Deformation of liquid metal under external force. Blue represents soft substrate and gray represents liquid metal: (**a**) Stretching; (**b**) Pressing; (**c**) Bending. (**B**) Wheatstone bridge to detect the resistance of the liquid-metal pressure sensor, reproduced with permission from [95].

#### *4.3. Capacitive Sensors*

Some capacitive force sensors have been designed as well, which mainly work based on the variation of capacitance [66,112–115]. The theory of capacitive sensors is similar to that of resistive sensors. In principle, the capacitance is decided by the formula below:

$$C = (\varepsilon S) / d\_\prime \tag{4}$$

where ε is the dielectric constant; *S* is the facing area and *d* is the vertical distance (Figure 8A). When applying an external force, parameters of the capacitor will be altered, causing the change of capacitance.

More complex and precise sensors can be fabricated through the series and parallel combination of capacitors. Figure 8B is a stretchable sensor with two liquid metal fibers, whose capacitance would change when being twisted. It could be used to measure torsion, strain and touch through the change of capacitance. Moreover, an inertial sensor based on a liquid metal droplet was designed as well. The liquid droplet was not fixed, so that it could modulate the capacitance between electrode 1 and 2 during movement. Compared to rigid sensors, these capacitive sensors based on liquid metal has good flexibility and stability.

**Figure 8.** Working principle of capacitive sensor. (**A**) Illustration of capacitor: s is the facing area and d is the vertical distance. (**B**) Stretchable capacitive sensors with double helix liquid metal fibers, reproduced with permission from [114].

#### *4.4. Electrochemical Sensors*

Electrochemical sensors utilize the chemical reaction occurring in solution. EGaIn shows good electrochemical behavior. Liquid metal with nickel and aluminum has been proven to be propelled by hydrogen in NaOH solution [78,116]. The movement can be controlled by an external electrical or magnetic field. Combined with saline solution, liquid metal can oxidize quickly and become gray gradually. Based on these reactions, liquid metal can be used to detect the existence of some substance and electrical field.

Moreover, Sivan et al. found that liquid-metal marbles could enhance the sensitivity of heavy metal ion sensors [117]. They produced liquid-metal marbles with controlled coating density and proved that these marbles showed enhanced electrochemical sensitivity towards heavy metal ions with excellent selectivity. It had been used to detect some specific heavy metal ions, such as *Pb*2<sup>+</sup> or *Cd*2+. Compared to resistive sensors and capacitive sensors, studies on electrochemical sensors are relatively fewer.

#### *4.5. Metamaterial Biosensors*

With the development of metamaterials, liquid-metal metamaterials are on the rise. Among all the metamaterials, split-ring resonator (SRR) arrays and complementary split-ring resonator arrays (CSRR) are the most commonly used structures for electromagnetic metamaterials, as shown in Figure 9A,B. Single SRR unit and SRR arrays based on liquid-metal have been fabricated successfully by printing (Figure 9C,D). All those results prove that liquid-metal based metamaterials are possible. The primary limitation of a liquid-metal SRR structure is that the fabrication size is not fine enough to generate electromagnetic response. Besides SRR and CSRR structures, several other types of liquid metal electromagnetic metamaterials have been proposed as well. For instance, Xu et al. fabricated a switchable metamaterial exploiting liquid metal [118]. The interference between copper and liquid metal resulted in an electromagnetically induced transparency (EIT)-like spectrum, which could be switched on or off on the basis of the fluidity of liquid metal. Moreover, a flexible metamaterial absorber based on liquid metal and PDMS was proposed [119–121]. Reichel et al. also designed a terahertz signal-processing device with liquid metal, which was electrically reconfigurable [122].

Recent studies have also applied metamaterials to biosensors for molecule detection. The sensing mechanism of metamaterials is that a change in the resonant frequency would happen when different biomolecules bind onto metamaterial resonators [123]. It is hoped that metamaterials can be combined with with liquid metal to design soft metamaterial biosensors. Generally, research on electromagnetic metamaterials based on liquid metal are still in the initial stage and more efforts are needed. Upcoming research could be conducted to find out the feasibility of liquid-metal metamaterial biosensors.

**Figure 9.** Metamaterial. Schematic diagram of split-ring resonator (SRR) (**A**) and complementary split-ring resonator (CSRR) (**B**), where the gray color represents metal and the white represents gaps. (**C**) An SRR unit fabricated by liquid metal. (**D**) SRR arrays fabricated by liquid metal.

#### *4.6. Liquid-Metal Antenna*

Antennas are important components in wireless devices for remote communication and sensing [124]. Several flexible antennas fabricated by liquid metal have been designed for implanted medical devices, interactive gaming and aeronautic remote sensing to avoid the mechanical failure of solid parts due to material fatigue, creep or wear [124–127]. A radiation pattern reconfigurable antenna was designed by Rodrigo et al., which could operate at 1800 MHz with 4.0% bandwidth [124]. Mazlouman et al. produced a frequency-reconfigurable antenna as well by injecting liquid metal into a square reservoir. Their antenna allowed frequency tuning from 1.3 to 3 GHz. Hayes et al. proposed a flexible antenna with a novel multi-layer construction based on liquid metal [126]. It was proven to be mechanically flexible and perform stably during flexing. Cheng et al. fabricated a flexible planar inverted cone antenna for ultrawideband applications [127]. It is expected that antennas based on liquid metal with improved durability and stability will be presented in the near future.

#### **5. Typical Applications**

Sensors play an important role in biomedicine and bionic area. So far, the application of soft sensors includes force sensors, temperature sensors, blood glucose sensors and so on. Based on the fluidity and metallicity of liquid metal, sensors enabled from them possess good flexibility and sensitivity. Obviously, sensors fabricated by liquid metal will become critical components for wearable devices and robots.

#### *5.1. Force Sensors*

Soft force sensors based on liquid metal, which can tolerate large deformation, have shown advantages in force detection and motion-monitoring.

For instance, pressure sensors have been researched vigorously. Previous experiments showed that when directly applying pressure to a printed liquid metal pressure sensor, the resistance would change accordingly [110]. As shown in Figure 10A, when there existed external pressure, the resistance of the liquid-metal sensor would increase. Moreover, Jung et al. manufactured a pressure sensor embedded in a microfluidic channel, which could be used to measure the viscosity of various fluids [95]. The change of the fluids' viscosity would lead to a pressure to the membrane and then the electrical resistance would increase due to the decrease in the cross-sectional area (Figure 10B). A Wheatstone bridge circuit was applied in their research to detect the change of electrical resistance accurately. Kim et al. integrated liquid metal microchannels with a 3D-printed microbump array to increase the sensitivity of the pressure sensor, which has been proven to have a wide range of applications in health monitoring [93]. Oh et al. injected liquid metal into PDMS and designed a pressure-conductive rubber sensor. This could be used to detect wrist pulse and respiration [128]. Besides pressure sensors, strain sensors are also important types of force sensors. For instance, Zhou et al. fabricated a flexible strain sensor with asymmetric structure. This could detect various human joint motions simultaneously [129].

Vogt et al. have designed a multi-axis force sensor with microfluidic channels recently [106]. This could detect not only normal shear force, but also in-plane shear force. Meanwhile, Shi et al. reported a flexible force sensor to detect normal and shear forces as well based on the piezoresistive effect and further designed an artificial hair cell sensor [130]. Figure 10C shows the working principle of the hair cell sensor. The polymer hair would be deformed when there was a shear or normal force, causing the deformation of circuits connected to the hair. The change of capacitance between liquid metal could reflect the movement of polymer hair. It perfectly imitated the operating mode of hair cells in the inner ear, which might further propel the research of an artificial ear.

Cooper et al. proposed a stretchable capacitive sensor with double-helix liquid-metal fibers, which was capable of detecting torsion, strain and touch [114]. When being twisted or elongated, the geometry of fibers as well as the capacitance between the fibers would change accordingly. Moreover, a soft inertial sensor was introduced by Varga et al. based on the alteration of capacitance between two electrodes when the liquid-metal droplet moved [66]. This inertial sensor was sensitive to motion and position detection.

Recently, wearable transient epidermal sensors were fabricated by liquid-metal hydrogel [131]. Liquid-metal marbles were distributed evenly into polyvinyl alcohol by sonication to form liquid metal hydrogel. This material was flexible and self-healing. Epidermal sensors made by it could detect some human activities, like swallowing and finger bending.

Furthermore, soft force sensors are expected to be applied to electronic skin and intelligent robots to further promote the function of artificial products. Combining the advantages of traditional artificial skin and sensors, a soft electronic skin can provide not only aesthetic function, but also a perception function [132,133].

#### *5.2. Temperature Sensors*

Several types of temperature sensors based on liquid metal have been studied. Hong et al. applied liquid metal as the interconnection between temperature-sensing units to fabricate soft temperature sensor arrays [106]. Their sensor array could be easily attached to the skin, which was important for electronic skin as well. A thermocouple based on liquid metal was proposed by Li et al. [109]. Ga with 0.25 wt. % Ga oxides and EGaIn21.5 with 0.25 wt. % Ga oxides were chosen as the thermocouple elements. The thermocouple was directly printed on paper (Figure 10D) and could be bent and twisted to some extent.

Currently, temperature sensors based on liquid metal have not been investigated deeply because compared to existing rigid metal temperature sensors, liquid metal temperature sensors are not sensitive and convenient enough. Tremendous efforts will be made in the near future.

**Figure 10.** Photos of force sensors, temperature sensors and blood glucose sensors. (**A**) The resistance of a printed pressure sensor according to external pressure, reproduced with permission from [110]. (**B**) A micro-channel pressure sensor: (**a**) Structure of the sensor; (**b**) Working principle of measuring viscosity of various fluids; Reproduced with permission from [95]. (**C**) Schematic of artificial hair cell sensor: (**a**) No external force; (**b**) Deformation under a shear force; (**c**) Deformation under a normal force. (**D**) Thermocouple based on liquid metal printed on paper, reproduced with permission from [109].

#### *5.3. Blood Glucose Sensors*

Yi et al. have designed an electrochemical sensor for wireless glucose-detecting based on liquid metal [134]. They fabricated liquid-metal electrodes by direct hand printing on polyvinyl chloride (PVC) substrate. In their research, those electrodes were modified by enzyme to detect glucose. When the enzyme reacted with glucose, the electrodes could transfer it to an electrical signal. Overall, this flexible sensor has been proven to be reliable by cycle voltammetry. They also designed a wireless system to transmit testing data to a smart phone via Bluetooth.

#### *5.4. Sensor Array*

An array is a common circuit arrangement for sensors, which can improve their sensitivity and accuracy. For instance, the temperature sensors [106] and the pressure sensor [95] mentioned above are all in the form of arrays. Specially, Zhang et al. developed a sensor array for application in robotic electronic skin [135]. They used cross-grid liquid-metal layers as electrodes. Both capacitance mode and triboelectric nanogenerators (TENG) [136] mode could be realized by this sensor array. There were two layers of cross-grid arrays constructed by liquid-metal electrodes. The capacitance between the top and bottom electrodes would increase or decrease when the environment changed. The adoption of liquid metal made the whole device highly flexible. It could function as an E-skin sensor with complementary capacitance mode or TENG mode, respectively. Compared to a single

sensing element, sensor arrays can provide large-area and multi-functional detecting, especially for human health-monitoring.

#### *5.5. Pneumatic Artificial Muscles*

Pneumatic artificial muscles have been increasingly investigated owing to the potential in wearable devices. Inspired by biological muscle, Jackson et al. designed an artificial muscle with both force and position sensors [137], as shown in Figure 11A. The Golgi tendon organ and muscle spindle were fabricated by EGaIn. Figure 11B shows the form alterations when the sensorized, flat, pneumatic artificial muscle was at rest and inflated to the pressure of 82.8 kPa. Liquid metal was injected into micro-channels to form contractile and force sensors (Figure 11C). It has been proved that this artificial muscle could mimic the behavior of biological muscle well. The progress in artificial muscle is meaningful to wearable devices, as well as soft robots.

**Figure 11.** Artificial muscles and liquid metal microsphere sensors. (**A**) Schematic comparison between biological muscle and artificial muscle, reproduced with permission from [136]. (**B**) The sensorized, flat, pneumatic artificial muscle at rest and inflated to 82.8 kPa, reproduced with permission from [137]. (**C**) Illustration of the contraction and pressure sensors, reproduced with permission from [137]. (**D**) The electrical environment of a liquid-metal microsphere in NaOH solution (above) and two methods of communicating for liquid-metal microspheres in a circuit (below), reproduced with permission from [77]. (**E**) Microscopy images showing liquid-metal microspheres produced by capillary-based microfluidics, reproduced with permission from [77]. (**F**) Schematic diagram of the photothermal conversion of liquid-metal microspheres and the near-infrared (NIR) sensor based on the liquid metal microsphere, reproduced with permission from [138].

#### *5.6. Liquid-Metal Microsphere Sensors*

The potential application of liquid-metal microspheres is increasingly attractive as well. The concept of liquid-metal enabled droplet circuits in solution was proposed previously [77]. Liquid metal droplets, ions and carbon nanotubes were combined in electrolyte solution to enhance electrical conductivity. Figure 11D shows the electric charge distribution of a liquid-metal microsphere in NaOH solution, and the communication methods between liquid-metal microspheres and carbon nanotubes in a circuit. The quantum tunneling effect is the fundamental principle lying behind liquid-metal droplet circuits. Besides the theoretical liquid-metal droplet circuits, sensors based on liquid-metal microsphere were researched recently [138]. Wei et al. designed a method to fabricate liquid-metal microspheres with different sizes through a capillary effect, as shown in Figure 11E. The size of liquid-metal microspheres could be precisely and stably controlled by the co-flowing configuration in the micro-channels. These liquid-metal microspheres had been proven to exhibit outstanding photothermal properties, which could be utilized to produce optical sensors, such as near infrared ray (NIR) sensors (Figure 11F). There is a photo-thermal conversion in liquid metal microsphere, which is promising to not only liquid-metal optical sensors, but also light-driven actuators and as an energy conversion medium. Moreover, Zhang et al. presented a flexible capacitive sensor based on liquid-metal microsphere as well [138]. It worked through the multi-plateau capacitance waveform when liquid metal sphere passed through the sensing area. It is believed that more applications of liquid-metal microspheres could be explored.

#### **6. Perspective**

In the near future, we may expect the appearance of metal-enabled liquid sensors. By contrast with traditional rigid and soft sensors, metal-enabled liquid sensors will be in the form of fluidity on the basis of liquid metal. The environment of the human body is liquid as well and all organs work in this environment. According to the ideas of bionics, research on liquid sensors may be an inevitable trend.

#### *6.1. Concept of Liquid Sensors*

As shown in Figure 12A, a liquid sensor is fabricated by wrapping liquid metal with a flexible membrane. Liquid metal inside the membrane has greater freedom than that printed on soft substrate or injected into micro-channels. Thus, the liquid sensors are more flexible and transformable. To ensure good mechanical property of liquid sensors and avoid leakage of liquid metal, the membrane must exhibit excellent flexibility and compactness. Figure 12B gives the details of a liquid-sensor system. Processing circuits are required in order to amplify, transform and filtrate the electrical signal output by the liquid sensors. Then, the processed signal could be measured by the detecting device. Compared to existing soft sensors, liquid sensors are more suitable for in vivo detection because it is highly transformable. Injectable electrodes based on liquid metal have been proposed for tumor treatment already [65]. It is possible to design injectable liquid sensors with the application of liquid metal.

**Figure 12.** Metal-enabled liquid sensors. (**A**) Illustration of liquid sensor: (**a**) Front view; (**b**) Sectional view. (**B**) Working principle of liquid sensor. (**C**) Diagram of a liquid sensor system.

#### *6.2. Working Principle of Liquid Sensors*

Metal-enabled liquid sensors can make a response to external force, electrical field, magnetic field, sound, light, heat and some chemical substances, on the basis of the unique physical and chemical properties of liquid metal (Figure 12C). External stimulations will result in the changes of resistance or capacitance, which can be converted to measurable electrical signals, such as voltage or current signal. To reflect the alteration of the environment, the liquid sensors should be connected to a processing unit, as shown in Figure 12C.

Qualitative liquid sensors based on liquid metal are not hard to realize. The difficulty is how to create a quantitative liquid sensor with high sensitivity and precision. More efforts are required to propel the development of liquid sensors.

#### **7. Discussion**

Liquid metal is characterized by good fluidity, perfect metallicity and unique electromagnetic properties. Liquid metal-based soft sensors have attracted increasing attention. Generally, soft sensors fabricated by liquid metal imply the following advantages.

Firstly, sensors fabricated by liquid metal own excellent flexibility owing to the fluidity of liquid metal. They can remain working when being twisted, stretched and bent within limits. The existence of liquid metal ensures that the circuits are still connected during deformation and improves the reliability of those soft sensors. This allows the application of liquid-metal sensors in wearable devices with great promise.

Secondly, fabrication methods of liquid-metal based sensors are diverse. Due to the unique physical and chemical properties of liquid metal, patterning methods includes printing, microfluidic technology and others. Compared to traditional micro-/nano-machining technology, these new methods are convenient and economical. However, the resolution should be further improved by optimizing the fabrication process.

Last but not least, liquid metal-based sensors provide better conformability and sensing sensitivity. Since the flexibility of liquid metal circuit is guaranteed, it can well fit different parts of human body and promote the wearing experience. Furthermore, since sensors based on liquid metal can be tightly attached to objects to be tested, the sensing sensitivity is higher than that of rigid sensors or common soft sensors.

So far, liquid metal-based sensors have been vigorously researched because of their outstanding performance. It is hoped that more soft sensors based on liquid metal, especially metamaterial biosensors, can be designed. Through increasing endeavors, more sensing mechanisms enabled from liquid metals are also possible in the near future. Microsphere sensors based on liquid metal, and even metal-enabled liquid sensors, may cause a revolution in the development of soft sensors and in vivo human signal monitoring.

#### **8. Conclusion**

This article provides a systematic review on liquid metal-enabled soft sensors as well as possible future development. Typical sensors made in this way include force sensors, temperature sensors, electrochemical sensors and so on. They work mainly depending on the alteration of resistance and capacitance of liquid metal. Electromagnetic metamaterials fabricated by liquid metal may propel the development of metamaterials and flexible metamaterial biosensors. With the excellent fluidity of liquid metal, those sensors show excellent stretchability, which have potential in wearable device and health-monitoring. Electrode arrays can be combined with soft sensors for large-area detection and sensitivity improvement.

In summary, liquid metal has shown great potential in the area of soft sensors. Based on the progress achieved, more efforts are necessary and the efforts will be worthy. Future research may focus on resolution and repeatability as well as designing the testing principles of liquid metal-based sensors. Microsphere sensors based on liquid metal and metal-enabled liquid sensors will be important research fields in the future. It is believed that the soft sensors fabricated by liquid metal could be widely applied in diverse biomedical areas and soft robots.

**Author Contributions:** Conceptualization, Y.R. and J.L.; literature review, Y.R.; writing—original draft preparation, Y.R.; writing—review and editing, Y.R., X.S. and J.L.; visualization, Y.R.; supervision, J.L.; project administration, J.L.; funding acquisition, J.L. All authors have read and agreed to the published version of the manuscript.

**Funding:** This work was funded by the NSFC Key Project under Grant No. 51890893, No.91748206, Dean's Research Funding and the Frontier Project of the Chinese Academy of Science.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Review* **AC Electrothermal E**ff**ect in Microfluidics: A Review**

#### **Alinaghi Salari 1,2,3,\*,**†**, Maryam Navi 1,2,3,**†**, Thomas Lijnse <sup>4</sup> and Colin Dalton 4,5,\***


Received: 11 October 2019; Accepted: 28 October 2019; Published: 11 November 2019

**Abstract:** The electrothermal effect has been investigated extensively in microfluidics since the 1990s and has been suggested as a promising technique for fluid manipulations in lab-on-a-chip devices. The purpose of this article is to provide a timely overview of the previous works conducted in the AC electrothermal field to provide a comprehensive reference for researchers new to this field. First, electrokinetic phenomena are briefly introduced to show where the electrothermal effect stands, comparatively, versus other mechanisms. Then, recent advances in the electrothermal field are reviewed from different aspects and categorized to provide a better insight into the current state of the literature. Results and achievements of different studies are compared, and recommendations are made to help researchers weigh their options and decide on proper configuration and parameters.

**Keywords:** electrothermal; microelectrode; microfluidics; micromixing; micropump

#### **1. Introduction**

Microfluidics is the precise control, and manipulation of fluids that are geometrically constrained to small, typically sub-millimeter, manufactured systems. Over recent decades, microfluidics has gained a great deal of attention in multiple fields, including medicine, chemistry, and biomedical engineering, due to its ability to perform multiplexing, automation, and high-throughput screening tasks. [1,2]. Due to the high surface-to-volume ratio of the fluid, and thus, the dominance of surface forces over inertial forces (i.e., low Reynolds number), fluid flow generation is a major challenge in microfluidic devices, as conventional pressure driven methods have poor efficiency in such devices [3,4]. The mechanisms of micro scale manipulation of fluids and particles can be categorized into two groups: mechanical, such as diaphragm-based devices, and non-mechanical, such as electrokinetic-based techniques. The presence of moving parts increases the risk of mechanical failure and can be incompatible with particulate flows, and thus, can limit the application of mechanical pumps for lab-on-a-chip devices [1,4,5]. Non-mechanical strategies, however, do not have these limitations. They can be integrated with microfluidic devices and also be used with particulate fluids [4,6,7]. Examples of non-mechanical methods include ultrasonic, direct current (DC) charge injection, and travelling wave driven electrohydrodynamic (EHD) micropumps [7–10].

Electrokinetics is a popular non-mechanical technique used for microfluidic fluid manipulation applications owing to its simple design and electronic automation [3]. Electrokinetic phenomena result from the interaction of an external electric field and induced electric charges. DC electrokinetics (DCEK), which has been studied over decades, requires relatively high voltages (i.e., on the order of several kilovolts) to operate which can limit its application in lab-on-a-chip devices [4,11,12]. alternating current (AC) electrokinetics (ACEK), however, which operates in low voltages (i.e., 1–20 Vrms), has led to the development of devices being portable and capable of handling biofluids without engaging in unwanted chemical reactions [13,14]. Furthermore, with non-uniform fluid flow streamlines generated by ACEK, this mechanism can be used to mix fluids [6].

AC electrokinetics mainly includes the dielectrophoresis (DEP), AC electroosmosis (ACEO), and AC electrothermal effects (ACET) [15], each of which is explained briefly in the following sections. There have been many substantive review articles on micropumps [16,17], electrohydrodynamics [18], electrokinetics [3,19,20], and their subcategories [2,21], but a comprehensive review focused on the electrothermal effect in microfluidics is still missing in the literature. This review intends to study the advances in the utilization of the electrothermal effect in microfluidics from different aspects, namely: the electric field, temperature field, and velocity field. In addition, channel properties, conditions of numerical simulations, experimental setup, and applications of ACET effect are presented. Finally, we will mention the ongoing research directions and future potential opportunities for the electrothermal effect. The majority of the publications cited throughout the manuscript that have made major contributions are also summarized in Table A1. It should be noted that the study of the electrothermal effect in other mechanisms such as ACEO is not in the scope of this paper and can be found elsewhere [5,22]. Furthermore, strategies which are based on DC electric fields (e.g., DC electrophoresis) or non-electrical (e.g., magnetic) forces are not discussed here.

#### **2. AC Electrokinetics**

#### *2.1. Dielectrophoresis*

Dielectrophoresis arises from the interaction between a dipole moment on a particle and a non-uniform electric field [23]. If the particle has a polarizability higher than the surrounding medium, the DEP force exerting on the particle will be towards regions with a high electric field (positive DEP). For particles with lower polarizability, however, this force will be towards regions with a low electric field (negative DEP). This is demonstrated by the Clausius–Mossotti factor, which specifies the direction of the DEP force with respect to the electric field [24]. In addition to the permittivities of the particle and the medium, the magnitude of DEP force is also a function of the particle size. DEP force is directly related to the third power of the particle radius, and thus, is an ideal tool for separating, concentrating, and sorting particles, cells, and viruses [25–31]. Moreover, since DEP force scales with the gradient of the electric field, it decreases with the distance from electrodes, and the generated velocity is inversely proportional to the third power of distance [13]. Therefore, DEP is not an effective technique for handling particles of relatively small size (e.g., ≤1 μm) far from a strong electric field (e.g., a few micrometers away) [32]. Two excellent reviews on dielectrophoresis are [33,34].

#### *2.2. AC Electroosmosis*

AC electroosmosis is dependent on the formation of an electric double layer (EDL) at the interface of a liquid and solid substrate [35–37]. At an interface of a solid object and electrolyte fluid, due to the adsorption of ions onto the object surface it acquires charges, and as a result, an EDL forms inside the fluid near the surface [3,38]. When an electric field is applied to this system, the charges in this layer experience an electrostatic force, which can cause fluid motion. The rest of the fluid is then dragged into motion due to viscous forces.

Since the EDL thickness is inversely related to the fluid electrical conductivity, at relatively high conductivities (e.g., 84 mS·m−1), the EDL thickness becomes very small (e.g., <1 nm), which makes ACEO ineffective for the manipulations of biological fluids (1–2 S·m−1) [1,13,15,39–41]. In addition, ACEO is frequency-dependent, and increasing the actuation frequency beyond 100 kHz causes the

ACEO effect to become invisible, since at high frequencies, the electric double layer is unable to form, and no fluid flow is generated [35]. Similarly, at very low frequencies, the double layer can completely screen the electric field, and thus, no net flow can be generated. ACEO has been developed and used in many forms to pump fluids or manipulate particles, namely biased ACEO for particle assembly and micropumping [32,42], micropumping of fluids [43,44], Travelling wave ACEO [45] and asymmetric ACEO micropumping [40], and DEP electrohydrodynamic particle trapping (i.e., ACEO in conjunction with DEP) [46,47]. As ACEO is only effective in relatively low frequencies, it is more prone to bubble generation and electrode deterioration resulting from electrochemical reactions, which can affect the electric field distribution and eventually damage the device [39]. Despite these limitations, there are many application-driven papers in the literature using ACEO [48–51], where modifications have been suggested to enhance ACEO applicability in fluids with electric conductivities up to 0.1 S·m−<sup>1</sup> under the actuation of high frequencies and voltages. These modifications include the utilization of polarizable walls in induced-charge electroosmosis [52], AC faradic polarization [50], and nonlinear electroosmosis on curved surfaces [53], to name a few. An advantage of ACET is that it can be used for higher conductivities, i.e., over 1 S·m<sup>−</sup>1. More details of different strategies for implementation of ACEO in microfluidics can be found in [54,55]. DEP and ACET effects can be combined [13,56–59] to improve particle manipulations, as DEP has difficulty in manipulating submicron particles, and also is weak in areas far from electrodes where ACET is strong [13,56].

#### *2.3. AC Electrothermal*

Unlike ACEO and DEP, ACET has been shown to be very effective in biomedical applications which involve high conductivity biofluids, such as blood, urine, and saliva [60]. This is due to the fact that the ACET effect originates from a temperature gradient in the bulk of the fluid and not the fluid-electrode interface (i.e., the EDL). Fluids with higher conductivities can generate stronger microflows, and therefore, it is the most efficient electrokinetic mechanism for manipulating biological fluids with conductivities above 0.7 S·m−<sup>1</sup> [1,4,13,14,39,60–63].

Emerging in 1960s [64,65], ACET has been widely used for fluid manipulations over the years [7,66], and is also referred to as induction EHD [67,68]. Despite similar flow patterns, the physics behind ACET and ACEO are different. ACEO is the result of the interaction of an electric double layer at the interface of the fluid-electrode and a non-uniform AC electric field, while ACET arises from the interaction of a temperature gradient in the bulk of the fluid and a non-uniform AC electric field. The source of the temperature gradient may be internal (i.e., Joule heating) or external (e.g., strong illumination, microheaters, etc.). Temperature gradients in the fluid lead to gradients in the electrical properties of the fluid, i.e., conductivity and permittivity, which induce free charge density. An electric force arising from the non-uniform electric field causes the free charges to move. As a result of shear stress, the surrounding fluid is also dragged into motion which produces microflows. Unlike ACEO, the ACET effect shows plateaus in force in a wide frequency range (10–10<sup>11</sup> Hz) [39,69]. With ACET, the fluid velocity is steadier and more predictable at different frequencies compared to ACEO and DEP. In general, ACET flow can be generated in frequencies above 100 kHz and salt concentrations of above 10−<sup>2</sup> mol·dm−3, whereas ACEO is more common at low frequencies and salt concentrations of 10−<sup>2</sup> mol·dm−<sup>3</sup> and below [62]. Despite the fact that fluid heating is crucial for the ACET effect, ambient heat conduction helps dissipating energy so that the temperature rise in the bulk of the fluid is typically low (Δ*T* < 5 K), which is safe for biofluids [11,13,14,41]. The ACET force is proportional to the temperature gradient |∇*T*| and not the temperature rise [1].

In order to generate AC electric fields required for inducing the electrothermal effect, microfabricated electrode arrays are commonly used. Employing a symmetric pair of electrodes at the bottom of a microfluidic channel can induce two symmetric sets of microvortices above the electrodes, and thus, no net flow can be generated [70]. For pumping applications, however, the electrode symmetry needs to be broken. Since the electrothermal force is a function of the electric field and temperature gradient, asymmetry may be achieved by manipulating either or both of these factors. This will

be discussed in more details in the following sections. Typically, due to its simple implementation, imposing geometry asymmetry to microelectrodes is the most common approach for breaking the symmetry of microvortices. In addition, manipulating the temperature field with the help of external heat sources, such as strong illumination [69,71–73], embedded microheaters [74,75], and heating electrodes [1], can also be used for creating a net flow. Although a common ACET microdevice implements an array of electrode pairs placed at the bottom of a microchannel with a rectangular cross section, more complicated configurations with electrode arrays placed on the top, bottom, and sidewalls of channels with different cross sections have also been studied [69,76–78]. Studies with the use of grooves on the channel surface to induce further asymmetry and increase flow have also been addressed, but fabrication of these designs suffers from serious challenges.

Similar to other electrokinetic mechanisms, ACET suffers from some drawbacks, most of which have been addressed to some extent in the literature, as will be shown in this paper. In microfluidic devices, miniaturization can be hindered as the ACET effect originates from the bulk of the fluid and decreasing the channel dimensions can decrease the volume of the fluid flowing inside the channels [3,35]. In addition, ACET depends on the formation of temperature gradients, and therefore, cannot be used with low conductivity fluids. As such, its application in conjunction with DEP, which requires low conductivity fluids for efficient particle sorting, is limited [1,4,5]. Importantly, an excessive temperature rise in fluids with high conductivities can cause the buoyancy force to dominate over the ACET force [4]. The reason is that the ratio of electrothermal force to buoyancy force is proportional to |∇*T*|/Δ*T*. Thus, when Δ*T* > |∇*T*|, the buoyancy force becomes the dominant force. Finally, increasing the actuation voltage to achieve high fluid velocity can lead to electrochemical reactions which can limit the application of ACET effect on biofluids [1]. There have been some reports on how to mitigate this issue [79].

#### **3. Theory**

As stated in the previous section, the ACET effect results from the interaction of a non-uniform electric field and a temperature gradient in the bulk of the fluid. The energy balance equation governs the amount of Joule heating as follows [15]:

$$k\mathbf{V}^2T + \frac{1}{2} < \sigma|\mathbf{E}|^2 > = 0\tag{1}$$

where, *k* and σ are the thermal and electrical conductivities of the fluid, respectively, and *E* is the electric field, which can be obtained from the Laplace equation in a homogeneous medium as below:

$$\mathbf{V}^2 V = 0\tag{2}$$

where, *E* = −∇*V*, and *V* represents the electric voltage.

An order of magnitude estimation of Equation (1) gives [15]:

$$
\Delta T \approx \frac{\sigma V^2}{k} \tag{3}
$$

Based on Equation (3), the temperature rise is directly proportional to the fluid electrical conductivity and actuation voltage squared, which, in most applications, is the control parameter.

The ratio of heat convection to heat conduction in a microchannel is very low (i.e., Peclet ≤ 0.07) [15,67,80]. Furthermore, it has been numerically shown that, compared to electrical forces, natural convection in micro-channels can be neglected [15,80]. However, for cases with high thermal Peclet numbers, heat convection cannot be neglected [81]. The temperature gradient in the fluid can

change the fluid properties, including permittivity ε and conductivity σ, and can be calculated as follows [15]:

$$
\nabla \varepsilon = \left(\frac{\partial \varepsilon}{\partial T}\right) \nabla T \tag{4}
$$

$$
\nabla \sigma = \left(\frac{\partial \sigma}{\partial T}\right) \nabla T \tag{5}
$$

In most ACET applications, it is assumed that the rate of change of permittivity and conductivity with the change in temperature is very small [15]. Otherwise, a temperature coefficient needs to be defined to account for the changes in fluid properties [81,82]. As a result of this assumption, the perturbed electric field can be neglected, and the charge convection can be assumed to be much smaller than the charge conduction [15].

The change in electrical properties of the fluid leads to the generation of electrical charge density as follows [15]:

$$
\rho\_{\mathbb{E}} = \nabla \cdot (\varepsilon \mathbb{E}) \tag{6}
$$

$$\frac{\partial \rho\_E}{\partial t} + \nabla \cdot (\sigma \mathbf{E}) = 0 \tag{7}$$

where, ρ*<sup>E</sup>* is the charge density.

Under the effect of the electric field, there is a force applied to the charge density which is [15]:

$$f\_{\mathbf{E}} = \rho\_{\mathbf{E}} \mathbf{E} - \frac{1}{2} \mathbf{E}^2 \mathbf{V} \varepsilon \tag{8}$$

The first term in Equation (8) is the Coulomb force and the second term is the dielectric force.

As charges move in the electric field, they drag the surrounding medium into motion. Therefore, microflows are generated in the fluid and are governed by:

$$
\nabla p + \eta |\nabla|^2 u + f\_E = 0 \tag{9}
$$

where, η, *p*, and *u* are the dynamic viscosity, pressure, and velocity field, respectively. Furthermore, from the conservation of mass for an incompressible fluid, we have:

$$
\nabla \cdot \mathbf{u} = 0 \tag{10}
$$

With an order of magnitude estimation from Equation (9), the flow velocity can be written as |*u*| ≈< *fE* > · *l* 2 <sup>η</sup> , where *l* is the characteristic length of device, which is usually the electrode spacing [4,15,60].

Charge density can be calculated by combining Equations (6) and (7) as follows [83]:

$$\rho\_E = \frac{\sigma \varepsilon}{\sigma + i\alpha\varepsilon} (\alpha - \beta) (\nabla T \cdot E) \tag{11}$$

where, ω is the angular frequency of the AC electric field, and:

$$a = \frac{1}{\varepsilon} \left(\frac{\partial \varepsilon}{\partial T}\right) \tag{12}$$

$$
\beta = \frac{1}{\sigma} \left( \frac{\partial \sigma}{\partial T} \right) \tag{13}
$$

For aqueous solutions and temperatures around 293 K, <sup>α</sup> and <sup>β</sup> can be estimated as <sup>−</sup>0.4% K−<sup>1</sup> and 2% K<sup>−</sup>1, respectively [84].

With the above approximations, the electrothermal force can be simplified as below [15]:

$$ = \frac{1}{2} \frac{\varepsilon \left(\alpha - \beta\right)}{1 + \left(\omega \tau\right)^2} (\nabla T \cdot E)E - \frac{1}{4} \varepsilon \alpha |E|^2 \nabla T} \tag{14}$$

where, τ = <sup>ε</sup> <sup>σ</sup> is the charge relaxation time of the liquid and is in the range of 0.7–35 ns for conductivities in the range of 0.02–1 S·m−<sup>1</sup> [41,85]. As stated above, the first term represents the Coulomb force, and the second term is the dielectric force. These forces act in different frequency ranges (i.e., the Coulomb force dominates at low frequencies and dielectric force dominates at high frequencies) and are in different directions [83]. Near a certain frequency, known as the cross-over frequency *fc*, the two forces compete, and flow reversal can occur as a result of switching from a Coulomb force dominant to a dielectric force dominant regime or vice versa [15]. The cross-over frequency can be calculated as below [15]:

$$f\_c \approx \frac{1}{2\pi\pi} \left( 2 \frac{\left| \frac{1}{\mathcal{F}} \left( \frac{\partial \mathcal{G}}{\partial T} \right) \right|}{\left| \frac{1}{\mathcal{E}} \left( \frac{\partial \mathcal{G}}{\partial T} \right) \right|} \right)^{\frac{1}{2}} \tag{15}$$

For example, the cross over frequency for a biofluid with a conductivity of 1 S·m−<sup>1</sup> is roughly 200 MHz [61]. ACET force, and thus ACET flow velocity, is higher (up to 11 times) at low frequencies and has no dependence on frequency except for the transition region (i.e., near crossover frequency) [13,41,69].

By taking a closer look at Equation (11), as α, β, and ω are constants, we can conclude that ρ*<sup>E</sup>* ∝ ∇*T*·*E* [1]. Commonly, in electrokinetics, frequencies much lower than 10 MHz (usually around 200 kHz) are used, where 1 + (ωτ) <sup>2</sup> <sup>≈</sup> 1 and dielectric force is negligible (i.e., the Coulomb force is ~11 times larger than the dielectric force) [41]. At these frequencies, there is not enough time for the double layer to form, and thus, the dielectric force is neglected [13]. As a result, the flow direction is determined by the Coulomb force and Equation (14) is reduced to the first term on the right hand side. With a similar argument, as α, β, and ωτ are constants, we can conclude that |< *FE* >| ∝ |*E*| <sup>2</sup>|∇*T*|, and according to Equation (1), when Joule heating is implemented, <sup>∇</sup>*<sup>T</sup>* <sup>∝</sup> *<sup>E</sup>*2, and therefore, |< *FE* >| ∝ |*E*| <sup>4</sup> [69]. This means that electrothermal force is proportional to the fourth power of the electric field. According to the order of magnitude estimation of velocity obtained from Equation (9), ACET fluid velocity also has a quartic relationship with the actuation voltage. This means that a small increase in the electric field can cause a large increase in electrothermal force, and thus, a significant increase in the resultant flow velocity. As stated above, since electrothermal and electroosmotic flows have similar patterns, one way to distinguish them is to use this proportionality. Since electroosmotic velocity is proportional to the square of voltage, by plotting velocity against applied voltage, the source of microflows can be revealed [14,15].

The theory discussed here is based on an uncoupled model developed by Ramos et al. in 1999 [15]. In such a model, an assumption of a small temperature rise Δ*T* < 5 K renders the change of fluid properties and electric field with temperature insignificant. However, if the temperature rise is considerably higher, then the fluid properties will change by temperature variations, and therefore, a fully coupled model needs to be used [74,86–88]. Hong et al. developed a coupled model, compared it with the classical model, and found that only at small temperature rises do the results of the two models match [87]. More recently, pumping and mixing of non-Newtonian fluids have been studied [89–91].

#### **4. Electric Field**

In lab-on-a-chip applications, the maximum limit for actuation voltage is typically within the range of 5–7 Vrms, beyond which electrochemical reactions can harm the nature of biofluids. Since, as stated in the previous section, electrothermal force, and thus, fluid velocity is highly dependent upon the applied voltage (i.e., quartic dependence), the increase in velocity of electrothermal based biomedical microdevices is hindered accordingly. To overcome this issue, many studies have proposed and developed new ACET designs with modifications to the electric field, which include either changes to the geometry of the electrode array or introducing asymmetry to the electric potential, using techniques such as travelling wave, multiphase, DC biased, etc. In this section, different strategies proposed for modifying the electric field are reviewed.

#### *4.1. Introducing Asymmetry to Geometry*

To produce a strong electric field, microelectrodes are typically patterned on the inner surfaces of microchannels [11,66]. Due to changes to the electric field and temperature gradient strengths, by manipulating the electrode configuration, a wide variety of flow patterns and velocities can be achieved [41]. Microelectrode arrays with different shapes have been reported, which can be categorized into two-dimensional (2D) and three-dimensional (3D) geometries [41,85].

Due to the simplicity and ease of fabrication, 2D electrode geometries are the most common and can be further categorized into asymmetric rectangular [4,41,69,92], orthogonal [4], meandering [62], concentric [60], and triangular [56], with asymmetric rectangular being investigated the most in the literature (Figure 1a–e). Imposing asymmetry on the electrode pairs induces asymmetry in the resultant microflows, and therefore renders pumping action. Due to this feature of microelectrode configuration, more heat is hypothesized to be generated near the narrow electrode [1]. In general, in such a configuration, the width of the narrow electrode and the gap between electrodes in a pair are the two major design parameters which govern the magnitude of the fluid velocity [41].

**Figure 1.** Schematic of different electrode geometries. (**a**–**e**) 2D Electrodes: (**a**) asymmetric rectangular, (**b**) orthogonal, (**c**) meandering, (**d**) concentric, and (**e**) triangular. (**f**–**i**) 3D Electrodes: (**f**) 3D electrode array, (**g**) microgrooved configuration, (**h**) electrodes facing each other, and (**i**) spiral design. Reproduced with permission from [56,60,62,69,77,85,93–95].

Yuan et al. conducted a thorough study on the optimization of 2D rectangular electrode arrays and reported a set of ratios for the corresponding parameters [41]. 2D rectangular asymmetric arrays are used mostly for pumping applications, either alone or with other changes to electric or thermal fields [1,74]. In pumping applications, one or multiple arrays of asymmetric microelectrode pairs are placed perpendicular to the channel length in order to generate a net flow along the channel. In applications, where a lateral fluid mixing is also desirable, however, the electrode array needs to be placed at an angle <90◦ to the channel length [77,95]. In 2D electrode configurations, by decreasing the gap between electrodes up to a certain point, the resultant fluid velocity can be increased as the strength of the electric field increases. However, if the strength of the electric field is maintained at a constant, increasing the gap between electrodes can enhance the resultant fluid flow, as increasing the gap allows a larger volume of fluid bulk above the gap to experience the strong electric field [41]. One way to enhance the ACET effect is to decrease the width of the narrow electrode, increasing the non-uniformity of the electric field. However, if the narrow electrode is made too small, the strength of the electric field can decrease accordingly, and, as electrothermal force is proportional to the fourth power of the electric field, electrothermal force and, hence, velocity can decrease drastically [41]. As shown, the dimensions and electrode geometry are of great importance to the electrothermal effect. In addition, the channel size and cross section can also significantly affect the resultant flow rate [96]. More on this topic can be found in Section 9. Orthogonal electrode arrays were proposed by Wu et al. for pumping applications with velocities exceeding 1 mm·s−<sup>1</sup> [4,93]. A meandering electrode array, with a sinusoidal electrode gap, was proposed for rapid mixing of two biofluids [62]. Rapid mixing of high-conductivity (up to 22 S·m<sup>−</sup>1) fluids has been shown using concentric electrode designs [60]. A triangular electrode pattern has also been suggested for micromixing [56].

The backward flow generated above the narrow electrode can negatively impact the use of planar configurations of electrodes for pumping applications, as they cause the average flow in the microchannel to slow down [85]. In 2000, Ajdari, suggested a 3D electrode configuration to circumvent this problem [94] (Figure 1f). Expanding on the same idea, Du and Manoochehri suggested a microgrooved channel (instead of 3D electrodes) to impose spatial asymmetry to the electrode pairs [97] (Figure 1g). More recently, this effect has been numerically simulated with a number of substrate configurations [98]. Manufacturing a channel with modified roughness of substrate surface is easier than manufacturing 3D electrodes. In their work, different shapes of grooves are proposed and experimentally studied. This study showed that by using the microgrooved structure, based on the shape of the grooves, the pumping capacity increases by up to six-fold compared to a planar configuration. In another work, they reported an optimized configuration of such a design [85]. They managed to increase the net flow rate by further suppressing backflows and shortening the streamlines [85]. However, fabrication of such microgrooved structures is more complicated than conventional microchannels with planar electrode design [69]. A variation of such configuration in conjunction with opposing electrodes on the top surface is suggested for mixing applications [89,99]. A similar configuration is also reported where opposing castellated electrodes are utilized to eliminate the ACET vortex and thereby enhance pumping [100].

A two-layer microelectrode configuration, where microelectrodes are facing each other, has been proposed for particle trapping in order to investigate the DEP effect in conjunction with ACET [13]. In this electrode configuration, one electrode is placed on top and the other at the bottom of the channel. A similar configuration is also used for studying Joule heating effects [80], as well as for patterning of colloids when equipped with underlying microheaters [75]. Such an opposing microelectrode configuration with different patterns is also reported for mixing fluids [77,89,101] (Figure 1h). This configuration is also used in rapid electrokinetic patterning, where electrothermal effects play a significant role in particle manipulation [102,103]

More recently, studies have been carried out on utilizing electrodes on all walls of the microchannel in different patterns [104,105]. Using a particular configuration of multiple electrode arrays on side walls, simultaneous mixing and pumping of biofluids in one microchannel during a short time and over a short distance is feasible [104]. In a very recent publication, a spiral electrode pattern has been proposed for simultaneous mixing and pumping which is capable of achieving a velocity of 400 <sup>μ</sup>m·s−<sup>1</sup> for a biofluid with a conductivity of 0.224 S·m−<sup>1</sup> (Figure 1i) [95].

#### *4.2. Introducing Asymmetry in Electric Potential*

In addition to an asymmetric electrode geometry, the electric field can also be modified to enhance the electrothermal effect. Applying a DC potential to the electrode pairs in addition to the existing AC potential, using a multiphase AC signal in the form of a travelling wave, and utilizing a two-phase actuation system are examples of such modifications, each of which is briefly described here.

#### 4.2.1. DC Biased

In this method, symmetric electrodes are used but one electrode is always at a positive potential and the other is always at a negative potential subject to faradaic charging and capacitive charging, respectively (Figure 2a) [6]. Hence, positive charges on both electrodes lead to unidirectional flow. The advantage of this configuration is that pumping action can be achieved by a symmetric pair of electrodes.

**Figure 2.** Schematic of different mechanisms for imposing asymmetry to the electric field in alternating current electrothermal effects (ACET) devices. (**a**) Direct current (DC) biased configuration. The left electrode is always at a positive potential and subject to faradaic charging, while the right electrode is always at a negative potential and subject to capacitive charging. (**b**) Travelling wave configuration. The induced temperature gradient generates a charge density which is moved with the travelling electric field. Note that the direction of propagation of travelling wave is opposite to the direction of fluid flow. (**c**) Two-phase asymmetric configuration. The narrow electrodes in two adjacent pairs have different polarities. The superposition of the two configurations enhances the electric field in region B. Reproduced with permission from [6,68,69].

#### 4.2.2. Travelling Wave (TW)

For the first time in 1966, Melcher observed pumping of fluids with very small conductivities, by imposing a temperature gradient along the depth of the channel in conjunction with travelling wave induction [64]. In 1967, Melcher and Firebaugh further investigated the phenomenon and presented the governing equations [65].

Fuhr et al. studied a travelling wave induced micropump and modified the principles presented by Melcher and Firebaugh [7,65]. Later, in 1994, the same group explained the mechanisms in which temperature gradients (both externally applied and internally generated) are used in conjunction with a travelling wave [66]. With an induced temperature gradient in the fluid bulk, free charges can be generated, and the travelling wave can move them along the channel above the microelectrodes causing fluid flow. This principle is further illustrated in Figure 2b. The mechanism of applying a temperature gradient in this design can be external (i.e., bed heating) or self-generated (i.e., Joule heating).

In 1992, Fuhr et al. experimentally investigated a high frequency travelling wave induced micropump with self-generated temperature gradient, i.e., higher temperatures at the bottom of the channel in the space between electrodes and lower temperatures at the top of the channel [7]. In this case, the resultant fluid flow is in the opposite direction of the travelling wave, as also verified by Liu et al., [68]. Their measurements of flow velocity show a quadratic relationship with voltage caused by applying an external temperature gradient [13,106]. Following the study of Fuhr et al., several numerical investigations of travelling wave induced electrothermal flow were also carried out [68,107].

#### 4.2.3. Two-Phase Actuation

In 2011, Zhang et al. proposed a two-phase planar asymmetric electrode design which yielded an increase of 25–50% in flow rate compared to conventional single-phase asymmetric configurations [69]. In this design, an AC signal of 0◦/180◦ was applied to the narrow electrode in an asymmetric electrode pair (Figure 2c). This led to an enhanced electric field. As the temperature gradient has a quadratic relationship with the electric field, temperature, and therefore, flow rate increased accordingly.

#### **5. Temperature Field**

#### *Internal and External Heating*

In an ACET device, heating of the fluid is the source of thermal gradients which can eventually lead to electrothermal microflows. Heating can be either internal (i.e., Joule heating) or external.

Internal heating refers to Joule heating of the fluid upon activation of the electric field. In a rectangular asymmetric electrode configuration, the temperature rise above the narrow electrode is higher than that above the wide electrode. Joule heating scales with conductivity of the fluid and applied voltage (Equation (1)). As a result, fluids with low conductivity need a higher electric field to produce sufficient thermal gradients. However, at high applied voltages, electrochemical reactions can occur, which can damage the fluid and electrodes. This issue can be resolved by applying an external heat source and keeping the applied voltage low.

A few methods have been proposed for external heating, including strong illumination [69,71,73], integrated heating elements (ohmic heating element) [39], heating of the electrodes [1], and thin film resistive heaters [74], the latter of which is shown to be portable and more efficient [74,75].

Incorporation of external heating allows the electrothermal flow to be controlled independently of the electric field strength [39,74]. Therefore, the effect of fluid properties on the resultant flow is minimized [39], and manipulation of low conductivity fluids with the electrothermal effect becomes feasible [74]. In addition, by means of external heating, the electrothermal flow can be implemented in conjunction with other electrokinetic methodologies, which work under a uniform electric field, such as electrophoresis and electroosmosis [74,75]. External heating strategies also enable control of the direction of fluid flow [1,71,106].

Green et al. performed a thorough study on strong illumination as an external heat source [71,106]. The resultant velocity field in the case of external heating can be obtained from the equation below [13,106]:

$$|\mu| \approx 3 \times 10^{-3} \left(\frac{\varepsilon V^2}{\eta}\right) \left|\frac{\partial T}{\partial y}\right| \left(\frac{1}{\sigma}\right) \tag{16}$$

where, <sup>∂</sup>*<sup>T</sup>* <sup>∂</sup>*<sup>y</sup>* is the external thermal gradient along axis *y*. Unlike Joule heating, here the velocity has a quadratic relationship with voltage, which was verified experimentally by Stubbe et al., where ohmic heating elements were used [39].

Depending on the amount of heat flux introduced into the fluid bulk by the external heating, the effect of voltage on the flow velocity changes. In 2013, Yuan et al. found that if the heat flux is

small (around 104 <sup>W</sup>·m<sup>−</sup>2), it can be counteracted by the Joule heating of the fluid, and the relationship between fluid velocity and voltage can be between quartic and quadratic [1]. However, if the heat flux is relatively large and the fluid is mostly under the influence of external heating, the fluid velocity will have a quadratic dependence on the voltage.

Furthermore, thin film microheaters can be embedded in the substrate below the electrodes, separated by an insulating layer [75]. Such a heating mechanism can yield a thermal efficiency of close to 100% [74]. Velasco and Williams showed the application of thin film heaters for assembly of colloids [75]. The geometry of the assembly is governed by the geometry of the array of microheaters. Therefore, particles can be trapped in a larger area when compared to conventional particle trapping techniques [13]. The assembly of 1 μm and 2 μm particles was achieved in the frequency range of 1–200 kHz, while no particle aggregation was observed at other frequencies. Since the temperature gradient is the underlying mechanism, microheaters cannot be patterned close to each other. Furthermore, with thin film microheaters, sharper temperature gradients exist near the electrodes which result in a larger value of <sup>∇</sup>*T*·*E*<sup>2</sup> [74]. Compared to Joule heating, thin film microheaters alone require only 40% of the power to produce the same flow rate. Without a change in power, the flow rate can be increased to 250% of that of solely Joule heating [74]. Both the proximity of thin film resistive heaters to each other and their distance to the regions of maximum electric field strength determine the resultant flow regime [74,75]. The maxima of temperature gradient and electric field need to occur in the same region in the bulk of the fluid in order to maximize the product of <sup>∇</sup>*T*·*E*<sup>2</sup> [74]. More recently, Williams and Green applied the same idea to a symmetric pair of electrodes, which is more desirable for DEP applications, and carried out numerical simulations towards finding an optimum location for the heater with respect to the electrodes [108].

In spite of a fundamental factor for generation of electrothermal flow, Joule heating can be an unwanted effect in other electrokinetic devices [80]. For example, in a DEP based particle manipulation device, Joule heating can harm the biological fluid and form electrothermal microflows taking particles away from their intended spot of sorting or trapping [80,81]. For this reason, studies have been conducted to gain a better insight into Joule heating [80] and its effects on insulator-based DEP devices [81,109].

Almost all electrothermal designs studied in the literature are checked for excess temperature rise due to Joule heating, as this can negatively impact their potential application with biofluids. For example, Du and Manoochehri compared the Joule heating in their proposed microgrooved and planar configurations at a wide range of conductivities [85] and found no major difference between the two structures. In devices with thin film microheaters, temperature rise is shown to be half of that of Joule heating, which further corroborates the high efficiency of these devices [74].

Similar to Joule heating, external heating may not always lead to an enhanced electrothermal effect. Zhang et al. studied the effect of strong illumination coupled with their two-phase planar electrode configuration [69]. Interestingly, they observed a decrease in velocity when a strong illumination was applied. This observation was justified by assuming that the flow direction generated by the illumination is opposite to that generated by Joule heating [83]. As mentioned earlier, for external heating to be effective, heat flux introduced into the system must be significantly higher than the Joule heating generated by the system. Not clearly stated by Williams [74], however, is that their reported results show that by increasing the conductivity, the ratio of electrothermal force generated by thin film heaters to that generated by Joule heating gradually decreases. This indicates that the coupled electric and temperature fields need to be solved for higher temperature rises (>5 K) [82,86].

ACET flow velocity has a strong dependence on the applied voltage (*<sup>u</sup>* <sup>∝</sup> *<sup>V</sup>*4, when no external heating is applied). However, voltage can be increased up to a certain point, usually below 7 Vrms, to avoid thermal damage to the biofluid [1,11,14,41,60,71,75]. To solve this problem, Yuan et al. introduced a thermally biased configuration, where one electrode in a pair is at a higher temperature than the other one [1]. In this way, the temperature gradient in the fluid can be controlled independently from the voltage.

At frequencies below 10 MHz, where the Coulomb force is the dominant force, the electrothermal force can be simplified to *FE* = ρ*EE*. Since, due to the limitations discussed above, there is an upper limit for the applied electric field, the electrothermal force can be increased further by increasing the charge density, ρ*E*. As ρ*<sup>E</sup>* ∝ ∇*T*·*E*, Yuan et al. attempted to increase the electrothermal force by increasing the temperature gradient by means other than Joule heating [1]. In their design, unidirectional flow was obtained with symmetric electrodes, making it possible for their configuration to be used either as a micromixer or a micropump depending on the applied voltage, frequency, and heat flux. When imposing external heat flux to the narrow electrode in an asymmetric pair, they obtained a velocity 5.7 times higher than that of a regular asymmetric ACET micropump. Their simulation results show that by applying the heat flux to the wide electrode, the direction of the flow above the electrodes can be reversed. This finding corroborates their hypothesis that the generated heat on the narrow electrode is the primary cause of pumping in a conventional ACET device, and that external heating has a strong influence on the pumping performance.

#### **6. Fluid Flow Regime**

#### *6.1. Flow Velocity*

Almost all the work that has been performed dealing with the electrothermal effect is an attempt to increase the ACET velocity in high conductivity fluids while avoiding significant increase in temperature and voltage. Many studies have attempted to increase the electrothermal flow velocity by focusing on the governing parameters mentioned in Equation (14) and have proposed new designs to increase the velocity based on their interpretations of the formula. The maximum of the flow velocity reported in these studies differs both in its magnitude and location (i.e., height/distance from channel bottom/electrodes).

Since electrothermal velocity has a quadratic relationship with the temperature gradient, many studies have focused on increasing this parameter of electrothermal force to increase the flow velocity. External heating has been proposed and thoroughly investigated for this purpose, which is discussed in Section 5. By simplifying the equation of electrothermal force to |< *FE* >| = ξ(ω)|*E*| <sup>2</sup>|∇*T*|, Zhang et al. concluded that by increasing both the strength of electric field and the temperature gradient, a significant increase in electrothermal force and thus fluid velocity could be obtained [69]. Taking advantage of this combination, by developing a two-phase asymmetric planar electrode design in which a stronger electric field is obtained, they demonstrated flow velocities reaching 25–50% higher than those of conventional single-phase configurations. In this simplified equation, ξ(ω) is a function of frequency and the angle between the vectors of applied field and temperature gradient. They also found that fluid velocity is much higher at low frequencies, which was also verified by Williams who determined that pumping rates at high frequencies (>*fc*) drop to 10% of that at low frequencies [74].

To accurately measure the electrothermal velocity field, usually micro particle image velocimetry (micro-PIV) is used. The assumption of this method is that particles follow the fluid flow and are not under the influence of any other forces [110–112]. Therefore, ACET velocity needs to be measured at a height with a negligible DEP effect on particles [1]. At a distance close to electrode surface, the DEP effect on particles is significantly high, which can cause them to be trapped at the electrode edges [1]. As a result, depending on the channel height and size of tracer particles, ACET velocity is typically measured at a height of 10–50 μm above the electrode surface. A good control for this effect is carried out by Wu et al., where, they found, by using particles with a size of ~500 nm, an order of magnitude estimation yields that at ~5 Vrms particles move with a DEP-induced flow velocity of 0.11 <sup>μ</sup>m·s−<sup>1</sup> at the height of 10 <sup>μ</sup>m above the electrodes [4]. Since ACET velocity is often higher than 100 <sup>μ</sup>m·s-1, DEP velocity is negligible at this height. The highest electrothermal velocities reported are taken at a height of 20–50 μm above the electrodes [1,13,92]. The lowest electrothermal velocity occurs very close to the electrodes (~5 μm above electrodes in a microchannel with a height of 200 μm) [92]. These observations are in agreement with the physics of the electrothermal effect as it is created in the bulk of the fluid.

Increasing the number of microelectrode pairs has shown to result in increasing ACET fluid flow of high conductivity biofluids at voltages above 4 Vrms [92]. In addition, flow velocities measured experimentally are typically smaller than those predicted numerically, sometimes by orders of magnitude, which can be justified to be related to the experimental conditions. Here, we discuss some of the experimental issues reported in the literature.

In their experimental study, Sigurdson et al. showed an electrothermal flow velocity of 100 <sup>μ</sup>m·s<sup>−</sup>1, which was reported at a lower voltage in a numerical study [14]. Since their simulations are performed for a 2D geometry, they attributed this discrepancy to neglecting the out of plane heat transfer, which is significant in devices with a small channel height (i.e., ~200 μm). In their later work, since they encountered the same issue (i.e., observing an experimental velocity of 1.5 orders of magnitude lower than the numerical result), they defined an effective voltage [61]. For this matter, a reduction coefficient of 0.38 was introduced to correct the actuation voltage in the numerical simulations. With the modified voltage, the velocity–voltage curve of numerical study closely matches the corresponding curve of the experimental study.

Studying the Joule heating effect, Williams et al. [80] defined a coefficient, *Erel*, for medium conductivity, to account for the loss in applied electric field, which they attribute to the electrical resistance of their electrode material (i.e., indium tin oxide (ITO)). Therefore, by using a more conductive material for electrodes (e.g., gold), there will be less potential loss (i.e., larger value of *Erel*). Sin et al. showed a significant deviation of experimentally measured fluid velocity from the theoretically predicted one for fluids with conductivities on the order of 22 S·m−<sup>1</sup> [60]. The reason for this discrepancy was assumed to be the deterioration of the electrode surface due to electrochemical reactions at high conductivities, which is not accounted for in the numerical model. The same argument is also reported as the reason for deviation of experimental data from simulation in other studies [76,105]. With the surface of the electrode being altered, the electric field, flow patterns, and velocities are no longer predictable as the actual electric potential can be significantly decreased [76]. In addition to high conductivity, increasing voltage above 5.5 Vrms is also reported to cause electrode deterioration [76]. Evaporation, and thus a change of the medium's properties, is also mentioned as another possible factor for this discrepancy. In addition, the buoyancy effect can play a role in causing this discrepancy, since a high conductivity medium can experience high temperatures. The ratio of the electrothermal force *FE* to buoyancy force *FB* can be approximated as the following [4]:

$$
\left| \frac{F\_E}{F\_B} \right| = 7.95 \times 10^{-12} \left( \frac{\nabla T}{\Delta T} \right) E\_{rms}^2 \tag{17}
$$

As suggested by this approximation, for buoyancy to be negligible, temperature gradient ∇*T* needs to be much higher than temperature rise Δ*T*. As a result, for enhancing the ACET performance, device elements (e.g., electrodes) with relatively high thermal conductivities need to be implemented [4].

#### *6.2. Direction of AC Electrothermal (ACET) Flow*

In micropumps with planar electrode configurations, two microvortices on both electrodes in every pair are in competition to determine the direction of net flow [85]. In conventional rectangular asymmetric configurations, since the fluid flow spends more time on the wider electrode, the direction of ACET net flow is determined by the microvortices on the wider electrode [4].

It has been experimentally shown that the direction of flow can be controlled by external heating, i.e., switching the heat flux between wide and narrow electrodes [1,71,106]. This is true for both rectangular symmetric and asymmetric electrode configurations. Additionally, some studies have been performed on the use of unique device configurations to easily toggle the direction of net flow and introduce mixing [113,114], however, these models have not yet been rigorously validated.

#### *6.3. Flow Reversal*

One major advantage of ACET is that its velocity is generally independent from frequency, however, at around the crossover frequency, flow reversal occurs, and thus, lower velocities (~10–20% of velocity of low frequencies) are generated [39,69,74,115].

Flow reversal has also been observed in devices with external heating. In 2014, Liu et al. numerically investigated applying a temperature gradient along the channel length on rectangular symmetric electrode arrays and compared it to the conventional asymmetric array [68]. They observed unidirectional pumping in the direction from higher to lower temperatures at 100 kHz (i.e., from the narrow electrode to wide electrode). They also observed that by increasing the frequency to 500 kHz, the direction of flow in a symmetric array reverses but still maintains a unidirectional flow, whereas, in the asymmetric array, the unidirectional flow turns to vortices with no pumping action.

In general, factors other than frequency can also lead to flow reversal. For example, ACEO systems can face flow reversal at higher voltages due to faradaic charging [116]. Transition between ACEO and ACET mechanisms [93] and steric effect can also be important [117,118]. However, none of these reasons can justify the flow reversal in the DC biased AC electrothermal device of Lian et al. [6], which is still open for further investigation.

Flow reversal is considered by most studies as a disadvantage in ACEK due to the uncontrollable and unpredictable nature of this phenomenon. However, some studies have investigated controllable flow reversal by tuning the actuation frequency, switching electric field, and/or applying external heat sources [1,39,68,113,114].

#### **7. Application**

Most common applications of the electrothermal effect in micro systems include mixing and pumping of fluids and particle manipulations [1,13,60,62]. A good example of using the electrothermal effect for mixing is immunoassays [11,14,20,61,119–121]. Immunoassays are biochemical tests in which the concentration of macromolecules (e.g., ligands or proteins) in a biofluid is measured by the use of an antibody. The antibody is immobilized on a surface and then the biofluid of interest is introduced above the surface. The concentration of macromolecules is measured by the number of macromolecules attached to the antibodies. The key factor in this process is for macromolecules to contact the antibodies repeatedly so that all possible bindings take place and lead to a precise sensing result. Since this process is diffusion limited, incubation time can take hours [61]. By using the electrothermal effect, the chance of macromolecules reaching the sensing area can be increased, thereby significantly decreasing the response time, enhancing the bindings by seven to nine times within minutes [14,61]. Sigurdson et al. investigated this idea numerically and found that by applying a voltage of 6 Vrms, an increase of seven times in the amount of bound antigen can be reached. It should be noted that electrothermal stirring is effective only when Damkohler ≥100, i.e., when the process is diffusion limited and not reaction limited. Therefore, electrothermal stirring is not significantly effective for reaction limited systems such as DNA systems. Sigurdson et al. also found that the electrothermal stirring is especially effective in the space above the electrode gap, where the velocity and the concentration gradient are high, making it the optimum place for antibodies to be immobilized [11,14,61]. This was proven experimentally in their later work [61]. Huang et al. found the optimum reaction site to be closer to the negative electrode in a symmetric rectangular pair [122]. In this case, both Damkohler and Peclet numbers need to be considered. Electrothermal stirring is best for mass transport limited regimes, where the Peclet number is relatively low, and convection contributes more to the overall mass transport than diffusion. Therefore, by utilizing the electrothermal effect, the required sample volume can be reduced, leading to higher efficiency devices [14]. It has also been shown that in immunoassay applications, electrothermal force can be a better choice compared to electroosmotic force, as electroosmotic force may cause the antigen–antibody bounds to fall apart [14].

In 2012, Sasaki et al. used a meandering electrode configuration in a Y-shaped channel to mix two high salt content fluids [62]. They reported a fivefold reduction in mixing time compared to diffusional mixing. In their study, the dependence of mixing index on salt concentration, frequency, and mixing time was investigated. It was concluded that the meandering structure was suitable for salt concentrations of 10−<sup>3</sup> to 10−<sup>1</sup> mol·dm<sup>−</sup>3, provided that the frequency lied in the range of 100–200 kHz, which is typical for electrothermal devices. A long range ACET effect, where centimeter scale vortices are generated, can also be used for mixing purposes [70].

Further development on electrothermal based immunoassays was carried out by Liu et al. in 2011 [11]. They decreased the incubation time from 30 min to 3 min by implementing ACET in their immunoassay with the conventional asymmetric electrode array. A study on electrode geometry on capacitive immunoassays showed that electrode geometry is an important component in high electric field configurations with asymmetric geometries rendering higher detection efficiency [123]. Another study suggested that the placement of electrodes on the same surface as the reaction site renders the most efficient configuration for immunoassays [121]. Selmi et al. studied the effect of temperature on immunoassays with asymmetrical electrodes [124]. The use of a pulsed ACET flow and amplitude modulated (AM) sine waves has also been reported for enhancing mixing efficiency in immunoassays [125,126].

Electrothermal effect in conjunction with dielectrophoresis can be used for the trapping and assembly of particles, patterning of colloids, and preconcentration of biological samples for detection and characterization purposes [13,56,59,75,127–131]. Additional studies have shown the enhancement of DNA hybridization with ACET configurations [132]. Recently, the concept of an AC electrothermal micropump was applied to a cell-culture system (with culture media of conductivities up to 2 S·m<sup>−</sup>1) towards the development of organ-on-a-chip and human-on-a-chip systems [63,133].

Recent studies have also shown that, although other ACEK effects such as DEP can enhance trapping of biological samples, they render nonspecific responses [134].

#### **8. Substrate Material**

As electrodes are patterned on a substrate, and the substrate is one of the channel walls through which the heat is dissipated, the choice of substrate material is of great importance [135]. The materials typically used for the fabrication of microfluidic devices include silicon, glass, polydimethylsiloxane (PDMS), and polymethylmethacrylate (PMMA). As a commonly used substrate material, glass has a lower thermal conductivity compared to silicon and can yield higher temperature gradients, and thus, higher rates of electrothermal flow [105]. Heat transfer through the substrate can be controlled by other means too. For example, a thermoelectric cooler under the silicon substrate has been suggested for maintaining the temperature gradient of interest in an ACET device [61]. As opposed to glass, silicon has a higher heat transfer coefficient, and thus, is a good candidate for applications where excessive temperature rise is undesirable [61,85,87,92,136]. Silicon is also admired for its precise geometrical features and low surface roughness [85]. Compared to silicon, the thermal conductivity of PDMS is much lower, causing the temperature rise in PDMS microchannels to be significantly high [92,136].

#### **9. Channel Height**

Unlike ACEO, the origin of ACET is the charge density generated in the bulk of the fluid. Therefore, the height of the microfluidic channel plays an important role in the formation of microflows. The microflows will be suppressed when the channel height is too small (<200 μm in a typical microfluidic ACET device) [13,76]. In contrast, ACEO devices utilize a thin layer of electric double layer responsible for dragging the fluid bulk through the microchannel, meaning that reducing the channel height causes the ACEO effect to be more effective.

It has been shown that in two-phase actuation systems, the optimal channel heights are in the range of 500–1000 μm [69]. In a planar configuration, the maximum velocity is reached at the height of ~500 μm, above which no significant difference in the flow rate can be observed [76,85]. In microgrooved structures, however, it is shown that a velocity of five times higher than planar electrode configurations can be reached at a significantly smaller channel height, i.e., ~50 μm, which is 10% of the optimal height in conventional and two-phase systems. This feature of the microgrooved configuration helps further miniaturizing the ACET based devices. As mentioned above, miniaturization is hindered in ACET devices due to dependence of the microflows on the channel height. With increasing the channel height above 50 μm in a microgrooved configuration, the velocity drops but still stays at higher values compared to a planar configuration.

In the case of patterning electrodes both on the top and bottom of the microchannel in an ACET device, increasing the channel height above 200 μm causes the velocity profile to become more similar to that of an ACEO device [104]. A thorough study on the effect of channel height in micropumps can be found in reference [76].

#### **10. Numerical and Experimental Settings**

In this section, some key points and influential factors in numerical simulation and experimental setup of ACET devices are briefly reviewed. To assure that the same conditions are valid to apply to other devices of interest, the readers are advised to refer to the articles associated with each statement.

#### *10.1. Numerical Simulation*


#### *10.2. Experimental Setup*


• For generating effective electric field at the electrode surface, electrodes should be fabricated relatively thin, e.g., 50–100 nm [32].

#### **11. Future Work**

Although utilizing the ACET effect for various biomedical applications has been investigated extensively over the past two decades, more investigative work is yet to be carried out to further our knowledge of the phenomenon. Existing drawbacks must be better addressed in order to facilitate utilization of this effect in biomedical laboratory settings. For example, temperature rise, and its detrimental effects on biological samples, is of great concern while using the ACET effect in microfluidic devices. Substrate materials with high heat conductivities, which enable a low temperature rise while keeping high voltages, must be investigated to achieve strong electrothermal flow.

While plenty of novel and efficient electrode designs have been proposed for pumping and mixing applications, the majority of such reports are numerical studies. A lack of experimental studies that would reveal the hidden challenges in applying these novel strategies to real life applications exists in the literature. Such lack of experimental works is mainly due to the limitations in microscale fabrication and electrode degradation occurred at high voltages. To address such fabrication challenges, techniques for low cost fabrication of prototypes of such complicated designs and investigation of different electrode materials and coatings to withstand high voltages [79] are in great demand.

**Author Contributions:** Writing—original draft preparation, M.N. and A.S.; writing—review and editing, M.N., A.S., T.L., and C.D.; supervision, A.S., C.D.; funding acquisition, C.D.

**Funding:** This research was funded by a Canadian Natural Sciences and Engineering Research Council (NSERC) Discovery Grant.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **Appendix A**

The majority of the publications cited throughout the manuscript which have made major contributions are summarized in Table A1.


**Table A1.** A short summary of relevant articles studying the electrothermal effect, useful for new researchers in this field as a guide to what has been done and those who are working in the field.





#### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

MDPI St. Alban-Anlage 66 4052 Basel Switzerland Tel. +41 61 683 77 34 Fax +41 61 302 89 18 www.mdpi.com

*Micromachines* Editorial Office E-mail: micromachines@mdpi.com www.mdpi.com/journal/micromachines

MDPI St. Alban-Anlage 66 4052 Basel Switzerland

Tel: +41 61 683 77 34 Fax: +41 61 302 89 18