**A Review on Optoelectrokinetics-Based Manipulation and Fabrication of Micro**/**Nanomaterials**

**Wenfeng Liang 1, Lianqing Liu 2,3,\*, Junhai Wang 1, Xieliu Yang 1, Yuechao Wang 2,3, Wen Jung Li 3,4,\* and Wenguang Yang <sup>5</sup>**


Received: 20 November 2019; Accepted: 8 January 2020; Published: 10 January 2020

**Abstract:** Optoelectrokinetics (OEK), a fusion of optics, electrokinetics, and microfluidics, has been demonstrated to offer a series of extraordinary advantages in the manipulation and fabrication of micro/nanomaterials, such as requiring no mask, programmability, flexibility, and rapidness. In this paper, we summarize a variety of differently structured OEK chips, followed by a discussion on how they are fabricated and the ways in which they work. We also review how three differently sized polystyrene beads can be separated simultaneously, how a variety of nanoparticles can be assembled, and how micro/nanomaterials can be fabricated into functional devices. Another focus of our paper is on mask-free fabrication and assembly of hydrogel-based micro/nanostructures and its possible applications in biological fields. We provide a summary of the current challenges facing the OEK technique and its future prospects at the end of this paper.

**Keywords:** optoelectrokinetics; optically-induced dielectrophoresis; micro/nanomaterials; separation; fabrication

#### **1. Introduction**

Accurate manipulation and fabrication of micro/nanomaterials in liquid is fundamental for a range of applications such as micro/nanoelectronics [1,2], biosensors [3,4], biomedicine [5,6], biosensing [7,8] and energy harvesting [9,10]. Various attempts have been made to make that happen. For example, silver nanoparticles were integrated into the luminol system to enable more efficient electrochemiluminescence and thereby allow for ultrasensitive detection of cardiac troponin [11]. Combining graphene and traditional integrated circuits, a high-mobility, high-resolution, and broadband image digital sensor was developed to capture ultraviolet, visible, and infrared light [12]. Gold nanoparticles were assembled into micro/nanowires to fabricate a flexible pressure sensor that offers a detection limit up to 25 Pa [13]. This sensor is also sensitive to pulses in different regions of the human body, offering an approach to facilitating the development of wearable devices.

To address real-world needs for manipulation and fabrication of micro/nanomaterials, a number of micro-/nano-scaled methods have been presented. Typical examples include microfluidic [14,15], acoustic [16,17], electrokinetics [18,19], magnetic [20,21], optical [22,23], thermal [24,25], and atomic force microscope [26,27] approaches. An emerging topic on the manipulation and fabrication of micro/nanoparticles is about how to develop a novel mechanism that can complement what a single technique can offer. The long-term aim is to create valuable techniques that can help extend the applicability of micro/nanomaterials. Combining acoustic and magnetic fields, an approach that can aggregate nanoparticles in the presence of a magnetic field and move them towards the wall of a microchannel under the action of an external acoustic wave was proposed [28]. This approach bears significant potential for application in target drug delivery and biomedical surgery. There is also a technique called the acoustofluidic tweezers, which integrates acoustic waves into a microfluidic system and offers a label-free and high-throughput way to isolate 110-nm particles from a mixture of micro/nanoparticles with a purity as high as 99% [29]. By incorporating white light source from light-emitting diode into a microfluidic system, the absorbance of micro/nanoparticles at different diameters as well as the bit error rate can be determined, which allows easy and rapid measurement of micro/nanoparticle concentration [30]. In addition, a hybrid isomotive dielectrophoresis (DEP)-microfluidic technique was proposed for label-free separation of same-/differently-sized micro/nanoparticles based on their crossover-frequency features [31]. This technique also enables reliable and repeatable separation and dynamically adjusting the purity and yield of separated beads.

A new hybrid and novel manipulation technique called optically-induced dielectrophoresis (ODEP) or optoelectrokinetics (OEK) was proposed for programmable, contact-free, flexible, automatic, dynamic, and rapid manipulation of micro/nano-entities [32]. OEK has attracted much attention in micro/nanomanipulation fields since it was invented in 2005. This technique combines the merits of optical, electrokinetics, and microfluidic schemes, offering a more versatile approach to micro-/nano-scaled manipulation and fabrication over other competing lab-on-a-chip techniques. Optically-projected images are custom-designed on a computer using graphic animation software. These images are then transmitted by a commercial projector and focused onto the surface of the OEK chip, thereby triggering the photosensitive material and changing the distribution of the external AC bias potential. Most of the AC bias potential will shift to the liquid layer, which generates a non-uniform electric field around the illuminated area. Meanwhile, various electrokinetics forces are produced. These forces are further exerted onto the micro/nanoparticles, driving, directing, and delivering them towards the intended destination programmatically and digitally. This is how optical-electrokinetics-microfluidics integration works. Featuring the use of optically-projected images as virtual electrodes to directly manipulate micro/nanoparticles and without referring to metal-based electrodes, OEK has been widely used in the manipulation, separation, assembly, and fabrication of micro/nano-entities as well as extraction of their intrinsic properties [33–38], which has shown a highly promising new strategy in the development of micro/nanomanipulation and fabrication communities. More recently, Berkeley Lights Inc., the inventor of OEK, has commercialized this technology in bio-related fields, offering a unique approach to take the development of biomedical and bioengineering sciences to a new level. Furthermore, it also shows the high potential for the use of this technology in material manipulation and fabrication to move from lab-oriented research to real-world applications.

Unlike previous review studies that only discuss some aspects of OEK-based microfluidics [39–41], we present a comprehensive review of OEK-based manipulation and fabrication of micro/nanoparticles. First, we summarize differently structured OEK chips and discuss their respective working principles. Then, we describe OEK-based manipulation and fabrication of micro-/nano-scaled materials without using any masks. Next, we review the use of OEK chips for fabricating various hydrogel-based structures and functional micro/nanodevices. Finally, we explain current and future challenges facing the OEK technique and possible innovations on it.

#### **2. Optoelectrokinetics (OEK) Chips and Their Working Principles**

The ODEP or OEK chip was firstly invented in 2005 [32]. The chip typically consists of four layers [32]: a top glass layer coated with a transparent and conductive indium tin oxide (ITO) film, which connects with one end of the AC bias potential; a bottom ITO glass layer serving as the bottom electrode that connects with the other end of the AC bias potential; a thin film (50 nm) n+ hydrogenated amorphous silicon (a-Si:H) layer, deposited onto the bottom ITO glass after generation of a 1 μm undoped a-Si:H layer and a 20 nm silicon nitride layer; a liquid layer that contains the manipulated micro/nanoparticles to function as a "gap" for the OEK chip. The doped a-Si:H layer can decrease the contact resistance between the liquid layer and the undoped a-Si:H layer, and the silicon nitride layer serves the purpose of passivation. It is worth noting that the undoped a-Si:H layer is the photosensitive layer from which optically-induced virtual electrodes are generated. In general, the ITO layer is sputtered onto the glass layer; the a-Si:H layer is deposited onto the ITO layer by plasma-enhanced chemical vapor deposition. Figure 1 is an illustration of its mechanism proposed in our group given in [42]. In our study, only a layer of a-Si:H was deposited, which was also classified into an a-Si:H based OEK chip.

When the OEK chip is illuminated by the optically-projected patterns shown on a digital micromirror display (DMD), the external AC voltage shifts to the liquid layer with suspended micro/nanoparticles. Then, an optically-induced non-uniform electric field is produced around the illuminated area. Under the action of this electric field, the micro/nanoparticles are polarized, producing an ODEP force. This force is, in turn, exerted onto the micro/nanoparticles, which is defined as [43]

$$
\left\langle \overrightarrow{F}\_{\text{ODEP}} \right\rangle = 2\pi R^3 \varepsilon\_m \text{Re}[K(\omega)] \nabla \left| \overrightarrow{E}\_{rms} \right|^2,\tag{1}
$$

where *R* is the radius of the micro/nanoparticles, ε is the permittivity of the liquid, *E* is the magnitude of the optically-induced non-uniform electric field, *rms* means the root-mean-square value, the subscript *m* is the liquid, ω is the angular frequency of the AC bias potential across the liquid medium, and *K*(ω) is the Clausius–Mossotti (CM) factor, which is further expressed as

$$K(\omega) = \frac{\varepsilon\_p^\* - \varepsilon\_m^\*}{\varepsilon\_p^\* + 2\varepsilon\_m^{\*\prime}} \, ^\prime \tag{2}$$

where <sup>ε</sup>\* = <sup>ε</sup> <sup>−</sup> jσ/ω, with <sup>σ</sup> representing the conductivity. The subscript *p* means the particle. The direction of the ODEP force, both positive and negative, is fully directed by the real part of the CM factor. Specifically, if the value of the real part of the CM factor is higher than zero, a positive ODEP force will be generated and then the micro/nanoparticles will be attracted and moved to the illuminated area. Otherwise, the micro/nanoparticles will be pushed away from the illuminated area. For micro/nanoparticles that vary significantly in their size, a common method to separate them is by making the ODEP magnitudes have three orders of difference.

Other OEK forces, including AC electroosmosis (ACEO) and AC electrothermal, were investigated using the finite element analysis (FEA) method [44]. Typically, those two kinds of electrokinetics forces arise from the interaction between the optically-induced non-uniform electric field and liquid solution at a given AC frequency. As an example, under the action of an ACEO force, the fluid flow can rapidly push a large number of micro/nanoparticles towards the illuminated area and assemble them into desired patterns or devices [45,46].

Recently, various types of photosensitive materials have been explored to manipulate and fabricate micro/nanomaterials in the same way as an a-Si:H based OEK chip works. A P3HT/PCBM bulk-heterojunction polymer was demonstrated as an alternative to a-Si:H for micro/nanomanipulation [47]. Figure 2 shows the structure of an OEK chip made up of this material. P3HT/PCBM serves as a photosensitive material to respond to optically-projected images and produce a non-uniform electric field, thereby manipulating and isolating differently sized polystyrene beads [48]. The P3HT/PCBM polymer for making this type of chip is created by spinning-coating at a relatively low temperature. Compared to a-Si:H based OEK chips, OEK chips based on this material are much easier to fabricate, except that the fabrication process of the latter requires preventing the polymer film from being collapsed by either water or oxygen. OEK chips based on this material can bend up and down freely to produce concave or convex curvatures, which leads to significantly higher efficiency in separating and concentrating differently sized polystyrene beads [49].

**Figure 1.** Schematic illustration of optoelectrokinetics (OEK) chip (reproduced from Ref. [42]).

**Figure 2.** Structure of P3HT/PCBM polymer-based OEK chip (reproduced from Ref. [47]).

Another kind of OEK chip was devised based on TiOPc, an organic photosensitive material. As shown in Figure 3 [50], the chip has the same structure as the one indicated in Figure 1. This chip can be easily fabricated only by spinning-coating and baking techniques. This chip is fabricated in a much easier way than an a-Si:H based chip. It has been demonstrated that this chip can perform real-time manipulation of picobubbles [51] and droplets [52] as well as cell patterning [53]. However, problems remain with the long-term stability of this OEK chip.

**Figure 3.** Structure of TiOPc-based OEK chip and simulation of the electric field (reproduced from Ref. [50]).

#### **3. OEK-Based Manipulation of Micro**/**Nanomaterials**

#### *3.1. Separation and Assembly of Micro-Scaled Particles*

The OEK chip has been commonly used to perform manipulations such as separation, concentration and assembly on micro/nanoparticles, with the aid of an ODEP and/or ACEO mechanism. The OEK chip permits the assembly of 2D colloidal microparticles by using the electrohydrodynamic flows [54–56] and the manipulation of droplets [57–59]. Differently sized polystyrene beads serve various functions in material science and biomedical research. Hence, finding a rapid and automatic method to manipulate and separate them is critical in investigating the performance of targets of interest.

The conductivity of polystyrene beads is size-dependent and expressed as σ = 2*KS*/*R*, where *KS* is the surface conductivity of polystyrene beads [42]. The relationship between the crossover frequency (Re [*K*(ω)] = 0) of the ODEP force and the size of the polystyrene beads can be further derived as [42]

$$f\_{\text{crossover}} = \frac{1}{2\pi} \sqrt{\frac{(\sigma\_m R - 2K\_S)(2K\_S + 2\sigma\_m R)}{R^2(\varepsilon\_p - \varepsilon\_m)(\varepsilon\_p + 2\varepsilon\_m)}}.\tag{3}$$

The size-dependent crossover frequency as a function of the liquid conductivity, i.e., Equation (3), is illustrated in Figure 4 [42]. Then, separation of 1 μm and 10 μm polystyrene beads was successfully demonstrated by using ODEP forces in different directions, i.e., a positive ODEP force exerted onto the 1 μm polystyrene beads and a negative one onto the 10 μm ones, as shown in Figure 5 [42].

**Figure 4.** Crossover frequency vs. liquid conductivity of three differently sized polystyrene beads (reproduced from Ref. [42]).

Moreover, polystyrene beads with three different diameters, i.e., 500 nm, 1 μm, and 10 μm, were separated simultaneously. Figure 6 shows the detailed experimental process [42]. A negative ODEP force would be exerted onto the 10 μm diameter polystyrene beads, while a positive ODEP force would be exerted onto both the 500 nm and 1 μm ones. However, the 1 μm polystyrene beads would experience a much higher magnitude of ODEP force than the 500 nm ones because the magnitude of the ODEP force is proportional to the third power of the particle radius.

**Figure 5.** Separation of 10 μm and 1 μm polystyrene beads. (**a**) Initially, polystyrene beads were suspended in the liquid solution; (**b**) the OEK chip was illuminated by the optical ring pattern, with the AC bias potential switched on simultaneously; (**c**) 10 μm polystyrene beads were pushed towards the central area of the ring under the action of a negative ODEP force, while the 1 μm ones were attracted into the ring under the action of a positive ODEP force as the ring size decreased; (**d**,**e**) are captured images showing the positions of the polystyrene beads with two different diameters as the ring size decreased dynamically; (**f**) shows the final positions of the 10 μm and 1 μm polystyrene beads (reproduced from Ref. [42]).

**Figure 6.** Simultaneous separation of polystyrene beads with diameters of 500 nm, 1 μm, and 10 μm. (**a**) When the OEK chip was illuminated by the optical rectangle pattern under the action of an external AC bias potential, both the 500 nm and 1 μm polystyrene beads were attracted to the illuminated area. By contrast, the 10 μm ones were pushed away from the optical ring; (**b**) as the optical rectangle size decreased, the 10 μm polystyrene beads moved towards the central area of the optical rectangle and those with the other two diameters still resided within the optical ring; (**c**) when the gap formed by the long sides of the rectangle was around 10 μm, the 10 μm polystyrene beads were aligned; (**d**) the 10 μm polystyrene beads were separated as the rectangle size further decreased; (**e**) the rectangle was decreased to a line pattern; (**f**) the 500 nm and 1 μm polystyrene beads were finally located at different positions of the OEK chip (reproduced from Ref. [42]).

Manipulation and assembly of metallic microspheres into patterns were proposed [60]. It was observed that conductive silver-coated poly(methyl methacrylate) (PMMA) microspheres (50 μm diameter) could move in an OEK chip at a maximum velocity of 3200 μm/s, much quicker than

non-conductive microparticles such as polystyrene beads [61]. As the motorized XY stage moves at an increasing velocity, the microspheres reach their maximum velocity when they cannot follow the movement of the stage. Simulation on the electric field distribution and the ODEP force attributes the strong ODEP force to the local interaction between the optically-induced electric field and the silver shells surrounding the microspheres. The microspheres were then experimentally assembled to validate their high-accuracy positioning capabilities. Figure 7 shows how the microspheres were assembled with different spaces, i.e., from 1.39 μm to 20.7 μm, relative to a stable bubble [60]. The assembly of the "O," "E," "T" pattern was realized, demonstrating the ability of the OEK chip to precisely and parallelly pattern metallic microspheres into arbitrary shapes. In addition, the manipulation and assembly of hybrid metal-polymer microparticles were achieved [62].

**Figure 7.** Captured images of manipulated metallic microparticles with different spaces to a bubble: (**a**) 1.39 μm, (**b**) 2.5 μm, (**c**) 3.3 μm, (**d**) 5.3 μm, (**e**) 8.4 μm, (**f**) 10.6 μm, and (**g**) 20.7 μm; parallel assembly of metallic microspheres into images of (**h**) "O," (**i**) "E," (**j**) "T" (reproduced from Ref. [60]).

It has been further demonstrated that OEK can serve as a versatile and programmable microrobot to perform typical micromanipulations such as "load," "transport," and "deliver" [63]. Firstly, OEK was used to manipulate custom-designed microstructures. Then, the OEK based microrobot could manipulate secondary microparticles in a parallel, multistep, contact-free and programmable manner. Figure 8 is a series of images showing the dynamic process of the OEK-based microrobot across large distances. This partially enclosed microrobot could load one 15 μm diameter polystyrene bead of interest with the aid of a negative ODEP force (Figure 8A–C), followed by a translational movement of 300 μm/s (Figure 8D). In addition, the OEK-based microrobot delivered this bead to the target location (Figure 8E,F). Figure 8G schematically presents the three-step process, which exhibited a higher moving velocity than when OEK was used alone. This OEK-based microrobot was also validated to be suitable for isolating single cells for clonal expansion and other biomedical applications.

**Figure 8.** A series of OEK-based robotic micromanipulations. (**A**) A fully enclosed microrobot; (**B**) the load mode of a partially enclosed microrobot; (**C**) the partially enclosed microrobot loaded one bead of interest; (**D**) the fully enclosed microrobot transported the bead; (**E**) the partially enclosed microrobot delivered the bead; (**F**) the bead was unloaded; (**G**) schematic illustration of the "load," "transport," and "deliver" micromanipulations (reproduced from Ref. [63]).

#### *3.2. Manipulation of Nano-Scaled Particles*

OEK has also been used to dynamically manipulate nano-scaled entities, including the separation of nanowires [64–67], the patterning of two-dimensional nanomaterials [68], and manipulation of nanoparticles [69–74].

OEK allows the use of ODEP forces to effectively separate individual nanowires with different conductance levels, which has facilitated the development of nanodevices. Dynamic separation of semiconducting and metallic nanowires based on their difference in translational velocity was reported [64]. When the external voltage was higher than the "separation voltage," i.e., the threshold one for moving the silicon nanowire, the silver nanowire could be separated from the silicon nanowire. Due to the high polarization of metallic nanowires in a non-uniform electric field, the silver nanowire moved at a much higher velocity than the silicon nanowire. When the OEK chip was dynamically illuminated by an optically-projected laser line at a scanning velocity greater than 2 μm/s and an AC bias potential of 8 Vpp, both of these two types of nanowires were trapped by the laser line initially. The silicon nanowire, however, could not follow the movement of the laser line and became "uncontrollable." Hence, these two types of nanowires were successfully separated. In addition, it was experimentally validated that real-time and large-scale assembly of silver nanowires is possible using an array of image-defined traps, suggesting the potential for massively parallel assembly at the nanoscale.

Furthermore, the manipulation and assembly of nanoparticles have been studied both theoretically and experimentally. First, a rapid and automatic assembly of 100 nm diameter gold nanoparticle (AuNP)-based microstructures was experimentally investigated [72]. Figure 9 indicates the experimental process in which AuNP-based microstructures were rapidly assembled. It was observed that each geometry of the four AuNP-based microstructures could reflect the optical pattern. However, using the triangle pattern as the virtual electrodes attracted most of the AuNPs into the locations of angular bisectors and formed a circle structure with three lines pointing toward the center of the triangle. The square pattern pushed the AuNPs to the diagonals of the pattern and most of them were moved into the central area.

**Figure 9.** Rapid assembly of various microstructures using gold nanoparticles (AuNPs) with a diameter of 100 nm. (**a**) Four different optical patterns served as virtual electrodes, with an external AC bias potential switched on simultaneously; (**b**) after 30 s, the AuNPs were attracted into the areas illuminated by the four patterns; (**c**) AuNP-based microstructures formed when the optical patterns were moved and AC bias potential was switched off; (**d**) SEM images of (**c**); (**e**) SEM image of an AuNP-based microstructure in triangle pattern; (**f**) SEM image of an AuNP-based microstructure in square pattern (reproduced from Ref. [72]).

Fabrication of various nanomaterial-based microelectrodes in an OEK chip was achieved using two different methods [73]. A composite solution consisting of conductive polyaniline (PANI) nanoparticles and multi-walled carbon nanotubes (MWCNTs) was used to examine the possibility of creating microelectrodes by OEK forces. Microelectrodes with various geometrical sizes could be fabricated within 1–3 min. Compared to the resistance change of the MWCNTs bundles, that of the microelectrodes could be neglected; the ethanol concentration was successfully reflected by the resistance change of the sensitive elements.

Rapid assembly of carbon nanoparticles (CNPs) with a diameter of 50 nm into electrical elements was realized [74]. A series of experiments were conducted to rapidly assemble CNPs into various electrical elements within 45 s. The results demonstrated that the microstructures had resistance properties. Specifically, their resistance value could be controlled by the width and length of the microstructures: inversely increasing with the width (Figure 10a–c) and linearly increasing with the length (Figure 10d–f).

**Figure 10.** The curve-fitting function of the assembled carbon nanoparticle (CNP)-based microstructures. Resistance with respect to different widths of three microstructures at three different measurement voltages (**a**–**c**); Resistance with respect to different lengths of three microstructures at three different measurement voltages (**d**–**f**) (reproduced from Ref. [74]).

#### **4. Mask-Free Fabrication of Electrodes and Devices**

An optically-induced electrochemical reaction and deposition scheme was presented to enable dynamic, rapid and mask-free fabrication of microelectrodes [75–79]. When the OEK chip was illuminated by optically-projected images, an electrical field would be produced in the illuminated area due to the creation of electron-hole pairs. Then, only the metal ions of liquid solution in the illuminated area were reduced by trapping electron from a-Si:H when an external AC bias potential was switched on. Furthermore, the reduced metal atoms would be attached to the illuminated a-Si:H surface and formed into metallic microstructures with the same shape as the incident light.

Compared to the OEK-force-based method, this scheme could dynamically fabricate microelectrodes in 10 secs with a liquid conductivity as high as 2 <sup>×</sup> 107 S/m. For instance, silver ion-based microelectrodes were reported, as shown in Figure 11 [75]. These microelectrodes exhibited more even distribution and lower roughness. Their heights could be adjusted by the linearly increasing deposition time and the solution concentration. In addition, the entire experiment was conducted at room temperature and atmospheric pressure without using conventional photolithographic techniques and metal nanoparticles and/or inks. In addition, integrated CuO/ZnO-based/single-walled-nanotube [76] nanowire-based field-effect transistors were presented, further validating that this scheme could facilitate the manufacturing of integrated nanodevices and offer an alternative to micro/nanosensor fabrication.

**Figure 11.** (**a**) Fabricated microelectrodes with respect to deposition time; scale bar: 10 μm. (**b**) Surface roughness of microelectrode by atomic force microscope (AFM) and SEM. (**c**) Thickness and (**d**) surface roughness of microelectrodes with respect to deposition time and solution concentration (reproduced from Ref. [75]).

Silver-based nanostructures with various geometrical topographies were synthesized in an OEK chip, as shown in Figure 12 [77]. The incident light excited the generation of electron-hole pairs in the a-Si:H layer and then the electrons migrated from the valence band to the conduction band. With the aid of an existing electric double layer, an electrochemical reaction occurred between the electrons and the suspended silver solution in the liquid chamber under given parameters in the solution; meanwhile, crystallization occurred with the silver-based nanostructures. In this case, silver polyhedral nanoparticles, nanoplates in triangle and hexagon patterns, as well as nanobelts were fabricated by projecting various optical patterns. Experimental parameters that affected the fabrication process, including time duration, AC frequency and voltage, were also investigated and optimized. In sum, this OEK-based method opens up a new path to mask-free and rapid synthesis of nanostructures and nanobelts.

**Figure 12.** Synthesis of various silver nanostructures by optically-projected patterns in OEK chip. (**a**) Silver crystal nanoparticles. (**b**) Stacked silver hexagonal nanoplates. (**c**) Crystallized silver octahedra and hexagonal silver nanoplates. (**d**) An array of silver nanoparticles-based micropillars (reproduced from Ref. [77]).

Using this method, MoS2 with excellent optical and electronic properties was fabricated into thin-film transistors without relying on any conventional microlithography, such as nanoimprint lithography, laser patterning, or photolithography [78]. The MoS2 material was first loaded into the OEK chip. Then, the Au and Ag electrodes were rapidly fabricated onto the target MoS2 film. Accordingly, MoS2 thin-film transistors with Au and Ag electrodes were obtained, respectively, as shown in Figure 13. These transistors performed the best when they were 30–40 nm thick, exhibiting a low subthreshold swing of 0.75 V/decade and high mobility of 41 cm2·V−1·s−1, much better than conventional Si-based thin-film transistors.

**Figure 13.** SEM images of MoS2 thin-film transistors fabricated with Au (**a**) and Ag (**d**) electrodes; (**b**,**e**) are the enlarged view of the rectangle areas in (**a**,**d**), respectively; (**c**,**f**) are the AFM characterization results of the heights of MoS2 thin-film transistors (reproduced from Ref. [78]).

#### **5. Fabrication and Assembly of Hydrogel-based Micro**/**Nanostructures**

Mask-free and non-UV based polymerization and prototyping of high-aspect-ratio 3D hydrogel microstructures, such as poly (ethylene glycol) (PEG)-diacrylate (PEGDA), was demonstrated in an OEK chip [80–85].

A laser or UV-based method was proposed to obtain PEGDA-based micro/nanostructures [86,87]. The abovementioned OEK-based mask-free method could meet the same purpose without using any UV light sources or lasers [80]. Figure 14 shows the experimental results with controlled sizes, shapes and thicknesses of PEGDA-based structures. Parallel micro/nanostructures were flexibly patterned onto the chip using optical images. Furthermore, by projecting a series of custom-designed and dynamic optical patterns that served as digital masks, 3D microstructures were fabricated rapidly

in a layer-by-layer manner, with their thickness varying from tens of nanometers to hundreds of micrometers [81].

**Figure 14.** SEM images of poly (ethylene glycol) (PEG)-diacrylate (PEGDA)-based micro/nanostructures fabricated using optically-projected spot patterns with different exposure times. The heights in (**a**–**f**) measured by AFM were 233 nm, 455 nm, 687 nm, 950 nm, 1.34 μm, and 1.62 μm, respectively (reproduced from Ref. [80]).

In addition, hollow and circular tubes were fabricated with a length, diameter, wall thickness, and high-aspect-ratio tuned by the exposure time, as shown in Figure 15 [82]. These tubes could manipulate and trap polystyrene beads when they grew longer as the exposure time increased. Then, the PEGDA-based tube continued to elongate and the beads were trapped and moved away from the initial position. Arrays of hydrogel micropillars were also fabricated. These micropillars were capable of serving as micro-scaled cavity molds for casting polydimethylsiloxane, demonstrating potential of finding applications in microfluidics-related fields. The abovementioned results indicate this technology is well-suited for the rapid fabrication of microfluidic chips.

**Figure 15.** (**a**) PEGDA-based hydrogel tube with a length of ~20 μm, an outer diameter of ~10 μm, and an inner diameter of ~5 μm. (**b**) PEGDA-based hydrogel tube with a length of ~35 μm, an outer diameter of ~25 μm, and an inner diameter of ~20 μm (reproduced from Ref. [82]).

Assembly of PEGDA-based microstructures was also realized, followed by an application of bottom-up functional tissue construction and engineering [83]. Arbitrary PEGDA-based microstructures were polymerized in a high-throughput and on-demand manner by a DMD. Then, the fabricated microstructures were transferred into an OEK chip for assembly, which was achieved by using microfluidic flow. Figure 16 illustrates the assembly of PEGDA-based microstructures with different sizes and shapes. Before moving into the OEK chip, the microstructures were either fabricated individually or patterned on a per-array basis. Then, microstructures were aligned into horizontal lines under the action of an ODEP force. Figure 16A–G present the dynamic process of translating microstructures into a single structure for further biomedical and tissue applications. Furthermore, microstructures with the same shape could be assembled into the same layer, as shown in Figure 16H–L.

**Figure 16.** (**A**–**G**) Captured microscope image showing the dynamic process of assembling microstructures with different sizes and shapes; (**H**–**L**) the experimental process of assembling two different microstructures (reproduced from Ref. [83]).

An extended study was made on the microfabrication of 3D hydrogel scaffolds [84]. A layer-by-layer solidification mechanism was presented to fabricate hydrogel scaffolds, which involved a polymerization-delamination-polymerization loop. This loop was determined by a competition between the adhesive force and the water-absorbency-induced swelling force. The hydrogel scaffolds' thickness ranged from tens of nanometers to hundreds of nanometers. Thin hydrogel layers were cured at the interface of the a-Si:H and PEDGA solution layers in a layer-by-layer scheme with nano-scaled thickness. Furthermore, those layers were continuously stacked along the normal direction of the OEK chip to finally construct 3D structures. Figure 17 is a microscopic characterization of the dynamic multilayered electro-polymerization of 3D hydrogel structures. Two alternating lines perpendicular to each other were designed to fabricate hydrogel microstructures, as shown in Figure 17a. The optical microscope, SEM, and AFM were employed to illustrate the fabrication results at a Δ*t*<sup>0</sup> of 4 secs, as shown in Figure 17b–d, respectively. The results indicated that these mesh-like hydrogel scaffolds came with controlled pores and gaps. The folded PEGDA hydrogel grids shown in Figure 17e,f further validated the existence of pores and gaps, which could greatly facilitate the spreading, migration, and proliferation of cells as well as the mimicking of cell-cell communications.

**Figure 17.** (**a**) Projected patterns alternating at a time interval (Δ*t*0); (**b**–**d**) are the optical microscope, SEM and AFM observation results, respectively; (**e**,**f**) are the optical microscope and SEM images of folded PEGDA hydrogel grids, respectively (reproduced from Ref. [84]).

#### **6. Conclusions and Prospects**

As discussed above, the OEK technique has been widely used by the microfluidic community to make a list of research achievements in terms of the manipulation and fabrication of micro/nanomaterials Although some of these achievements are also accomplished by other competing techniques, OEK is superior to the others in that it consumes ultra-low power and offers a programmable, flexible and versatile approach for parallel and multi-scaled micromanipulation. OEK has also been found to be capable of detecting cellular properties and statuses and measuring drug concentration in a label-free and dynamic manner, without relying on any other techniques [88–94]. These advantages suggest that OEK is well-positioned to promote the use of microfluidic tools for micro/nanomanipulation. Nonetheless, challenges remain for OEK to transform from a lab-based technique to one that is widely useful in practical applications. Here is a summary of these challenges.

Firstly, the 3D manipulation mechanism of OEK should be further explored. Early studies on the OEK technique were mostly focused on 2D manipulation schemes, which proved to be more effective than other competing techniques. With the evolvement of OEK chips, however, researchers have realized the necessity of shifting their focus to 3D manipulation. One example of this shift in focus is a two-layer a-Si:H based OEK chip, in which the top ITO glass is replaced with a-Si:H to enable spatial and adsorption-free manipulation of beads using a negative ODEP force [95]. However, this OEK chip cannot manipulate micro/nanoparticles using a positive ODEP force and requires using other OEK forces. Besides, the manipulation is liable to be affected by dynamic fluid flow and electrothermal motion, which is because the manipulated micro/nanoparticles tend to be located in the center of the chip. The poor light transmission can significantly compromise the quality of observation. A lateral-field OEK device was developed to enable parallel single-cell manipulation, which can also assemble nanowires and integrate with other microfluidic components [96]. Nevertheless, the fabrication process is complicated and involves high costs. By dynamically projecting a series of optical patterns, a layer-by-layer approach was built to create 3D microstructures in a mask-free manner [81,84]. However, these incident light patterns are only suitable for fabricating hydrogel microstructures and are insufficient for manipulating nanoparticles and assembling nanowires into functional devices. All the above problems warrant continued efforts to further enhance capabilities of OEKs in 3D manipulation.

Secondly, to facilitate the adoption of OEK-based microfluidics, it is important to develop an integrated chip that is ready for a complete and conventional laboratory-level process. This requires the integration of OEK with conventional microfluidic systems and/or components. A number of attempts have been made on such integration to move OEK further beyond its typical functions such as manipulation, separation, assembly, and fabrication of micro/nano-entities. For instance, to enhance the detection performance of OEK, a pair of optical fibers were embedded into the microchannel to rapidly and accurately count the number of particles and cells [97,98]. A lens-free holographic microscope was incorporated into the OEK chip to allow observing microparticles and cells in a large field of view [99]. This approach offers the ability to rapidly capture the holographic images of microparticles and cells across an ultra-large area of up to hundreds of square millimeters. In addition, on-chip continuous medium exchange and electroporation of cells [100] were achieved by incorporating a conventional microchannel bonded onto a top ITO glass layer. The combination of an OEK chip and surface acoustic wave elements was utilized [101,102] to provide functions such as concentration, guiding, focusing, trapping, and sorting of polystyrene beads and cell lysis. Although the abovementioned works represent tangible advances in the OEK technique, there is still a lack of effective integration of micro/nanochannels or micro/nanostructures into the liquid layer of OEK chips. This severely hinders the use of OEK in fabricating arrays of micro/nanosensors and sophisticated micro/nanodevices for bio-detection and tissue engineering applications. The fabricated micro/nanostructures mentioned in this paper still reside within the OEK chip. They cannot be delivered or directed out of the chip due to the lack of a conventional micropump or microchannel integrated into the liquid layer. Hence, there is much more to do to better integrate OEK chips with microfluidic systems or elements.

Finally, increased efforts are required to identify new possible applications of the OEK technique, which is the ultimate challenge and also the key to address the above two challenges. We believe that a technique can only flourish by bringing real benefits to end-users. Although a variety of OEK-based applications, such as separation, assembly, patterning, fabrication, and synthesis of materials, have been reported, this technique still has a long way to go to become a truly useful and practical tool in these endeavors. For example, there has not been an OEK-based method to fabricate functional micro/nanodevices for industrial applications, such as field-effect transistors and nanosensors. In sum, if upcoming studies can focus more on solving the abovementioned challenges, the OEK technique will soon be able to find wide applications in real-world situations.

**Author Contributions:** W.L., L.L. and W.J.L. proposed the original idea, and planned the configuration. W.L. wrote the manuscript. J.W., X.Y., Y.W. and W.Y. revised the paper for language and quality. All authors have read and agreed to the published version of the manuscript.

**Funding:** This work was supported by the National Natural Science Foundation of China (grant numbers 61973224, 51805336, U1613220, U1908215 and 61803323); the Natural Science Foundation of Liaoning Province (grant numbers 2019-KF-01-15 and 2019-ZD-0673); and the Scientific Research Innovation Cultivation Project of Shenyang Jianzhu University (grant number CXPY2017012).

**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* **Autonomous and In Situ Ocean Environmental Monitoring on Optofluidic Platform**

**Fang Wang 1,2, Jiaomeng Zhu 1,2, Longfei Chen 1,2, Yunfeng Zuo 1,2, Xuejia Hu 1,2 and Yi Yang 1,2,\***


Received: 30 November 2019; Accepted: 7 January 2020; Published: 8 January 2020

**Abstract:** Determining the distributions and variations of chemical elements in oceans has significant meanings for understanding the biogeochemical cycles, evaluating seawater pollution, and forecasting the occurrence of marine disasters. The primary chemical parameters of ocean monitoring include nutrients, pH, dissolved oxygen (DO), and heavy metals. At present, ocean monitoring mainly relies on laboratory analysis, which is hindered in applications due to its large size, high power consumption, and low representative and time-sensitive detection results. By integrating photonics and microfluidics into one chip, optofluidics brings new opportunities to develop portable microsystems for ocean monitoring. Optofluidic platforms have advantages in respect of size, cost, timeliness, and parallel processing of samples compared with traditional instruments. This review describes the applications of optofluidic platforms on autonomous and in situ ocean environmental monitoring, with an emphasis on their principles, sensing properties, advantages, and disadvantages. Predictably, autonomous and in situ systems based on optofluidic platforms will have important applications in ocean environmental monitoring.

**Keywords:** optofluidics; ocean monitoring; colorimetric method

#### **1. Introduction**

The ocean is a vast repository of resources for human society and an essential material foundation for the sustainable development of society and environment. However, with continuous development and utilization of oceans by human beings, the ocean ecological environment has been gradually destroyed, and the marine ecological system has been severely damaged. Marine natural disasters, such as red tides and tsunamis, occurred frequently, bringing substantial economic losses and social impacts. Therefore, it is urgent to protect the ocean environment. Ocean environmental monitoring is an important link in protecting the marine environment. Developing the technology of ocean environmental monitoring is of great significance for the early predicting of marine disasters, preventing and reducing harmful disasters [1,2].

Recently, increasing attention has been paid to the autonomous and in situ ocean monitoring. The primary parameters of ocean environmental monitoring include nutrients (e.g., nitrate, phosphate, silicate), pH, DO, and heavy metals [3–9]. Traditional ocean environmental monitoring mainly relies on manual processing of a large amount of seawater samples and laboratory analysis, which consists of multiple steps (e.g., sampling, transport, pretreatment, instrument analysis) and is rather costly and time-consuming. Besides, the obtained results may be inaccurate since the seawater samples could undergo unexpected reactions during the long-time operation. The relatively low representative and timeliness make it challenging to meet the requirements of rapid detection of marine indicators. Moreover, bulky and expensive high tech instruments and professional personnel are required for traditional marine monitoring. As a result, it is difficult to support ocean environmental monitoring for early warning and management.

Optofluidics is a relatively new technology field that synergistically integrates the optics and microfluidics; it provides many particular characteristics for enhancing the sensing performance and the minimization of systems [10–16]. Fluid is the natural carrier of various micro-nano particles (including chemical macromolecules and biomolecules). Together with the fluid, particles can directly flow into the optofluidic system, where chemical reactions can complete quickly. Optofluidics has the potential to easily detect various optical parameters with high sensitivity and accuracy, such as refractive index, fluorescence, and absorbance. It avoids the complicated processes in traditional detection methods. Besides, the most favored materials for optofluidic chips are polydimethylsiloxane (PDMS) [15] and polymethyl methacrylate (PMMA), both of them are cheap and easily replaceable. These advantages of high integration, sensitivity and accuracy, and low cost make the optofluidic technology widely used in environmental monitoring [5,17] and biochemical sensing.

The development of an autonomous and in situ ocean environmental monitoring system is of high priority in oceanographic research [18]. Optofluidics bring new ideas for the miniaturization of detection systems and has been increasingly considered as a powerful technology to realize ocean environmental monitoring. At present, although a few integrated systems based on optofluidic platform have been reported, a detailed review on its applications on ocean environmental monitoring is absent. Here, we describe the optofluidics-based autonomous monitoring of nutrients, pH, DO, and heavy metals in oceans, with a focus on their principles, sensing properties, advantages, and disadvantages compared with other reported ocean monitoring methods. Much attention has been paid to the most commonly used colorimetric/spectrophotometric detection method, as it has the capability of integrating all of the processes involved in the analysis into a small chip. At last, we forecast that an autonomous and in situ system based on optofluidic platforms will play important roles in the development of ocean environmental monitoring.

#### **2. Nutrients**

Soluble inorganic nitrogen, phosphorus, and silicate in seawater are essential nutrients for the survival of marine organisms. A moderate amount of nutrients in seawater promotes the growth of biology and microorganisms, while inadequate nutrients restrict the growth of phytoplankton and excessive nutrients are prone to cause eutrophication and even further lead to harmful algal blooms, extreme depletion of DO, and even death of aquatic organism [19–21]. Generally, the nitrite concentration in seawater is very low and stable. A too high concentration of nitrite or dramatic changes of nitrite concentration often indicate changes in the ocean environment. Phosphate is an important indicator of eutrophication, and the distribution of silicate affects the community structure of planktonic algae. Accurate quantification of such nutrients in oceans is essential for comprehending the dynamics of marine ecosystems and forecasting the occurrence of harmful red tides [5].

Various techniques, such as colorimetry [22–34], chemiluminescence [35,36], fluorimetry [37–40], electrochemistry [41–43], and chromatography [44–46], have been proposed for nutrient determination. Among them, the colorimetric method using chromogenic agents is one of the most favored detection approaches due to its stability, excellent detection limits, simplicity, high cost efficiency, and analytical feasibility. It operates by adding a color-developing agent to samples, and the absorbance spectrum of the product solution is proportional to the concentration of the nutrient based on the Beer–Lambert law:

$$\mathbf{A}(\lambda) = -\text{lg}\left(\frac{\mathbf{I}}{\mathbf{I}\_0}\right) = \varepsilon(\lambda)\text{cl} \tag{1}$$

where, A(λ) is the absorbance of the solution at a wavelength λ, I0 is the intensity of the initial monochromatic light, I is the transmitted intensity of monochromatic light, ε(λ) is the molar absorption coefficient, c is the concentration of the analyte, and l is the length of light path. To improve the reliability and reduce cost, the discrete and auxiliary optical devices, such as light-emitting diodes (LEDs) and photodetectors, are usually used to construct the autonomous detection systems.

#### *2.1. Nitrate and Nitrite*

Among various colorimetric assay chemistries, the Griess assay method has been the mainstay for nitrite analysis for over a century [22–34]. The mechanism of the Griess assay method is that under acidic conditions, nitrite reacts with sulfanilic acid to produce a diazonium salt, which is then coupled to *N*-(1-naphthyl) ethylenediamine (NED), resulting in pink azo compounds that can be detected at 543 nm. For nitrate determination, nitrate should be reduced to a more reactive nitrite by copperized cadmium [6,23,47], zinc [25], vanadium chloride [7,48], etc.

Figure 1 shows several examples of optofluidic chips and assembled systems using the Griess method for nitrate and nitrite detection. Beaton et al. firstly reported a new generation of miniaturized, in situ, and stand-alone systems with sufficient stability as well as analytical performance for the determination of nitrate and nitrite in natural waters based on optofluidics [33]. As shown in Figure 1a–c, the optofluidic platform consists of a tinted PMMA substrate (5.0 mm thick) micromilled with three absorption cells, three pairs of green LEDs (525 nm), and high sensitivity photodiodes for the detection of nitrate/nitrite with different concentrations. Dark PMMA was used for reducing the background light of LED reaching the photodiode. The optofluidic platform integrated with a custom electronics package for the operational control, data collection, analysis, and transfer. The whole monitoring system has a small size (100 mm diameter and 200 mm height), and low power consumption (about 1.5 W). Autonomous and in situ determination of nitrate and nitrite was deployed with this integrated system, the linear range is up to 350 μM, and the limit of detections (LODs) are as low as 0.025 μM and 0.02 μM for nitrate and nitrite, respectively, making it suitable to be applied in almost all natural waters.

**Figure 1.** The optofluidic platforms and integrated devices used for the determination of nitrate and nitrite. (**a**–**c**) are the flow diagram, the corresponding photograph, and autonomous integrated device of the microanalyzer used to determine the total nitrate and nitrite. (**d**) The optofluidic chip and the photographs of the autonomous integrated device designed by Cisneros et al. Images reproduced from [6,33].

Although the nitrate and nitrite detection systems based on optofluidic technology have been successfully applied in marine online monitoring, there are some drawbacks to be overcome. For example, optofluidic chips are prone to having the problems of bubble formation and blockages due to the narrow microchannels. Martinez-Cisneros et al. developed a compact and automated lab-on-a-chip (LOC) device that integrated microfluidic platform, a highly sensitive colorimetric detection module, and electronics platform for online determination of nitrate and nitrite, as shown in Figure 1d [6]. An embedded and monolithic microcolumn, which was internally installed with copperized cadmium granules, was used as a sample pretreatment module for the reduction of nitrate to nitrite. The device also integrated a bubble removal structure to minimize blockages and avoid potential interferences caused by bubbles, and was successfully applied to the continual monitoring of nitrate and nitrite.

All of the above systems for nitrate detection employ the indirect method, i.e., the concentration of background nitrite is firstly obtained directly after its reaction with the chromogenic agents, the nitrate is then reduced to nitrite, and the second measurement of the total nitrate and nitrite is conducted. The concentration of nitrate finally could be obtained by subtraction. This indirect and time-consuming method adds complexity to the monitoring system. Cogan et al. developed a highly integrated system for nitrate determination. The system was based on the direct and simplified chromotropic acid reagent method, which eliminates the reduction step of nitrate to nitrite [30], as shown in Figure 2. The principle is that the chromotropic acid reacts with nitrate ions in a sulphuric acid medium, resulting in yellow products that can be detected at 430 nm. With this method, a detection range for nitrate from 0.9 to 80 mg/L was achieved with a LOD of 0.73 mg/L. The advantages of simplicity, low cost, low reagent consumption, compact design, and high sample throughput make it an ideal candidate for applying in in situ detection of nitrate.

**Figure 2.** (**a**) Assembly diagram of microchip layers. (**b**) The chip layers that are fully assembled and the micro-cuvette. (**c**) The PDMS microfluidic chip. (**d**) Setup of the microfluidic chip and the optical detection module with LED and photodiode. Images reproduced from [30].

The traditional colorimetric analysis for nitrate and nitrite detection is time-consuming as it requires determining the calibration curve of the system, which means that a couple of standard samples need to be prepared and detected firstly. To overcome this drawback, some modified approaches based on the Griess assay have been proposed. Recently, Shi et al. demonstrated a robust differential colorimetric method for nitrite detection [24], as shown in Figure 3. The nitrite samples and Griess reagent were pumped into the designed tree-like network on a microchip made from PDMS, forming a unique concentration profile as well as a color gradient network, which contained sufficient information for the determination of nitrite. Only one sample is required for this differential colorimetric method, leading to less consumption of both time and power. The measuring errors for the used two nitrite solutions (0.50 mM and 0.33 mM) are 1.16% and 0.50%, respectively. Compared with the classic method that required a calibration curve, the stability and accuracy of the system are improved by approximately ten times and six times, respectively.

**Figure 3.** (**a**) Schematic diagram for forming the bidirectional differential concentration. (**b**) Optofluidic platform setup. (**c**) Left: optical detection system with green LEDs and CCD image sensor. Right: physical diagram of the integrated system. Images reproduced from [24].

#### *2.2. Phosphate*

A variety of approaches have been used for the determination of phosphate in seawater, including colorimetric detection, fluorescent detection, and electrochemical detection. All of these are reagent-based methods, as phosphate cannot be detected directly. Generally, autonomous systems tend to employ the colorimetric method rather than fluorescent or electrochemical methods [49,50].

Legiret et al. reported a high-performance optofluidic system for marine phosphate determination using the vanadomolybdate method [51]. The basic principle of this method is that under acidic conditions, orthophosphate reacts with vanadium polymolymolybdate reagent directly and rapidly, resulting in a stable yellow product that can be detected at 375 nm [52,53]. Figure 4a–c shows the optofluidic chip and the assembled system with this approach. The system incorporated an optofluidic chip in tinted PMMA, an optical detection module with high power UV-LEDs as a light source, and photodiodes as absorbance detectors, an embedded control electronics and syringe pumps. The integrated optofluidic analyzer has a small physical size of 22 cm (height) by 10 cm (diameter). Experimental results showed that it has a wide linear range from 0.1 μM to 60 μM and a LOD of 52 nM for phosphate detection. This integrated microanalyzer features high measurement accuracy and resolution, while requires low power and reagent consumption, and it has been successfully deployed in autonomous and in situ phosphate monitoring.

*Micromachines* **2020**, *11*, 69

Among all reagent based colorimetric approaches for the determination of phosphate, phosphomolybdenum blue method is the most widely used one [5,51–56]. Its principle is that under acidic conditions, orthophosphate reacts with ammonium molybdate, producing a light yellow ammonium phosphomolybdate, which is then reduced by the reducing agent (e.g., ascorbic acid) into a blue compound with strong chromogenic capacity—'molybdenum blue'. Duffy et al. reported a portable, compact lab-on-a-disc device based on the phosphomolybdenum blue method for in situ quantitation of water phosphate [52]. As shown in Figure 4d,e, the integration of a microfluidic disc allows the use of an optical path length as long as 75 mm for improving sensitivity. The device also integrated a low-cost optical detection system formed by a pair of LEDs and photodiodes. The total mass and the physical size of the device is 2 kg and 20 cm × 18 cm × 14 cm, respectively. The experimental results showed that the detection range for phosphate is 14–800 μg/L, and the LOD is as low as 5 μg/L. Grand et al. reported a newly developed LOC analyzer for long-term and in situ phosphate monitoring [56]. To obtain the best analytical sensitivity, the influence of sample temperature and salinity on the reliability and accuracy was studied, and the reaction parameters were evaluated and optimized. The analyzer is a viable candidate to be deployed in ocean phosphate monitoring as it owns many merits such as perfect stability and robustness, small size, simple to operate, and low energy consumption (1.8 W). It also features a LOD of 30 nM and a wide linear range from 0.05 to 13 μM, with a precision less than 10%.

**Figure 4.** The optofluidic chips and assembled system used for phosphate detection. (**a**–**c**) Flow diagrams, the corresponding photographs, and the autonomous integrated devices of the microanalyzers used to determine the total phosphate ion concentrations. (**d**,**e**) The rendered images showing each layer of the microfluidic disc and the phosphate sense system. Images reproduced from [51,52].

For many optofluidic autonomous monitoring systems based on the colorimetric method, the LOD is not comparable to the results obtained by traditional methods because of the limitations on the size of the absorption cell on microchips. To enhance the phosphate absorption and obtain better measurement accuracy, Zhu et al. proposed an innovative on-chip Fabry–Pérot microcavity, which consists of two parallel reflectors made by a pair of fiber facets coated with Au films (Figure 5) [5]. Light is reflected many times in the microcavity, thus lengthening the optical path and enhancing

absorption. The length of the absorption cell is shortened from several centimeters to 300 μm, and the device still features a LOD as low as 0.1 μM for phosphate detection. Combining the specially designed passive microreactor with semilunar barriers for rapid and enough chromogenic reaction, the required time for detection is shortened from several minutes to six seconds.

**Figure 5.** A Fabry–Pérot resonator-based optofluidic device used for monitoring of phosphate in oceans. (**a**) Setup of the optofluidic device. Microchannels with crescent batteries are designed for rapid and enough mixing of regents. (**b**) The microcavity is formed by a pair of optical fibers, on which surfaces are coated with gold. Images reproduced from [5].

#### *2.3. Silicate*

Silicomolybdenum yellow method and silicomolybdenum blue method are the two standard methods that are commonly used for detection of silicate in seawater. The former method is rapid but features poor sensitivity and is easily interfered with by salinity, making it unsuitable for low concentration analysis. The silicomolybdenum blue method is widely used for the determination of marine silicate as it has the advantage of high sensitivity. Its mechanism is based on the reaction of silicate with ammonium molybdate to form a yellow silicomolybdate complex, which is further reduced to a stable and detectable blue silicomoIybdenum product by ascorbic acid. Cao et al. reported a new generation of optofluidic sensor that has been successfully applied for analyzing the silicate in seawater [57]. The silicate sensor contained four functional modules, including the fluid driving system, a microchip made of tinted PMMA for the mixing and reaction of reagents, a control circuit, and a sensor peripheral. Three absorbance cells were designed for a wide detection range, and three pairs of 810 nm LEDs and photodiodes were used accordingly to perform spectrophotometry analysis. The LOD of the sensor was as low as 45.1 nM, and the linear determination range of the sensor was 0 to 400 μM. An offshore experiment proved that this optofluidic sensor has the advantages of low regent consumption, high accuracy, and robustness.

#### **3. pH**

Human activities produce a large amount of carbon dioxide (CO2). CO2 released into the atmosphere could be absorbed by the ocean, leading to significant chemical changes like ocean acidification. The absorption of CO2 also regulates the pH of seawater, which is a key parameter for the aquatic organism and influences the ecosystems and biogeochemical cycles of ocean [8,58–62]. A water environment with pH from 6.5 to 8.0 is required for aquatic life, the quantity of water-based life reduces outside of this range because the physiological systems that organisms lived by are affected. Toxic heavy metals (Cd, Pb, Hg, etc.) become more soluble at low pH, thus increasing the toxicity levels of the living environment for organisms [62]. As a result, accurate and timely determination of pH is dispensable for fully understanding the marine carbon cycle and changes of the ecosystem [4,63,64].

Glass electrode method is one of the most popular methods used for pH measurement; accurate results could be achieved if the glass electrodes are maintained and operated properly. However, the glass electrode method also has several drawbacks, such as physical fragility, leakage of the reference electrode buffer, and various responses of the glass electrode with salinity and temperature. All of these disadvantages limit its applications for long-term ocean monitoring. The principle of the most commonly used colorimetric detection of pH is adding indicator dyes, such as thymol blue [65,66], phenol red [67], cresol red [68], and meta-cresol purple [69] to seawater samples, resulting in colored compounds that indicate the pH value.

There are a few examples of optofluidic devices for in situ colorimetric pH measurements. Florea et al. developed a low cost and accurate optofluidic device for pH determination based on polyaniline (PAni), the detectable range for this device is from pH 2 to 12 [70]. The device integrated PAni-based coatings and spectrophotometry method to measure pH. The absorbance spectrum of PAni coatings changes when solutions at various pH values flow along the microchannel. This microfluidic sensor requires no more indicator reagent, thus reducing the complexity of pH detection. Besides, this work evaluated the feasibility of using a digital color camera (e.g., mobile phone with integrated digital cameras) instead of a spectrophotometer to perform colorimetric analysis, which will dramatically extend its applications [71,72]. However, the system is not fully autonomous and requires manual input, like taking photographs and sampling.

More fully developed autonomous systems for seawater pH measurement have been reported. Rérolle et al. reported a colorimetric pH sensor based on optofluidics for autonomous ocean monitoring. The pH indicator dye of thymol blue was employed in this case [66]. The optofluidic chip is made of tinted PMMA, it consists of a static mixer formed by a serpentine channel as long as 2.2 m, and an absorption cell with a size of 700 μm (length) by 700 μm (width). Its principle is that samples and indicator dye solutions were effectively mixed in the static mixer, then the mixer flows into the absorption cell for optical detection. The detection module consists of a light source with a tricolored LED and a detector with a linear array photodiode spectrometer. The system was successfully implemented in shipboard deployment and a precision of 0.001 pH unit for short-term pH monitoring could be obtained. Lai et al. designed an autonomous optofluidic chemical analyzer for both pH and partial pressure of CO2 (pCO2) [17]. A pH indicator was used for the determination of the pCO2. The indicator was pumped into the gas permeable membrane and then given an optical response when reacting with pCO2. With the beam-splitter design shown in Figure 6, the analyzer has a precision of ~±0.5 μatm for pCO2 and ~±0.0005 pH for pH detection. With the specially designed regent bag and adequate power supply, over 8500 measurements could be obtained for every scuba diving. However, the primary defect of this optofluidic analyzer is that it is more complicated compared to other pH and pCO2 sensors, such as the single-ended electrode or optode sensor.

**Figure 6.** (**a**) Schematics of the detection system for SAMI-CO2. (**b**) Schematics of the detection system for SAMI-pH. Images reproduced from [17].

#### **4. DO**

Dissolved oxygen is an indispensable substance for the aquatic organism and a vital parameter to characterize the metabolism of marine ecosystems [73–75]. The DO content in seawater is influenced by biochemical and physiological activities. As the DO content of polluted seawater is lower than that of natural seawater, the determination of DO also helps to evaluate the hypoxia and pollution in marine environments.

Many methods have been used to detect DO in oceans, including the iodometric titration method, which is internationally recognized as a benchmark [76], the most widely used electrochemical method, and the optical method, which is mainly based on the principle of fluorescence quenching. Each method has its own merits and drawbacks. The iodometric titration determination has the advantage of high accuracy, but it features cumbersome detection procedures and is not suitable for continuous online detection [77]. Electrochemical method characterizes for rapid detection and simple operation, but it has limitations such as the requirements for calibration and regular maintaining, and the electrode is easily poisoned. The optical DO sensor based on fluorescence quenching has the advantages of versatility and high sensitivity, as well as low toxicity, but the detection accuracy is still needed to improve [78].

Innovative research on the applications of optofluidics in DO detection have been reported. Mahoney et al. proposed a multilayer optofluidic device based on measuring the fluorescence quenching in a Ruthenium-based oxygen sensitive dye. Enhanced sensitivity was achieved by employing total internal reflection (TIR) of the excitation light. The optofluidic sensor exhibits high sensitivity for the detection of 0–20 ppm DO in water [79]. However, the automation and miniaturization of DO sensors for ocean monitoring based on optofluidic technology remain to be developed [80,81].

#### **5. Heavy Metals**

It is of great importance to monitor heavy metals in oceans as they can seriously affect the environment and human health. Generally, heavy metal refers to metals with specific gravities greater than 5 g/cm3. It includes essential metals that are indispensable for the normal physiological activities of organic life such as iron (Fe); copper (Cu); magnesium (Mg); selenium (Se); zinc (Zn); manganese (Mn); nonessential metals including cadmium (Cd), mercury (Hg), silver (Ag), lead (Pb), and gold (Au); and some unusual metals with high atomic weight [82]. With the accumulation of essential metals in organisms, toxic side effects will occur if its concentration exceeds a certain threshold [83]. Nonessential metals could be hypertoxic even at a trace level, although they do not participate in the metabolic activities of organisms.

Many analytical techniques have been proposed for the determination of heavy metals in seawater, for example, the inductively coupled plasma mass spectroscopy (ICP-MS) method and atomic absorption spectrometry method. However, both of them require tedious sample preparation and pretreatment, expensive equipment, and professional personnel, making them insuitable in applications for autonomous, in situ, and continuous monitoring of heavy metals. In this case, ocean monitoring sensors based on colorimetric, fluorescent, and chemiluminescent methods appear as promising technologies due to the high sensitivity and feasibility, versatility, and reproducibility [84–95], and all of these methods have been truly applied for online analysis of heavy metals in seawater.

The colorimetric method for heavy metal monitoring is also based on Lambert–Beer's law. The principle is that under certain conditions, heavy metal ions react with a specific reagent, producing a new colored chemical solution, and the absorbance of the solution is related to the concentration of the heavy metal. Different heavy metals need different color developing agents when employing the colorimetric method. For example, silver salt is used for arsenic determination [96], dithizone is generally used for lead and zinc determination [97], while dimethylglyoxime is generally applied for nickel determination [98]. Milani et al. developed an autonomous LOC colorimetric system with high performance and low-cost for the in situ monitoring of dissolved Fe(II) and Mn in natural water [88]. The optofluidic platform consists of a microfluidic chip made of PMMA and two pairs of LEDs and photodiode detectors for detection of Fe(II) and Mn, as shown in Figure 7a,b. With this portable device, 12 samples of Fe(II) and 6 samples of Mn could be analyzed per hour with LODs of 27 nM and 28 nM, and precisions of 2.1% and 2.4% (*n* = 19), respectively. Lace et al. reported an innovative optofluidics system for monitoring of arsenic in water with the leucomalachite green dye method [91]. The principle involves the reactions of arsenic and potassium iodate under acidic conditions, and leucomalachite green is oxidized to malachite green by the liberated iodine, producing a green color with an absorption peak at 617 nm. Chromogenic agents and water samples are mixed and reacted in a microfluidic chip, as shown in Figure 7c. The leucomalachite green method was optimiszd and water arsenic within the range of 0.3–2 mg /L could be detected, the LOD of the system was 0.32 mg/L, and the average relative standard deviation (RSD) was 21.1%.

The fluorescent method is an excellent option for autonomous and in situ ocean heavy metal detection. Compared with the colorimetric method, higher sensitivity and lower LOD could be achieved with the fluorescent method. The basic mechanism of the fluorescent method is that the fluorescence parameters (such as fluorescence intensity, lifetime, and spectrum) change in response to the concentration variations of some ions. Until now, numerous fluorescent sensors based on chelation-enhanced fluorescence [98,99] photo-induced electron transfer [92], aggregation-induced emission effect [100], and intramolecular charge transfer [101] have already been reported for heavy metal monitoring. Leray's group developed several optofluidic devices that incorporated fluorimetric detection for the monitoring of heavy metals such as aluminum and cadmium [92,93,102]. Recently, the aluminum-sensing mechanism by PSSA–4-propoxysulfonate salicylaldehyde azine was proposed [92]. As shown in Figure 7e–g, the sensing system includes a microfluidic chip made from PDMS/glass, and a fluorescence detection platform consists of a LED (365) nm and a photomultiplier (PMT) detector. In brief, the PSSA and Al(III) are injected and mixed in a Y-shaped junction, the complexation of the ligand with the analyte occurs upon mixing, producing a fluorescence signal, which is collected and detected by the PMT. Regularly monodispersed and spaced microdroplets are formed, and a highly reproducible and reliable monitoring environment was provided as the system to adopt the microfluidic droplet technology. The experimental results showed that the system has a LOD of 153 nM. The newly designed fluorescent sensor for cadmium detection adopts a commonly used water-soluble commercial dye Rhod-5N, whose fluorescence intensity increased with the existence of toxic cadmium. The optofluidic device is made from PDMS chip and a glass substrate, as shown in Figure 7d. The famous staggered herringbone structure [98,102] was used to form a passive mixer for the thorough mixing of the Rhod-5N reactant and the Cd2<sup>+</sup> analyte. The optofluidic device obtained a promising LOD of 0.45 μg/L in 3-(*N*-morpholino) propanesulfonic acid buffer at pH 7.

**Figure 7.** Spectrophotometric and fluorescent detection of heavy metal by optofluidic systems. (**a**,**b**) The photograph and the autonomous integrated device of the microanalyzers used to determine dissolved Fe and Mn with a spectrophotometric method. (**c**) The PMMA optofluidic chip used for arsenic detection with a colorimetric method. (**d**) The photograph of the optofluidic chip for fluorimetric detection of cadmium. (**e**) The scheme of the optofluidic set-up following the droplets approach for fluorescent detection of water-soluble aluminium. (**f**,**g**) The photographs of the microchip and the forming droplet of the fluorescent sensor. Images reproduced from [88,91,92,102].

The chemiluminescent method for quantitative analysis of heavy metals is mainly based on the linear relationship between the concentration of the analyte and the chemiluminescence intensity of the system under certain conditions. The most commonly used chemiluminescent reagents include luminol, acridine esters, and 1,2-dioxygen cycloethane. Based on the chemiluminescent reaction of Mn contained in the seawater sample with a luminol-based reagent, Christophe et al. developed an integrated in situ analyzer for detecting Mn (IISA-Mn) in the environment of the deep sea [89]. A miniaturized PDMS device with Tesla mixer structures was used to enhance the mixing and reaction efficiency, and a PMT detector was used to measure the chemiluminescent intensity. As a result of the integration of fluidic elements such as mixers, valves, and flow regulators, this analyzer based on optofluidics occupies less volume (3 L) and consumes less reagent (<130 mL/24 h) compared with its macroworld counterparts. The IISA-Mn gives a linear range of 0 to 500 nM and a LOD of 280 nM in seawater, making it sufficient for detecting geochemical anomalies such as hydrothermal ore deposition surveys. The system has been proven to be able to work flawlessly and continuously during the 8 h of an actual remotely operated vehicle dive.

#### **6. Discuss and Outlook**

There is more and more research on the optofluidic devices as optofluidic technology has advantages of low cost, small size, low reagent, and power demands. Although we emphasize the applications of optofluidic platforms for the monitoring of chemical parameters in ocean environment in this review, they could also find uses in biological monitoring in the marine ecosystem. For example, a variety of in situ bio/biochemical analyzers (e.g., adenosine triphosphate analyzer and gene analyzer) have been developed on optofluidic platforms, most of them have been evaluated in the real ocean environments including the deep sea [103,104].

As marine environmental pollution is increasingly severe, there is a growing demand for the further development of autonomous and in situ ocean monitoring. To realize continuously offshore monitoring with high practicability, stability, and credibility, smaller and lighter portable instruments with less power consumption and higher accuracy are needed. Optofluidic technology is one of the most promising approaches to realize the miniaturization and automation of ocean monitoring. This paper reviewed the optofluidic devices integrated with microelectromechanical systems that are applied for in situ ocean monitoring; up-to-date developments of optofluidics for autonomous ocean environmental monitoring were covered. Optofluidics has the challenge of ensuring that the overall system can achieve the required sensitivity for ocean applications where analytes exist at only trace levels. However, we firmly believe that based on the development trend of full automation and miniaturization of marine monitoring equipment, and with the development of the supported technology such as three-dimensional printing, inkjet printing, power-free pumping and micromachining technologies, the low-cost lab-on-chip systems based on optofluidic technology will inspire a promising future in the field of ocean observations.

**Author Contributions:** F.W., Y.Y., and J.Z. proposed the original idea, and planned the configuration and key topics. F.W. wrote the manuscript. J.Z., L.C., Y.Z. and X.H. revised the paper for language and quality. All authors took part in regular discussions and were involved in the completion of the manuscript. All authors have read and agreed to the published version of the manuscript.

**Funding:** This work was funded by open financial grants from Qingdao National Laboratory for Marine Science and Technology (No. QNLM2016ORP0410), National Natural Science Foundation of China (No. 11774274), National Key R&D Program of China (2018YFC1003200), and Foundation Research Fund of Shenzhen Science and Technology Program (No. JCYJ20170818112939064).

**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* **Silver Nanoprism Enhanced Colorimetry for Precise Detection of Dissolved Oxygen**

**Yunfeng Zuo 1,2, Longfei Chen 1, Xuejia Hu 1, Fang Wang <sup>1</sup> and Yi Yang 1,2,\***


Received: 18 March 2020; Accepted: 2 April 2020; Published: 4 April 2020

**Abstract:** Dissolved oxygen (DO) content is an essential indicator for evaluating the quality of the water body and the main parameter for water quality monitoring. The development of high-precision DO detection methods is of great significance. This paper reports an integrated optofluidic device for the high precision measurement of dissolved oxygen based on the characteristics of silver nanoprisms. Metal nanoparticles, especially silver nanoprisms, are extremely sensitive to their surroundings. In glucose and glucose oxidase systems, dissolved oxygen will be transformed into H2O2, which affects the oxidation and erosion process of nanoprisms, then influences the optical properties of nanoparticles. By detecting the shift in the plasma resonance peak of the silver nanoparticles, the dissolved oxygen (DO) content can be determined accurately. Great reconfigurability is one of the most significant advantages of the optofluidic device. By simply adjusting the flow rate ratio between the silver nanoprisms flow and the water sample flow, real-time continuous adjustment of the detection ranges of DO from 0 to 16 mg/L can be realized dynamically. The detection limit of this device is as low as 0.11 μM (3.52 μg/L) for DO measurement. Thus, the present optofluidic system has a wide range of potential applications in fields of biomedical analyses and water sensing.

**Keywords:** optofluidics; dissolved oxygen; silver nanoprisms; colorimetry

#### **1. Introduction**

Dissolved oxygen provides the necessary biochemical environment for the survival of aquatic organisms and is an indispensable material for aquatic life activities [1,2]. Dissolved oxygen monitoring plays an important role in aquatic ecosystem quality assessment, aquatic science experiments and aquatic resource exploration [3,4]. Besides, dissolved oxygen is also a key parameter in on-chip biochemical systems and chemical processing [5]. Generally, the three primary testing methods for the determination of dissolved oxygen in water are the iodometric method, the electrode polarography method (so-called Clark electrode method) and the fluorescence lifetime method. The iodometric method has a wide detection range and high detection accuracy [6]. However, the iodometric method has the defects of cumbersome operation, long operation time and high professional requirements for the operator. The electrode polarography method is simple and fast [7,8]. However, the presence of telluride, oil, carbonate and algae in the water sample may cause clogging or even damage to the gas permeable membrane. In the actual measurement process, it requires frequent maintenance. The fluorescence lifetime method is a simple and fast approach [9]. However, the system-building is complicated, and the price is relatively high. Therefore, an economical, efficient, accurate and simple measurement of dissolved oxygen is of great value for meeting the needs of production and scientific research. For this reason, it is of vital importance to develop an accurate detection system for dissolved oxygen monitoring.

Metal nanoparticles in different shapes and sizes have various unique properties [10], which enable them to be applied in diverse fields such as biosensors and nanomedicine [11]. Compared with other kinds of metal nanoparticles, silver nanoparticles have a stronger plasmonic interaction with light [12]. Particularly, triangular silver nanoprisms exhibit more LSPR bands due to their anisotropic morphology than spherical or quasi-spherical silver nanoparticles [13]. The extraordinary properties of silver nanoprisms make it a powerful tool for biosensing and molecule detecting [10,14,15]. Optofluidics is a new interdisciplinary research field that focuses on the amalgamation of optics and microfluidics [16]. This new field provides lots of unique advantages for enhancing the performance and simplifying the design of micro-electromechanical systems [17]. Over the past decades, this new field has been developing rapidly, and it has been applied in many areas such as biosensor [18], water purification [19], optical devices [20,21] and water sensing [22]. Nowadays, the field of soft optical materials has received wide attention [23]. Optical devices based on frontier soft optical materials exhibit high biocompatibility and special qualities [24]. It is promising and expected to broaden the horizon of optofluidics in biomedical utility [25]. Optofluidics integrates multiple disciplines, which can not only realize on-chip sample pretreatment, but also can be used for the rapid detection of various biochemical indicators quickly and accurately [26].

The combination of silver nanoparticles and optofluidic systems can bring new merits. In one way, optofluidics provides a stable condition for mental nanoparticles synthesis [27]. In the other way, silver nanoparticles will strengthen the capability of optofluidic systems for biochemical sensing [28]. Therefore, optofluidics and nanotech are perfect candidates to innovate the measurement of dissolved oxygen. Herein, we designed an integrated optofluidic chip combining nanoparticle synthesis and spectrum detection to realize the precision measurement of dissolved oxygen. Rapid synthesis of silver nanoprisms was realized in a single chip. Owing to the novel physical properties of silver nanoprisms, minute amounts of dissolved oxygen concentration changes in water can be detected. Compared to traditional methods, this optofluidic DO detector possesses advantages such as higher measurement precision (detection limit less than 3.52 μg/L), less sample and reagents consumption (level of microliter), small size, low cost, parallel processing of samples, adjustable detection range (0–16 mg/L), and easy integration. This method can overcome the deficiencies of the traditional equipment in practical applications and has an important application prospect in the field of water environment monitoring.

#### **2. Working Principle and Design**

As illustrated in Figure 1, this integrated optofluidic system based on Ag-nanoprisms etching consists of three major parts: a silver nanoprisms synthesis module, a sample processing module and an optical detection module. The synthesis module was designed to realize the rapid, on-chip synthesis of silver nanoprisms. The geometric and optical features of silver nanoprisms can be controlled by adjusting the flow rate ratio of synthetic components rapidly and continuously. Through fast mixing in the microfluidic channel and precise control of reactant flows, flash chemistry can be achieved to realize rapid synthesis of nanoparticles in a single chip [29,30]. Compared with traditional batch synthesis methods, microfluidic methods process better synthesis efficiency [31]. The second part, samples mixing and processing module, is designed to realize fast and efficient mixing of pretreated water sample and synthesized Ag nanoprisms based on special microchannel design and fluid dynamics. Before being injected into the optofluidic chip, water samples with different DO contents were treated by a H2O2-generating system. In the presence of glucose (GO) and glucose oxidase (GOD), dissolved oxygen in the water sample is converted to H2O2 with concentration directly related to DO content. The reaction process is as follows:

$$\beta \text{ glucose} + \text{O}\_2 + \text{H}\_2\text{O} \xrightarrow{\text{GOD}} \beta \text{ glucose acid} + \text{H}\_2\text{O}\_2 \tag{1}$$

where GOD is glucose oxidase that catalyzes the oxidation of β-glucose quickly and efficiently. The ratio of glucose and oxygen consumption to hydrogen peroxide product is 1:1:1. In this part, the pretreated water sample with generated hydrogen peroxide and Ag nanoprisms solution are injected into the system, and they are fully mixed and reacted in the channel. Micromixer allows reagents to be thoroughly mixed, which leads to controllable and stable reaction conditions. The Ag nanoprisms will be etched due to the presence of H2O2 in the pretreated water sample. The etching process is as follows:

$$2Ag + H\_2O\_2 \to 2Ag^+ + 2OH^- \tag{2}$$

Then, the mixture flows into the third part for optical detection. An optical detecting module combining the absorption cell is designed to capture the extinction spectrum of the eroded silver nanoparticles. The SPR (surface plasmon resonance) peak is extremely sensitive to the morphology of Ag nanoprisms. Different dissolved oxygen concentrations in the water will cause varying degrees of corrosion to the silver nanoparticles and affects the extinction spectrum, causing blue shifts of SPR peak. The detectable SPR shift signal can be used for the quantitative analysis of DO contents in water samples.

**Figure 1.** Schematic diagram of the optofluidic system based on Ag-nanoprisms etching for dissolved oxygen (DO) detection. This chip is consisted of three functional part: on-chip synthesis, sample processing, optical detecting. The detecting mechanism is based on the etching process of silver nanoprisms. The blue shift of the SPR peak of Ag nanoprisms can be used for the quantitative analysis of DO.

In this system, the full mixing of liquid reaction components or water samples and reagents is critical. Mixing efficiency affects not only the synthesis of nanoparticles, but also the accuracy of the system. Micromixers based on microfluidics are fundamental lab-on-a-chip components, which can realize the mixing of fluids within millisecond time scales to microsecond time scales, and have been applied in many fields [32]. Generally, micromixers can be divided into active and passive mixers. Active mixers rely on an external field, such as acoustic waves [33], to enhance the mixing and can achieve relatively high mixing efficiency [34]. Passive mixers realize fluid mixing based on structure design of the microchannel. Passive mixers are relatively simple and easy to integrate [35]. Among them, zigzag microchannel and microcylinders are usually applied to design passive micromixers based on chaotic fluid transformations [35,36]. Here, a passive micromixer based on a high-density Z-type (zigzag-type) hybrid structure was introduced, while microcylinders were designed in the Z-shape

microchannel to accelerate the liquid mixing efficiency, ensuring that the reactants or samples were thoroughly mixed and reacted with each other quickly. Liquids mixing in microchannel with different designs were simulated by finite element method (FEM), as shown in Figure A1 (see Appendix A). According to the simulation, the Z-shaped channel with cylinders enables the high-speed mixing of liquids.

According to the chemical equation, the oxidization of glucose consumes O2 and generates H2O2. When the concentration of glucose is insufficient to consume all the O2 in sample water, the conversion degree of dissolved oxygen increases with glucose content. When the concentration is sufficient, the SPR peak shifts will remain the same. Under natural conditions, the concentration of DO is far less than 20 mg/L. To realize the precise detection of DO in water, a sufficient or excess amount of glucose is necessary; 20 mg O2 needs at least 0.625 mmol glucose. The content of glucose oxidase (GOD) is relevant to the reaction efficiency and reaction time. In consideration of the response time of the whole optofluidic system and the contents of DO in actual water samples, sufficient amounts of GOD solution (5 mg/mL) and glucose solution (2 mM) were added to the water sample with the ratio of 1:1:1. To investigate the stability of sliver nanoprisms and the assay system, the effects of detection reagents, glucose, GOD, were tested. As shown in Figure A2 (see Appendix A), no detectable SPR shift was observed, showing that GO and GOD formed a stable system with Ag nanprisms.

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

#### *3.1. Fabrication*

The optofluidic device was fabricated by polydimethylsiloxane (PDMS) using soft-lithography processes. First, a 50-μm layer of SU8 photoresist (SU8-2050, Microchem, Westborough, MA, USA) was spin-coated onto a silicon wafer. After pre-baking, the master was exposed to UV light under a glass mask using a mask aligner (H94-37, SVC, Chengdu, China). Then, through the process of post-exposure bake, development and hard bake, a thick, chemically and thermally stable image was constructed on the silicon wafer as a mold. Subsequently, the mold was covered with PDMS (Sylgard 184, Dow Corning, Midland, MI, USA) and heated at 75 ◦C for 1 h for polymer curing. Then the cured PDMS with microchannel pattern was removed from the silicon wafer and bonded onto a glass substrate after plasma oxidation (PDC-002, Harrick Plasma, Ithaca, NY, USA). Then, the fabricated PDMS chip was stored in an oven at 75 ◦C for 30min to increase the bonding strength. After inserting fluidic tubing and optical fibers into the microfluidic chip, the optofluidic chip was fabricated. Reagent flow streams were injected by using micropumps (LSP01-2A, LongerPump, Baoding, China). The width of the inlet and mixing channel was 100 μm, and the width of the main channel for the reaction was 300-μm. The height of the microchannel is 50 μm. A large image was obtained by inverted microscope (Ti-E, Nikon, Tokyo, Japan) and microscope image stitching technology (NIS-Elements AR, Nikon, Tokyo, Japan) to illustrate the fabricated optofluidic device, as shown in Figure A3a (see Appendix A). Figure A3b shows the microscopes of the channel design at the positions of the inlets, zigzag-channel, microcylinders and absorption cell.

#### *3.2. Fast Mixing*

The performance of micromixers in the optofluidic systems is related to the control of Ag-nanoprisms synthesis and the reaction process between regents. To visualize the process of fast mixing in the microchannel, laser scanning confocal microscopy (A1R, Nikon, Tokyo, Japan) was used to capture the three-dimensional distribution of liquids in the microchannel. The core flow and sheath flows were dyed with Rhodamine B and Rhodamine 6G, respectively. The core flow emitted red fluorescence, and the sheath flows emitted green fluorescence. Figure 2a,b show the confocal images of liquids mixing process in the microchannel. The flow rates of the three liquid flows were 20 μL/min, respectively. The Z-shaped channel with cylinders induced the secondary flows in microchannel and increased the mixing efficiency [36]. Figure 2c,d show the intensity distribution profiles of the dotted

lines in Figure 2a,b. It was obvious that the liquids were thoroughly mixed in the Z-shape channel. This microchannel design can realize the quick mixing of liquid reagents at a low flow rate, and the mixing time is about 40 ms.

**Figure 2.** Mixing process of regent flows in microchannel. (**a**,**b**) Images of the three-dimensional liquids distributions capture by laser scanning confocal microscopy before and after the mixing process. The flow rates are fixed at 20 μL/min for each reagent inlet. (**c**,**d**) Cross-sectional liquids distribution corresponding to (**a**,**b**).

#### *3.3. On-Chip Synthesis of Silver Nanoprisms*

The silver nanoprisms were synthesized real-time and dynamically in the optofluidic chip according to a standard synthetic procedure at room temperature and in a neutral pH environment. [37]. Briefly, there are four fluid inlets in the synthesis module, three of which (i1, i2, i3) are at the front of the efficient micro-mixer. Another inlet (i4) was designed at the middle of the mixer. Silver nitrate (AgNO3, 0.4 mmol/L), sodium citrate (4 mmol/L), and H2O2 (0.6 wt%) were injected into the optofluidic device through inlet i1, i2, i3 with different flow rates (Qi1 = 20 μL/min, Qi2 = 30 μL/min, Qi3 = 10 μL/min), respectively. These three streams were fully and quickly mixed as soon as they flowed through the micro-mixer. Then sodium borohydride (NaBH4, 4 mmol/L) was injected through inlet i4 with flow rate at 20 μL/min.

Through the synthesis module, on-chip synthesis of silver nanoprisms was realized, the geometric and optical features of silver nanoprisms can be controlled by adjusting the ratio of synthetic components rapidly and continuously. Silver nanoprisms with various diameters and SPR peaks can be synthesized through lab-on-chip systems, shown in Figure 3a. Figure 3b–e shows the TEM (Transmission Electron Microscope) micrographs of the Ag nanoprisms under different synthesis conditions. The TEM images were obtained using a JEM-2010 HT transmission electron microscope. Various shapes of silver nanoprisms were synthesized using the microfluidic method by simply changing the flow rate ratio between each regent (Qi1:Qi2:Qi3:Qi4 = 3:3:1:2, 3:3:1.5:2, 2:3:1:2), as shown in Figure 3b–d. Figure 3e shows the batch synthesis of silver nanoprisms in macroscale with the same reagents concentration as that in Figure 3d. The particle size distribution of the Ag nanoprisms synthesized by means of microfluidics and batch approach were studied, as shown in Figure A4 (see Appendix A). Owing to fast mixing in the microchannel and the precise control of reactant flows, the silver nanoprisms synthesized through the microfluidic method possess excellence reproducibility and narrow size distribution compared with batch synthesis method. The resultant Ag-nanoprisms solution was incubated for 24 h at room temperature before use, in order to eliminate the influence of borohydride, hydrogen peroxide and air bubbles. Then, the Ag nanoprisms solution was injected into the sample processing and optical detecting modules with an adjustable flow rate for DO measurement application.

**Figure 3.** Synthesis of nanoprisms with different diameters. (**a**) The photograph of AgNPs synthesized through the lab-on-chip systems. TEM micrographs of the Ag nanoprisms under different synthesis conditions. Microfluidic synthesis with different flow rate ratios: Qi1:Qi2:Qi3:Qi4 = (**b**) 3:3:1:2, (**c**) 3:3:1.5:2, (**d**) 2:3:1:2. (**e**) Batch synthesis in macroscale.

#### *3.4. Procedures for Dissolved Oxygen Sensing*

To begin with, water samples with different dissolved oxygen concentrations were pretreated by injecting GOD solutions (5 mg/mL), glucose solution (2 mM), and oxygen-free water. The water sample with different DO contents and oxygen-free water were prepared through the aeration process by ultra-pure nitrogen and oxygen based on Henry's law. Besides, all the reagents were prepared in a nitrogen glove box by using oxygen-free water and sealed hermetically before use to minimize the impact on detection accuracy. The pretreatment process was accomplished in a Z-shape microchip. The flow rate ratio between sample water, glucose, GOD and oxygen-free water was fixed at 1:1:1:7. Then, the mixed solutions were stored into an airtight glass syringe and incubated at room temperature for 20 min. According to the chemical reaction formula, DO was transformed into H2O2. Then, the post-treated water samples and Ag nanoprisms solution were injected into the optofluidic chip with different flow rate ratio (*r* = *QNPs*/*QW*S) according to the concentration of DO content. Under the condition of different flow rate ratio, the total flow rate was fixed at 60 μL/min. The reaction mixture solutions were mixed and reacted along with the microchannel. H2O2 etches the Ag-nanprisms to silver ions. Then the solution flowed through the absorption cell. A supercontinuum whitelaser (WhiteLase SC400, Fianium, Southampton, UK) was applied to be the light source for colorimetric, and the optical signal was received and transmitted by a multimode optical fiber. Then, the SPR spectra of the mixtures were measured by a spectrograph system including a CCD camera (Newton 920, Andor, Oxford, UK) with Andor's line of Shamrock imaging spectrographs (Shamrock 303i, Andor, Oxford, UK).

Figure 4 shows the images and spectra of the Ag nanoprisms illuminated by supercontinuum whitelaser. Scattering images of silver nanoprisms before and after DO measurement (50 μM, 1.6 mg/L) are shown in Figure 4a,b. Their corresponding normalized extinction spectra captured through the optical detection module are shown in Figure 4c,d. The flow rates of post-treated water samples and Ag nanoprisms solution were fixed at *QWS* = 30 μL/min and *QNPs* = 30 μL/min, respectively (*r* = 1). The spectra of the sample flow were collected 5 times, 10 s apart. By detecting the SPR peak shifts, the concentration of dissolved oxygen in the water will be determined. TEM was employed to characterize the morphological transition of Ag nanoprisms after DO sensing. The shape of Ag nanoprisms changed owing to the etching, as shown in Figure A5 (see Appendix A).

**Figure 4.** Images and spectra of the Ag nanoprisms illuminated by supercontinuum whitelaser. (**a**,**b**) Scattering images of silver nanoprisms before and after DO measurement (50 μM, 1.6 mg/L). And their corresponding normalized extinction spectra captured through the optical detection module (**c**,**d**). The spectra of the sample flow were collected five times, 10 s apart.

#### *3.5. Device Performance*

In glucose and glucose oxidase systems, dissolved oxygen will be transformed into H2O2 and further affects the oxidation and erosion process of Ag nanoprisms. The precise and sensitive detection of dissolved oxygen is practicable based on the detection of blue shift in the plasma resonance peak of the Ag nanoprisms. The shifts of the SPR peak of silver nanoprisms were measured under different dissolved oxygen concentrations, and the integration time of the spectrometer was 20 ms, as shown in Figure 5a. The blue shift of the SPR peak increased with the concentration of DO. The flow rate ratio of Ag-nanoprisms solution and the post-treatment water sample (*r* = *QNPs*/*QWS*) was fixed at 1. The corresponding relationship between the SPR peak shift (Δλ) and concentration (C) of DO was shown in Figure 5b. A linear relationship was formed between the SPR peak shift and DO concentration ranging from 0 to 50 μM (1.6 mg/L). The linear relationship was expressed as Δλ = 2.3794C + 3.721 with a correlation coefficient R<sup>2</sup> of 0.9889. By changing the flow rate ratio, *r* = *5*, another linear calibration curve was acquired based on the SPR peak shift versus DO concentration, ranging from 0 to 250 μM (8 mg/L), as shown in Figure 5c. The linear relationship was expressed as Δλ = 0.4554C + 9.0308 with a correlation coefficient R<sup>2</sup> of 0.9944. High reconfigurability is one of the most significant advantages of optofluidic systems. By simply adjusting the ratio (*r*) between silver nanoprisms flow and sample flow, the continuous adjustment of the detection ranges of DO from 0 to 16 mg/L can be realized dynamically, shown in Figure 5d. The blue line shows the relationship between SPR peak shift (Δλ) and flow rate ratio (*r)* when the concentration of DO was fixed at 50 μM. The orange line shows the detection range adjustment by changing *r*. The insert is an enlarged graph of the relationship between detection range and *r* (0–1). When detecting a water sample with the same DO concentration, Ag nanoprisms will be etched to a greater degree at a low flow rate ratio, which leads to a more obvious blue shift of SPR peak. A better detection limit can be achieved due to the greater response and sensitivity of the system at a low flow rate ratio. For the case of *r* = 0.5, a relatively low detection limit can be achieved. Under the established experimental conditions, a blank water sample was measured. After 10 measurements, taking the signal-to-noise ratio of 3, the system detection limit was calculated to be *D* = *3*σ/*slope* = 0.11 μM (3.52 μg/L); σ is the standard deviation of the blank sample.

**Figure 5.** (**a**)Normalized SPR extinction spectra of the Ag nanoprism with different DO concentrations. Plots of SPR peak shift (Δλ) vs. concentrations of DO at different flow rates, (**b**) *r* == 1, (**c**) *r* = 5. (**d**) The blue line shows the relationship between SPR peak shift (Δλ) and flow rate ratio (*r)* when the concentrations of DO was fixed at 50 μM (1.6 mg/L). The orange line shows the detection range adjustment by changing *r*. The insert is an enlarged graph of the relationship between detection range and *r* (0–1).

The spectral characteristics of the silver nanoprisms under different pH values of sample water were measured, and the SPR peak shift as a function of pH was obtained. The effect of pH on the measurement results is shown in Figure 6a. Under different temperature conditions, the optofluidic device was also used to measure water with the same dissolved oxygen content. The change in the resonance peak of the silver nanoprisms was observed and recorded, as shown in Figure 6b. Under the circumstance of increasing pH value and decreasing temperatures, the SPR peak shift (Δλ) decreased. This was probably due to the impact on the activity of GOD under a high pH value and low-temperature condition. Besides, the detection system was not affected by pH value and temperature in a wide range, and exhibited relatively high stability.

g yg y

**Figure 6.** SPR peak shift of triangular silver nanoprisms under different conditions. (**a**) SPR peak shift of Ag nanoprisms versus pH value of sample water. (**b**) SPR peak shift of Ag nanoprisms versus incubation temperature.

Since the ratio of nanoprisms and the water sample of the optofluidic device can be changed, a wide and adjustable detection range can be achieved. Compared with other iodometric method, electrode polarography method, and fluorescence method, the optofluidic method based on silver nanoprisms possesses better performance in dissolved oxygen detection Table 1 lists the performance of different DO detection techniques.


**Table 1.** Comparison of optofludic method with other DO detection techniques.

In addition, the waste solution containing silver after DO measurement was collected and treated by zeolite [38]. The wastewater treatment process is shown in Figure A6 (see Appendix A). Then the absorbed silver was recycled by the precipitation method [39]. The rate of recovery can reach up to 98.3%, and the purity of silver powder can achieve 99.8%. The recovered silver can be recycled in DO measurement. This saves reagents, reduces detection cost and avoids environmental pollution.

#### **4. Conclusions**

In conclusion, we demonstrated a high-accuracy, eco-friendly optofluidic dissolved oxygen detector based on the etching of silver nanoprisms. The characteristics of the system and its anti-interference ability were investigated. For the measurement of dissolved oxygen, the detection limit is as low as 3.52 μg/L. The device possesses repeatability and optical stability. It is suitable for the measurement of dissolved oxygen under different environmental conditions. Compared with traditional dissolved oxygen determination methods, the optofluidic detector possesses a higher measurement resolution with a lower cost. Besides, the optofluidic device has advantages such as less reagent consumption and simple operation, and can realize precise detection and high adjustability in its detection range. High precision and reproducibility in the detection of DO is beneficial for researchers in helping them discover deep phenomena, conduct aquatic analysis, and reveal new laws in water environmental science. Therefore, the optofluidic dissolved oxygen detector has a good application prospect in water quality analysis.

Under the established conditions, the accurate detection of DO has been realized. In order to illustrate the effect of more factors such as the kinetics of mixing, total flow rate, micromixer design on the performance of the optofluidic DO sensor, further systematical investigation is necessary and is under conducting in our laboratory. Besides metal nanoparticles, there are still some materials, such as frontier soft optical materials, that are worthy of exploitation and have great potential in terms of biomedical sensing. This study will be beneficial to simplifying the system and optimizing the chip structure for various sensing applications.

**Author Contributions:** Conceptualization, Y.Z. and Y.Y.; methodology, Y.Z.; writing—original draft preparation, Y.Z.; writing—review and editing, L.C., X.H. and F.W.; supervision, Y.Y; all authors contributed to manuscript writing. All authors have read and agreed to the published version of the manuscript.

**Funding:** This work was financially supported by the Foundation Research Fund of Shenzhen Science and Technology Program (no. JCYJ20170818112939064), the National Natural Science Foundation of China (no. 11774274), the Open Financial Grant from Qingdao National Laboratory for Marine Science and Technology (no. QNLM2016ORP0410), National Key R&D Program of China (2018YFC1003200). We also acknowledge the assistance with nanofabrication provided by the Center for Nanoscience and Nanotechnology at Wuhan University.

**Acknowledgments:** The authors would like to thank Xiaoqiang Zhu for beneficial discussions, and Yu Gao for manuscript preparation.

**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.

#### **Appendix A**

**Figure A1.** Simulation of liquids mixing in microchannel with different designs. (**a**) Cross-sectional concentration profiles of straight channel (red line), Z-shape channel (blue line) and Z-shape channel with blocks (green line) at x = 1000 μm. (**b**–**d**) Stable concentration distribution in main channel by simulation. (**e**–**g**) Corresponding height curves.

**Figure A2.** SPR extinction spectra of Ag nanoprisms in presence of glucose or GOD.

**Figure A3.** (**a**)Large Image of the fabricated chip obtained by image stitching technology and microscopes of the channel design at the positions of inlets (**b1**), zigzag-channel (**b2**), microcylinders (**b3**) and absorption cell (**b4**).

**Figure A4.** Particle size distribution of the Ag nanoprisms by means of microfluidics (**a**) and batch approach (**b**).

**Figure A5.** TEM micrographs of the Ag nanoprisms after the detection of dissolved oxygen (50 μM, 1.6 mg/L), the shape of Ag nanoprisms changed owing to the etching.

**Figure A6.** Absorption of the silver waste water by zeolite. Residual rate of silver nanoparticles as a function of time during the water purifying process.

#### **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* **Structural Stability of Optofluidic Nanostructures in Flow-Through Operation**

**Yazan Bdour 1,**†**, Juan Gomez-Cruz 1,2,**† **and Carlos Escobedo 1,\***


Received: 25 February 2020; Accepted: 31 March 2020; Published: 2 April 2020

**Abstract:** Optofluidic sensors based on periodic arrays of subwavelength apertures that support surface plasmon resonance can be employed as both optical sensors and nanofluidic structures. In flow-through operation, the nanoapertures experience pressure differences across the substrate in which they are fabricated, which imposes the risk for structural failure. This work presents an investigation of the deflection and structural stability of nanohole array-based optofluidic sensors operating in flow-through mode. The analysis was approached using experiments, simulations via finite element method, and established theoretical models. The results depict that certain areas of the sensor deflect under pressure, with some regions suffering high mechanical stress. The offset in the deflection values between theoretical models and actual experimental values is overturned when only the effective area of the substrate, of 450 μm, is considered. Experimental, theoretical, and simulation results suggest that the periodic nanostructures can safely operate under trans-membrane pressures of up to 20 psi, which induce deflections of up to ~20 μm.

**Keywords:** optofluidic; sensor; surface plasmon resonance; nanohole array; mechanical properties; nanofluidic; nanoplasmonic

#### **1. Introduction**

The development of new point-of-care (POC) diagnostic technologies requires low-cost, fully integrated sensing platforms capable of providing quantitative results in situ. At the same time, POC diagnostic platforms have a tremendous potential that is yet to be fully exploited. Telemedicine, for instance, aims to monitor the health of patients remotely through on-site sensing using personal devices, holding a global market of ca. \$20 billion USD (United States dollars) [1]. A trendy and increasingly demanded approach to in situ sensing is the use of lab-on-a-chip platforms enabled by cell phones to record, analyze, and transmit the results [2–4]. With the recent emergence of new pathogens, such as the Coronavirus and the Yaravirus, an on-site analysis will limit their health impact with a rapid sensing test, quantifying the severity of the infection, and assisting with the quarantine measures [5,6]. Periodic arrays of subwavelength structures fabricated in metal films enable surface plasmon resonance (SPR), which motivated their use as biosensors for several applications in different fields [7–12]. Ordered arrays of metallic nanoholes are optofludic structures that enable transport of both fluid and analyte via nanofluidic confinement and nanoplasmonic sensor. The plasmonic resonance signature obtained from nanohole arrays (NHAs) allows the detection of biologically relevant analytes in label-free fashion and real time. Toward the development of POC biosensing platforms, these optofluidic nanostructures are integrated into microfluidic environments in order to create fully integrated sensors compatible with portable electronics [13]. NHA-based sensors

are ideal for field applications due to their small footprint and integration abilities as evidenced by recent demonstrations for the detection of bacteria, such as *Chlamydia trachomatis* [14], viruses, such as Ebola [15], cancer biomarkers [16] and uropathogenic bacteria [17]. Flow-through optofluidic structures also enable the enrichment of analytes in liquids by an electrohydrodynamic effect occurring around the NHAs when an electric potential and a pressure bias are applied to the fluid in a closed system [18]. Despite their demonstrated potential in sensing, most applications involving nanohole arrays focus on exploiting the conventional optical capacities of these nanostructures. The mechanical stability of the nanohole membranes is an overlooked aspect of their properties that are key when functioning as nanofluidic structures. In analogy to porous silicon-based membranes, where permeability increases significantly as membrane thickness decreases, the volumetric flow across nanostructured optofluidic sensors increases with the open pore fraction. However, plasmonic nanostructures with built-in thin membranes may suffer from low mechanical stability which could limit, critically, their use as optofluidic flow-through sensors [19–21]. The membrane's mechanical properties change due to the change in the structural morphology of the porous membrane as it deflects under pressure. The stability decreases by a correction factor (1 − *P*), where *P* is related to the porosity of the membrane [22].

Recent studies demonstrate that through-nanoapertures fabricated in thin (~50 nm) gold-coated Si3N4 substrates offer additional fluidic abilities that can be used to target in-hole delivery of analytes when operated as optofluidic sensors [23,24]. However, flow-through operation results in transmembrane pressures that could potentially damage the rather brittle nanostructures. The mechanical properties of the organized nanohole arrays are not completely understood due to their sensitivity, brittleness, and nano-sized structures. Here, we present a study on structural aspects of Au-on-nitride optofluidic nanoplasmonic sensors operating in flow-through fashion at flowrates compatible with biosensing applications.

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

#### *2.1. Fabrication of Periodic through Subwavelength Apertures*

Through-nanohole arrays were fabricated using focused ion beam (FIB) milling using 100-nm-thick Si3N4 free-standing membranes (Norcada, Edmonton, AB, Canada) coated with a thermally evaporated 100 nm layer of gold via a 5-nm chromium adhesion layer. Milling was achieved using a gallium ion beam set at 40 keV with a beam current of ~30 pA, with a typical beam spot size of 10 nm, and the dwell time of the beam at one pixel was set to 20 μs. Two arrays of through-nanohole arrays with an area of 20 μm by 20 μm, diameter of ca. 230 nm, and pitch of 560 nm were fabricated.

#### *2.2. Fabrication of Microfluidic Chips*

The microfluidic chip was fabricated using a replica molding technique as described in detail elsewhere [25]. The general steps of the fabrication procedure are briefly described next. A mask with the microfluidic pattern was generated using SolidWorks CAD software (Dassault Systems Solidworks Corp., Waltham, MA, USA). The design included one inlet and one outlet of 1.5 mm, and a 5-mm-wide channel with 100 μm in height. A master was fabricated by spin-coating SU-8 100 photoresist (MicroChem Corp., Newton, MA, USA) on a clean three-inch silicon wafer (Silicon Quest International Inc., Santa Clara, CA, USA). The coated wafer was then prebaked for one minute at 65 ◦C and for 10 min at 95 ◦C. The mask with the channel pattern was then placed over the coated wafer and exposed to ultraviolet (UV) light for 90 s. Next, the exposed wafer was hard-baked at 65 ◦C for 1 min and at 95 ◦C for 10 min. The master was subsequently developed using a SU-8 developer (MicroChem Corp., Newton, MA, USA). A 12:1 mixture of Sylgard 184 elastomer to curing agent (Dow Corning, Midland, MI, USA) was mixed, degassed in a vacuum, and poured onto the master. After baking at 85 ◦C for 20 min, the replica was removed from the mold. Inlets and outlets were provided 1-mm punched holes for fluidic access. Microfluidic connections were achieved using polyether ether ketone

(PEEK) tubing (Upchurch Scientific, Oak Harbor, WA, USA). A schematic representation of the set-up is shown in Figure 1.

**Figure 1.** (**a**) SEM image of fabricated periodic subwavelength apertures via focused ion beam (FIB). The nanostructures were 230 nm in diameter and 560 nm in pitch. (**b**) Schematic representation of a nanohole array in a microfluidic chip in flow-through operation.

#### *2.3. Optofluidic Structure Deflection Analysis*

Finite element analysis (FEA) was used as a means to know the order of magnitude of the deflection and the mechanical stress that the optofluidic sensor may experience in flow-through operation. COMSOL Multiphysics (COMSOL, Stockholm, Sweden) was used to simulate a simplified model of the optofluidic sensor under a prescribed unidirectional and orthogonal pressure on one of the faces of the suspended membrane. The simulations were used firstly to estimate the order of magnitude of applied pressures that would result on the deflection of the substrate containing the optofluidic structures. This first model involved a stationary elastic model with default Lagrange–quadratic element type. The finite element analysis solves for the displacement field at a specific point on the membrane for every input force. For the linear model, the system is governed by three tensor partial differential equations: ∇·<sup>σ</sup> + *Fv* = 0, <sup>ε</sup> = <sup>1</sup> 2 (∇*u*) *<sup>T</sup>* <sup>+</sup> <sup>∇</sup>*<sup>u</sup>* <sup>+</sup> (∇*u*) *<sup>T</sup>*∇*<sup>u</sup>* , and *C* = *C*(*E*, *v*), where σ is the Cauchy stress tensor, *Fv* is the body force per unit, *u* is the displacement vector, ε is the infinitesimal strain tensor, *C* is the fourth-order stiffness tensor, *E* is the Young's modulus, and *v* is the Poisson's ratio. A second static, nonlinear stress–strain model was used to compare the experimental data and to validate the deflection values obtained for the prescribed pressure range. The nonlinear stress–strain behavior was achieved by using a power-law nonlinear elastic material model, accounting for geometric nonlinearities, which is governed by Ludwik's law, τ = τ<sup>0</sup> + *k*γ1/*n*, where τ is the shear stress, γ is the shear strain, and *n* is an integer [26,27]. A user-controlled mesh with Lagrange–quadratic element type was used for this nonlinear model, to guarantee an acceptable mesh size along the thickness of the modeled substrate. The finite element analysis solves for the displacement field at a specific point on the membrane for every input force. In both models, linear and nonlinear, the parameters of Si3N4 were mainly used, as the values for the mechanical properties for this material supersede those of the metal components in the sensor, namely, a Young's modulus of 250 <sup>×</sup> 109 Pa, a density of 3.1 <sup>×</sup> 10<sup>3</sup> kg/m3, and a Poisson's ratio of 0.23. The surrounding surfaces around the membrane that correspond to the areas that define the thickness of the substrate were set as fixed boundaries. The transmembrane pressures were varied from 1 to 20 psi, as this range corresponds to flow rates on the order of nL/min, which are commonly used in biosensing applications. The deflection of the substrate and the stress (von Mises criterion) were recorded.

In addition to finite element method (FEM)-based models, analytical models on the mechanical behavior of perforated membranes published in the literature were also used to estimate the deflection of the optofluidic sensors in this study, as detailed in the Section 3 [22].

#### **3. Results and Discussion**

Figure 2 shows a schematic representation of the experimental setup used to measure the deflection of the membranes. Figure 3 shows the computer-aided design (CAD) models used to study the deflection of the optofluidic sensors via COMSOL Multiphysics software. Figure 3a shows the simplified model with a single nanoaperture at the center, used in the linear elastic material

simulations. The model accounts for a 100-nm-thick membrane with a square surface with a side length of 500 μm, and a circular opening of 10 μm for surface coverage equivalency of the effective surface of the nanoapertures. Figure 3b shows an image of the CAD model used for the nonlinear simulations, a square 100-nm-thick membrane with side length of 500 μm and a 20 μm × 20 μm array of 230-nm-diameter holes with pitch-to-diameter ratio of 2. In both cases, linear and nonlinear models, the mesh curvature factor was 0.6, the maximum element scaling factor was 1.9, the resolution of narrow regions was 0.3, and the optimize quality feature was set to on. The linear model had a maximum element size at all boundaries of 30 <sup>×</sup> 10−9. The resulting mesh had ~210 <sup>×</sup> 103 domain elements with ~40 <sup>×</sup> 10<sup>3</sup> boundary elements and ~1.4 <sup>×</sup> 10<sup>3</sup> edge elements. The nonlinear model had ~1.4 <sup>×</sup> 10<sup>4</sup> domain elements, ~900 <sup>×</sup> 10<sup>3</sup> boundary elements, and ~6 <sup>×</sup> 103 edge elements. The models were solved for pressures applied to the bottom surface of the substrate, for 1 psi, and then using the sweep parameter feature for a pressure range of 2–20 psi with 2-psi pressure increments.

**Figure 2.** Schematic representation of the experimental set-up.

**Figure 3.** Computer-aided design (CAD) models used for the finite element method (FEM)-based simulations. (**a**) CAD model used for linear elastic simulations. (**b**) CAD model used for the nonlinear elastic simulations. A detail of the nanoapertures in the CAD model and the corresponding mesh are shown as insets. Scale bar represents 10 μm.

Figure 4 shows images of selected values for the deflection and stress distribution of the model of the membrane under an applied pressure of 20 psi. Figure 4a,b show the displacement in the *z*-direction for the linear and nonlinear models, respectively. The results are presented as non-deformed, with vectors representing the direction and magnitude of the deflection. The pattern of deflection observed from the simulations, as expected, is quasi-circular, with increasing magnitude toward the center of the free-standing membrane. The maximum deflection values, for the linear and nonlinear

simulations at an applied pressure of 20 psi, were 24.08 and 19.39 μm, respectively. Maxima were always obtained at the apex of the deformed membrane. Figure 4c,d show the von Mises stress distribution for an applied pressure of 20 psi. The maximum stress found in the simulations was on the order of 1 <sup>×</sup>108 to 10 <sup>×</sup>108 Pa, which suggests that the substrate which is housing the nanoapertures could adequately withstand the deformations resulting from the applied pressure. The simulation results were used to define a range of pressure that could be used experimentally, avoiding failure of the membrane.

**Figure 4.** Simulation results of linear and nonlinear models. Membrane deflection under an applied pressure of 20 psi for (**a**) the linear model and (**b**) the nonlinear model. The apertures are shown in the insets within yellow dashed boxes in both cases. Stress distribution (von Mises yield criterion) under an applied pressure of 20 psi for (**c**) the linear model and (**d**) the nonlinear model.

Figure 5 shows a bright-field microscopy image of the Au-on-nitride membrane before and after the application of a pressure of 10 psi. The substrate included two rectangular periodic arrays of nanoapertures, indicated with yellow dashed lines. The boundaries of the Si3N4 membrane are indicated by red dashed lines. The focal plane in both images is the same, which indicates the deflection of the substrate under the applied pressure.

**Figure 5.** Membrane deflection before and after the application of a pressure of 20 psi. Scale bar represents 100 μm.

In order to measure the deflection experimentally, the elevation difference at the apex of the membrane was used as reference, and the in-focus *z*-positions were recorded. The applied pressure on the surface of the substrate was monitored and regulated to achieve a constant value throughout the measurement of the deflection. Fringe patterns can be observed in the deflected membrane case, which correspond to the reflected light, confirming a level gradient along the surface of the substrate, and a maximum translation at the apex. The *z*-positioning precision of the inverted microscopy system used in this study was 0.2 μm, which allowed measuring deflections with micrometer precision, at 2-μm intervals.

Figure 6 shows experimental and simulations results for applied pressures of 1–20 psi. The trend from the linear simulation model was linear, as expected, with corresponding minimum and maximum deflections of 1.209 and 24.08 μm. In contrast, the deflection results from the nonlinear model decreased with the applied pressure, with minimum and maximum values of 2.584 and 19.39 μm. The same trend was found for experimental values, with the magnitude of the maximum deflection at the apex decreasing with the applied pressure. This can be explained by considering the physical restriction along the frame of the free-standing membrane and due to the mechanical properties of the material. The figure also shows the results from three analytical models that were used to obtain theoretical values, i.e., the Rijn et al. [22], Ugural [28], and Kovacs et al. [29] models, as well as an adjusted Kovacs model fit with the experimental values. These models are similar to each other, whereby they all consider the perforation in a membrane as an error factor affecting the Young's modulus of the membrane. The deflection of a membrane is given by Equation (1) [30].

$$w\_{\text{max}} = k\_0 L \sqrt[3]{\frac{P\_0 \, L}{E\_{eff} \, h'}} \tag{1}$$

where *w* is the z-axis displacement, *L* and *h* are the size and the thickness of the membrane, and *P*<sup>0</sup> is the applied pressure. The constant *k*<sup>0</sup> is equal to 0.318, 0.325, and 0.319 within Rijn's, Ugural's, and Kovacs' models, respectively [30]. *Eeff* is the effective Young's modulus, calculated as *Eeff* = (1 − *P*)*Eclosed*, where *Eclosed* is the Young's modulus of unperforated membrane, and *P* is the correction factor. *P* is dependent on the perforation and is defined as the fraction of the open areas over the total area of the membrane. As the models are similar, there is negligible difference between the deflection values obtained using the three different models [30]. In the case of the optofluidic sensor, the deformable section of the membrane is smaller than the 500 μm by 500 μm of the free-standing substrate, as observed in Figure 5. The theoretical models do not consider the frame around the deformable area. Therefore, there is an offset between the deflection values obtained using the models and those obtained experimentally, as shown in Figure 6.

**Figure 6.** Experimental, theoretical, and simulation results of the maximum membrane deflection (apex). Error bars indicate standard deviation (*n* = 5).

The experimental results have a similar trend compared to the theoretical models. Over the non-linear region (< 7 psi), the experimental values are on average ∼ 32% below the theoretical maximum, and ∼ 12% below the theoretical maximum within the linear region (> 7 *psi*). The slopes for the experimental and theoretical (Kovacs) values were 0.4831 and 0.4826 μm/psi, respectively within the linear region, with *R2* (coefficient of determination (COD)) values of 0.937 and 0.993, respectively. The slopes indicate that the models do not quantify the actual deflection of the membrane. However, they accurately represent the trend of the membrane's deflection. As such, the unperforated area around the nanohole arrays is influential on the mechanical stability of the membrane. With the assumption that some length of area around the unperforated area does not deflect, then the deflection of the membrane can be rewritten as follows:

$$w\_{\rm max\_r} = k\_0 L\_{eff} \sqrt[3]{\frac{P\_0 \ L\_{eff}}{E\_{eff} \ h}} \tag{2}$$

where *Leff* , is the effective length of the membrane based on the experimental values, calculated as *Leff* = *Area o f holes*/*Peff* . *Peff* is the effective correction factor based on the experimental values, where it is assumed that some length around the unperforated area does not deflect. *Eeff* is adjusted to the experimental values and calculated as *Eeff* = <sup>1</sup> <sup>−</sup> *Peff Eclosed*. The model found a range of *Peff* values based on each experimental deflection point from 2.138 <sup>×</sup> <sup>10</sup>−<sup>3</sup> to 5.09 <sup>×</sup> <sup>10</sup><sup>−</sup>4, corresponding to effective membrane lengths of 225 μm to 460 μm, respectively. Figure 6 illustrates that the model is incapable of fitting all the experimental values with one value of *Peff* . The initial deflection value of the experimental values has an *Leff* of 225 μm, where the *Leff* non-linearly increases until it plateaus to a constant value of 460 μm within the linear region of the experimental values. The effective length paints a clear image of the membrane's behavior under pressure. Initially, at low pressures, only the center area of the membrane deflects, while the majority of the membrane is not affected by the applied pressures. As the applied pressure increases, the deflected area grows until it reaches a maximum constant value (460 μm). Even at the maximum value of effective lengths, some outer areas of the membrane do not deflect, reassuring the limitations of deflection model. The experiment was not designed to bring the substrate to mechanical failure; however, the pressure value for the breaking point can be extrapolated from the theoretical model based on the material's properties. The inflection

point of the membrane is not at the edges of the membrane but limited to the effective length of the membrane (i.e., *Leff*). Based on the Rijn et al. and Timoshenko et al. models, the maximum pressure applied can be found based on the total stress of the material as shown in Equation (3) [22,31].

$$
\sigma\_{\text{total}} = \sigma\_{\text{tensile}} + \sigma\_{\text{bond}} = \frac{0.297}{1 - \upsilon} \left( 1 + \frac{1.439}{0.358} \right) \sqrt[3]{\frac{P\_0^{\text{-2}} L\_{eff}^{\text{-2}} E\_{eff}}{\left(1 - \upsilon^2\right) h^2}},\tag{3}
$$

where σ*total* is the total stress of the membrane, and σ*tensile* and σ*bend* are the tensile stress due to stretching and the maximum bending stress near the middle of the membrane's deflection edges, respectively. The model is valid when the substrate is under a substantial load that results in large deflections (i.e., *wmax*/*h* 1). Considering that the reported ultimate stress, σ*ultimate*, is on the order of 10<sup>9</sup> Pa, and the intrinsic tensile stress is 10<sup>8</sup> Pa for a silicon nitride membrane, then the internal stresses can be neglected since they are an order of magnitude lower than the total stress [22]. For a nonductile inorganic material, the σ*ultimate* is equivalent to its yield stress. Taking 2.5 GPa as σ*total*, based on the mechanical properties of the material, a pressure of 33.91 psi and deflection of 23.87 μm are obtained, corresponding to the maximum possible values at the verge of mechanical failure [32]. This theoretical maximum deflection value at the verge of failure, which adequately follows the trend of the adjusted theoretical curve, is shown in Figure 6.

#### **4. Conclusions**

This work presented an investigation of the deflection and structural stability of optofluidic nanohole array-based sensors operating in flow-through mode. The study was approached using experiments, theoretical models, and FEA via computer simulations through FEM. Linear and nonlinear material models were simulated using COMSOL Multiphysics software. The simplified linear model had an expected discrepancy with experimental values, but these were useful to obtain an estimation of the order of magnitude of transmembrane pressures that would allow studying the deflection of the substrate when used in flow-through operation, while avoiding mechanical failure. The discrepancies were up to ~20%. In contrast, the nonlinear model, accounting for a complete nanohole array, accurately described the deflection values obtained experimentally. The stresses corresponding to these deflections can be used to predict maximum operation values that could prevent failure of the optofluidic nanostructures. Three analytical models were used to analyze the deformation of the sensor. The models depicted the behavior of the deflected substrate under pressure but did not intrinsically fit the experimental results since only a fraction of the surface deflects due to the attachment of the free-standing substrate to the silicon frame. Even when the entire 500-μm membrane is under pressure, only a reduced square area, ranging from 225 μm to a maximum of 460 μm per side, deflects. Once adjusted, the theoretical model better fit the experimental deflection values. Based on the models, the fracture point was extrapolated from the maximum yield stress of silicon nitride membranes. As the membranes are composed of nonductile, inorganic material, their yield stress is equivalent their ultimate stress, which resulted with a maximum possible deflection of 23.9 μm, with the applied pressure of 33.9 psi. Although the optofluidic structures are limited by their fragile mechanical stability in flow-through operation, these results show that they are capable of withstanding transmembrane pressures compatible with sensing applications, where the analyte is required to be brought into the apertures. Simulations that could predict the deflection of the structures would greatly benefit the design needs of flow-through optofluidic platforms for specific applications in the context of biosensing.

**Author Contributions:** Conceptualization, C.E.; methodology, Y.B., J.G.-C., C.E.; software, Y.B., J.G.-C., C.E.; validation, C.E.; formal analysis, Y.B., J.G.-C., C.E.; investigation, Y.B., J.G.-C., C.E.; resources, C.E.; data curation, C.E.; writing—original draft preparation, Y.B., J.G.-C., C.E.; writing—review and editing, C.E.; supervision, C.E.; project administration, C.E.; funding acquisition, C.E. All authors have read and agreed to the published version of the manuscript.

**Funding:** C.E. acknowledges funding from the Natural Sciences and Engineering Research Council of Canada (NSERC) RGPIN-2010-5138, and Canada Foundation for Innovation John R. Evans Leaders' Fund Program (No. 319670). Y.B. acknowledges funding from Materials for Advanced Photonics and Sensing (MAPS) funded through NSERC CREATE and Ontario Graduate Scholarship (OGS). J.G.-C. acknowledges the national graduate scholarship provided by the National Council for Science and Technology of Mexico (CONACYT).

**Conflicts of Interest:** The authors declare no conflicts 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* **3D Hydrodynamic Focusing in Microscale Optofluidic Channels Formed with a Single Sacrificial Layer**

**Erik S. Hamilton 1,\*, Vahid Ganjalizadeh 2, Joel G. Wright 1, Holger Schmidt <sup>2</sup> and Aaron R. Hawkins <sup>1</sup>**


Received: 11 February 2020; Accepted: 26 March 2020; Published: 27 March 2020

**Abstract:** Optofluidic devices are capable of detecting single molecules, but greater sensitivity and specificity is desired through hydrodynamic focusing (HDF). Three-dimensional (3D) hydrodynamic focusing was implemented in 10-μm scale microchannel cross-sections made with a single sacrificial layer. HDF is achieved using buffer fluid to sheath the sample fluid, requiring four fluid ports to operate by pressure driven flow. A low-pressure chamber, or pit, formed by etching into a substrate, enables volumetric flow ratio-induced focusing at a low flow velocity. The single layer design simplifies surface micromachining and improves device yield by 1.56 times over previous work. The focusing design was integrated with optical waveguides and used in order to analyze fluorescent signals from beads in fluid flow. The implementation of the focusing scheme was found to narrow the distribution of bead velocity and fluorescent signal, giving rise to 33% more consistent signal. Reservoir effects were observed at low operational vacuum pressures and a balance between optofluidic signal variance and intensity was achieved. The implementation of the design in optofluidic sensors will enable higher detection sensitivity and sample specificity.

**Keywords:** 3D hydrodynamic focusing; optofluidic; lab-on-a-chip; biosensor; microscale channel; microfluidic; liquid-core waveguide; single layer; reservoir effect

#### **1. Introduction**

The optofluidic detection of particles in liquid waveguide channels has recently grown in significance [1–7]. The intersecting of solid-core waveguides with liquid-core waveguides enables light-matter interaction, such as fluorescence generation for single molecule detection. This is especially interesting for identification of particles in liquid, such as disease pathogens [8–14]. The small scale of the optofluidic channels is critical for single molecule detection in flow.

The nature of the microfluidic channels found on optofluidic sensor platforms sets some of the ultimate sensitivity and accuracy limits. The small cross-sectional area of the channels typically means operation in the laminar flow regime, which results in a parabolic velocity distribution. Fluid that is in the center of the channel flows faster than fluid at the walls. This means fluorescent particles, being distributed evenly through a channel, would spend different amounts of time in perpendicular excitation light beams as they flow past excitation points introducing variation in excitation times and fluorescent signal generation. Additionally, the excitation light exhibits an optical mode intensity profile that results in a distribution of generated fluorescence intensities. Additionally, the collection efficiency of fluorescence in a liquid-core waveguide is dependent on particle position [15,16]. Particles that are near the center of the channel experience higher collection efficiency. Moreover, velocity-based identification schemes suffer from the particle velocity distribution as variation limits the sensitivity and specificity capabilities of these methods. Taken together, all of these possible variations in signal intensity and particle velocity distribution introduce uncertainty in particle detection.

Hydrodynamic focusing (HDF) promises to enhance the detection capabilities in optofluidic channels by controlling the position of target particles in the fluid channel. This results in narrower velocity and excitation intensity distributions, which means less variation in the fluorescent signal intensity distribution [17]. Three-dimensional hydrodynamic focusing (3DHDF) aims to limit the position of target particles both horizontally and vertically. This ensures they pass through the excitation beam, flow at a more uniform velocity, and experience optimal collection efficiencies resulting in less fluorescent signal intensity variation. We have predicted a signal enhancement of between three and five times with 3DHDF, but the complexity of implementation should be considered.

HDF is commonly achieved using buffer fluid or sheath flows to surround and squeeze a sample fluid stream. The engineering challenge comes in designing and fabricating the intersection for these flows. Two-dimensional hydrodynamic focusing is relatively straightforward, even at the 10-μm scale. It simply requires flowing buffer fluid from the left and right sides of the sample stream to achieve horizontal, or in-plane, focusing, as in Figure 1a. The buffer fluid occupies the edges of the microfluidic channel volume, limiting the locations that sample particles (flowing in from a central channel) can occupy. Focusing the sample stream from the top and bottom, or vertically, is much more difficult to achieve, because the fluid must be directed from out-of-plane. This can lead to complex designs, difficult fabrication processes, and complicated operation. However, the combination of horizontal and vertical focusing resulting in a sample stream surrounded on all sides, or sheathed, by buffer fluid, as in Figure 1b, enables enhanced optofluidic performance and applications [18–22]. Our goal with this work was to develop a simple buffer fluid induced 3D focusing scheme at the 10-μm scale while using surface micromachining. Ultimately, we desired a design only requiring a single sacrificial layer of material to avoid complexity in fabrication and maximize device yield. We also wanted to keep device operation as simple as possible.

**Figure 1.** Hydrodynamic focusing occurs when sample fluid (drawn in solid blue) is squeezed and sheathed from (**a**) two dimensions (horizontal focusing) or (**b**) three dimensions (horizontal and vertical focusing) by a buffer fluid (drawn in striped red), controlling the position of sample fluid in the fluid channel.

A wide variety of interesting designs have been developed to perform 3DHDF while using buffer fluid at the microscale. One existing method literally injects sample fluid into a buffer fluid stream using a nanoneedle, or micronozzle [23]. The micronozzle cantilever is formed with oxide over silicon before the silicon is etched and the nanoneedle released. This fragile structure extends into the 100 μm scale microfluidic channel, where buffer fluid occupies the volume and performs the sheathing function for 3DHDF. Not only is the fabrication complex, but the structure is fragile and it might be impossible to shrink to the 10-μm scale as the oxide annealing process that forms the nozzle may lead to a closed nozzle. Alternatively, tilted lithography has been used to create fluid manifolds for directing buffer fluid around sample fluid for sheathing [24]. The fabrication process requires a rotatable, tiltable jig holder for the microfluidic device during multiple photolithography exposures. The design is similar to the micronozzle, in that it includes a sample injection nozzle. Again, the structures are on the 100 μm scale. They both require three or four fluid reservoirs to operate, depending on layout. Other designs requiring complex fabrication processes or operation include a femtosecond laser exposure design and a PDMS stack that requires five to seven layers to make and six fluid reservoirs to operate [25,26].

Simpler, "single layer" designs make use of clever fluid dynamics properties to induce 3DHDF with buffer fluid through secondary flow production. This takes the form of Dean Vortex generation in fast moving fluids. The vortices induce flow orthogonal to the main flow direction, drawing out and shaping the sample stream. The microscale fluid channel structures can be simpler because the fluid dynamics do some of the work, reducing complexity in fabrication and operation.

The curve design, sometimes called "microfluidic drifting", requires four or five fluid reservoirs to operate [27]. Fluid moves through a 90-degree curve in a microchannel. Sample fluid is located on the inside of the curve and buffer fluid on the outside. As the fast-moving fluids travel around the bend, secondary flow occurs, drawing the sample fluid out horizontally. Additional buffer fluid envelopes the vertically focused sample stream from the sides, performing horizontal focusing and completing the 3DHDF.

The contraction-expansion array design (CEA) works in a similar manner, inducing secondary flow in the form of Dean Vortex generation [28]. Rather than a curve or bend, the CEA consists of a chain of narrow and wide fluid channel sections that cause the fluid to contract and expand. The sample fluid is located on the inside of the narrow contraction regions. When it reaches the expansion region, the sample fluid is horizontally drawn out before being contracted again. The pressure change between the regions induces the secondary flow and the effect compounds. Unlike the Curve design, the CEA completes 3DHDF with no additional buffer fluid for horizontal focusing. In this case, three fluid reservoirs are required for operation.

The last of the single-layer, vortex generator designs discussed here is the microstructure stream-sculpting design [29]. More commonly known as micropillars, the structures are located within the volume of the microfluidic channel in the path of the already horizontally focused sample stream. When the sample stream comes in contact with the micropillars, its path is redirected around the structures and secondary flow occurs. The effect is much like that of the CEA where the contraction and expansion caused by the row of pillars has a compounding effect, thereby sculpting the stream from a vertical stripe to a horizontal ellipse sheathed by buffer fluid. This design requires three to four fluid reservoirs, as the sample stream must be horizontally focused prior to its contact with the microstructures.

Note that these designs are more typical at the 100-μm scale or larger. When shrunk to the 10-μm scale, the fluidic resistance of the small cross-section fluid channels dramatically increases and the flow velocities that are required to induce the secondary flow focusing effect become impractical due to the massive microfluidic channel backpressures [17]. Thus, low fluid velocity is required and the existing single-layer designs become impossible to implement. This means that we cannot rely on secondary flow to do the work of focusing. We must rely on a sheathing or "volumetric flow ratio" design that can be operated at low velocity and in small cross-section channels.

We present a 3DHDF design requiring just one layer of sacrificial material. The low-pressure chamber design, or pit, requires four fluid ports to operate and it promises improved device yield over multi-layer designs. Moreover, it operates by pressure driven volumetric flow ratio, enabling low velocity focusing in channels with dimensions around 10 μm. We will present computer simulations, fabrication processes, and optical characterization of the focusing completed in microscale channels. Fluorescent beads in a flowing aqueous solution are used to characterize the design and demonstrate the enhanced detection capabilities of the system.

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

#### *2.1. Design Concept*

We have previously reported 3DHDF at the 10-μm scale while using a volumetric flow ratio technique [30]. However, complexity in the fabrication and operation remained. The design required three consecutive layers of sacrificial SU-8 photoresist that formed the fluid channel volume. The dimension steps from layer to layer introduced crevices in the silicon dioxide overlayers that were used to form the walls of the channel leading to structurally weak devices. Moreover, the design required six fluid reservoirs to operate, taking up large amounts of chip real-estate and making operation challenging. A two-layer design was also implemented, improving the yield and decreasing operation complexity by using four fluid reservoirs [31]. However, the multi-layer structure remained less than optimal and resulted in low device yield.

Quantifying fabrication complexity is challenging, as it includes processing time and difficulty. For example, multi-layer photoresist designs are extremely difficult in that subsequent layers must be perfectly processed on the first try, whereas single-layer designs allow for layer stripping and repatterning. Additionally, each layer attempt requires between three and six hours. Moreover, multi-layer designs prove to be less robust, resulting in low device yields. Every parallel processed device can become simultaneously damaged near the end of the multi-week fabrication process, requiring complete device remanufacturing. Moving to a single-layer design greatly simplifies the fluid volume formation process.

The schematics presented in Figure 2a,b show the fluid junction region of the more easily fabricated, more robust design introduced by this paper. It is formed with one layer of sacrificial photoresist and requires four fluid reservoirs to operate by pressure-driven volumetric flow ratio. The key fluid features that cause the fluid focusing are a pit and trench pair. These are etched into the substrate before a sacrificial material is deposited, covered, and etched out leaving the channels hollow. The pit refers to a large low-pressure chamber that is in-line with the sample stream flow path that the stream comes in contact with, causing it to drop and flatten out. Simultaneously, buffer fluid is injected above the sample stream to occupy the volume above it before both rise out of the pit and they are met from all sides by additional buffer fluid provided by the trench feature. The combined fluid stream exits the junction through the outlet as a 3D focused sample stream with a form factor that is similar to the cross-section of the outlet channel and ready for optical interrogation. The fluids are typically drawn through the chip by applying vacuum pressure to the outlet fluid port, but fluid can alternatively be pushed through by applying pressure to the inlet ports.

Figure 3 shows a top down view of the optofluidic chip with the 3DHDF element integrated. The silicon-based chip is 1 <sup>×</sup> 1 cm2. Four fluid reservoirs manage the sample, buffer, and waste fluid on the chip. The fluid on the chip flows from left to right, with the buffer fluid coming in contact with the sample fluid near the center of the chip at the focusing junction. The fluid focusing occurs before the liquid-core waveguide fluid channel is intersected by solid-core waveguides that carry light for fluorescence excitation and signal collection. This intersection point is called the excitation region, or excitation volume when referring to the fluid.

**Figure 2.** Schematics show (**a**) an above oblique view of the focusing junction describing the operation principle in which sample flow is shown in red, buffer flow in black, and outlet flow in dashed red, and (**b**) a beneath oblique view showing the etched features.

**Figure 3.** The focusing junction integrates into the optofluidic chip design near the center of the chip, indicated with the red ring, drawn from a top view.

#### *2.2. Channel Dimension Determination*

The fluid channel dimensions between the focusing junction at chip center and the fluid ports were determined by solving a pair of volumetric flow ratios (*Q*/*Q*) while using the known hydrodynamic relation *P* = *QR* or *Q* = *P*/*R*, where *P* is pressure, *Q* is volumetric flow rate, and *R* is fluid resistance. This relation is derived from Newton's second law, to the Navier–Stokes equations, to the Hagen–Poiseuille relation. The relation assumes incompressible laminar Newtonian fluid flow through a circular pipe. The hydraulic radius is substituted to apply the relation to a rectangular duct. The pair of volumetric flow ratios solved were of the buffer fluid flow and sample fluid flow, and of the buffer fluid flow and the outlet fluid flow. The pressure drop was chosen to be approximately equal between the ports and the fluid focusing junction, thus canceling out and leaving just the *R* terms. The *R* terms were expanded into the Hagen–Poiseuille form, the constants canceled out, and the hydraulic radius substituted, leaving

$$\frac{Q\_2}{Q\_1} = \frac{L\_1}{L\_2} \frac{R\_{H2}}{R\_{H1}} = \frac{L\_1}{L\_2} \frac{(2w\_1 + 2h\_1)^4}{w\_1^4 h\_1^4} \frac{w\_2^4 h\_2^4}{(2w\_2 + 2h\_2)^4} \tag{1}$$

*L* is fluid channel length, *w* is fluid channel width, and *h* is fluid channel height. Subscript 2 indicates terms relating to the buffer fluid channel and subscript 1 indicates terms that relate to the sample (or outlet) fluid channel. This equation was input into an online graphical calculator, called Desmos (San Francisco, CA, USA), which allowed for us to keep track of parameters and view the solution visually. Each *Q* ratio value was input manually as calculated from the input velocities and cross-sectional areas of the inlets from the CAD model (*Q* = *VA)*. The ratio was subtracted from both sides of the equation to create the slope intercept form for graphing. One of the dimension variables in the equation was replaced with the variable *x* to act as the solution. The equation was multiplied by a large number in order to cause the graphed line to appear nearly vertical for ease of dimension determination. The variable values and solution variable were adjusted until the solutions were deemed as acceptable for fabrication tolerances.

The height of the microchannels is 6 μm. The etch depth is 12 μm. The sample inlet channel is 12 μm wide and 4700 μm long. The buffer channels are 2500 μm long and 20 μm wide, but they split in two halfway from the port to the junction; the pit directed channel is 7 μm wide and the trench channel 12 μm wide. The outlet channel is 12 μm wide and the length is 2500 μm. Determining the dimensions based on volumetric flow ratio means the device can be operated by negative pressure at the outlet over a range of flow velocities.

#### *2.3. Modeling*

The design that is outlined in Figure 2 was developed in the computational fluid dynamics software Fluent (ANSYS, Canonsburg, PA, USA). The design was meant to be integrated with previously developed optofluidic channels, so it was constrained in outlet channel cross sectional dimensions of 12 μm × 6 μm [32]. The pit and trench etch features were designed to have the same depth of 12 μm so as to simplify etching processes during fabrication. Note that the pit determines the vertical dimension of the focused sample stream and the trench determines the vertical location of the focused sample stream at the outlet.

The model, as visualized in Fluent, is shown in Figure 4, where sample fluid begins as uniformly distributed particle traces and the buffer fluid is invisible. Sample fluid flows in the X direction from left to right, first expanding laterally before coming in contact with the low-pressure chamber, or pit. At this location, buffer fluid travels in the Z direction toward the center and it comes in contact with the sample fluid, pressing it down and occupying the top space of the channel. The sample stream rises up out of the low-pressure chamber before coming in contact with additional buffer fluid in the trench, also traveling in the Z direction toward the center. This raises the sample stream from the bottom toward the center of the channel, effectively 3D focusing the sample stream, as seen at the outlet (see the inset of Figure 4a). Note that the sample stream exhibits a nose shaped cross-section, where a majority of particle streams are predicted to be contained in the center of the channel vertically, but a small portion extend over the main body toward the top of the channel. This is a result of optimizing the design geometry while using large finite elements. This caused streamlines to end abruptly inside the model, resulting in an incomplete stream profile that appeared to be better focused. Increasing the element resolution (decreasing element size) overcame this issue and resulted in the complete profile seen in the figure. The model in Figure 4 used a body fitted cartesian mesh method with half micron element size, resulting in 182,222 finite elements. Although the design was optimized, tooled, and fabricated before the incomplete profile was discovered, it will be shown that the device operation could be adapted for more optimal stream focusing.

The color visible in the Fluent model represents fluid velocity of the sample stream. Blue shows slower flow streams and red faster. The sample stream flow velocity increases through the fluid junction, because the buffer fluid occupies volume and both exit the junction together. The fluid must flow faster to move the combined fluids out with the same fluid volume flow rate with which the fluids are entering the junction.

**Figure 4.** Sample fluid flows in the X direction of the fluid junction model in an (**a**) oblique view, (**b**) side view, and (**c**) top view, where buffer fluid is transparent and sample fluid is colored from dark blue to red, representing the range of flow velocities from slow to fast. A uniformly distributed particle concentration at the sample inlet becomes a focused particle distribution at the outlet.

The simulation was performed with sample and buffer inlet velocities of 1 cm/s and an outlet pressure of 0 Pascals. Fluid dynamics of the microchannels dictate laminar flow, thus the shortened channels in the simulation. We expect focused flow to propagate down the channel for optical interrogation. However, if the particles are diffusing outside of the focused stream, this behavior would not be accounted for by the fluid dynamic software, as seen in hundred-micron channel length models.

That the modeling conditions do not resemble the experimental conditions was a conscious choice, because the channel length is more than two orders of magnitude greater than the cross-section. Modeling the entire length of the fluid channels from inlet reservoirs to outlet reservoirs, especially given the negligible inertial focusing effects, is unhelpful. Instead, the focusing junction was modeled with the inlet velocities and outlet pressure and the channel dimensions on chip were determined by volumetric flow ratio to achieve the modeled velocities. This was expected to result in the predicted hydrodynamic effect by using negative pressure-driven flow. The experimental results show focused stream flow velocities on the same order of magnitude as predicted.

We modeled the design with some variations to portray how certain changes to the design affect the hydrodynamic focusing. First, we removed the pit feature in Figure 5a to show how it functions for vertical focusing. Without the pit, strong horizontal focusing occurs as the buffer fluid squeezes the sample stream from the sides. Subsequently, the trench feature raises up and further squeezes the sample stream. The resulting sample stream is what might be called 2.5D focused, as it is, in fact, limited in position in the fluid channel, but only focused on three sides instead of four. The pit is necessary to move the sample stream downward as buffer fluid fills the top of the channel, as described earlier.

We investigated how controlling fluid flow might enable us to overcome the suboptimal nose shaped sample stream seen in Figure 4. By increasing the buffer fluid input velocity, we expected a greater amount of vertical focusing as a result of the sample stream being more fully covered in the pit region. Figure 5b shows two and a half times the velocity, 2.5 cm/s, where the bridge of the nose shape disappears. Note that sample flow velocity increases with an increasing buffer flow, as all of the fluid entering the junction must leave through one port at the same volumetric flow rate that it enters at the other five ports combined. The increase in buffer fluid limits the amount of sample fluid that can be processed through the device per time, which should be considered in any time-sensitive optofluidic application, for instance, when screening an entire volume of liquid, which contains particles of interest. However, the increase in fluid velocity might compensate for this.

**Figure 5.** Variations were modeled to show how fluid channel shape and function alter the hydrodynamic focusing effect. (**a**) No pit leaves the horizontally focused sample stream near the top of the channel, a sort of 2.5D HDF. (**b**) Increasing buffer flow by 2.5 times better shapes the sample stream and removes the bridge of the nose shape.

The current design could be adapted to induce optimal flows by increasing the volumetric flow ratio of the buffer fluid. This could be performed with fluid inlet velocity control rather than negative pressure driven flow at the outlet, or the straightforward addition of back pressure at the buffer inlets while using gravity or some other means of positive pressure to push more buffer fluid through [33]. However, high operational vacuum pressures dominate such gravity-driven back pressure effects, as will be discussed.

#### *2.4. Fabrication*

The optofluidic device was fabricated in a class 10 cleanroom while using standard silicon-based microfabrication processes. Figure 6 shows side view drawings of the critical process steps. First, commercial layers of silicon dioxide and tantalum pentoxide were deposited on a blank silicon wafer at thicknesses of 265 nm and 102 nm, respectively. These dielectric layers form the anti-resonant reflecting optical waveguide structure (ARROW) required for waveguiding in a low refractive index medium such as a liquid. Next, chrome was deposited on the wafer and patterned with AZ 3330 photoresist and chrome etchant to form a stop etch feature for a later step. Next, nickel was deposited on the wafer and again patterned with AZ 3330 and nickel etchant. The features were ICP-RIE etched, forming the low-pressure chamber (pit) and trench to 12 μm deep. Here, the sacrificial material of SU-8 10 photoresist was deposited and patterned at 3000 rpm spin speed and 30 s exposure time, filling the etched features and forming the liquid-core channel volume. Note that the 6 μm tall resist overlaps all of the edges of the etched features by about 5 μm to ensure coverage and allow for some fabrication tolerance. Next, a self-aligned pedestal was created by depositing and developing AZ 4620 off the top of the SU-8 resist, then depositing nickel before removing the remaining resist with acetone. This liftoff step leaves the nickel mask on top of the defined pedestal to act as a dry etch mask for a 6 μm pedestal. The oxide was deposited next as low stress, high refractive index plasma-enhanced chemical vapor deposited (PECVD) silicon dioxide. Nickel was deposited again, patterned, and the

oxide was then etched to form the liquid-core walls and solid-core bodies. An additional low stress, yet low refractive index PECVD silicon dioxide layer was deposited as an environmental barrier [34]. Finally, the sacrificial material was exposed at the ends of the channels with buffered hydrofluoric acid while using AZ 4620 as a mask, and the SU-8 was etched out with a mixture of hydrogen peroxide and sulfuric acid at a ratio of 3:2.

**Figure 6.** The fabrication flow chart shows cross-section drawing on critical silicon-based microfabrication processes used to build the device, with silicon substrate in gray, thin metal etch masks in gray, anti-resonant reflecting optical waveguide structure (ARROW) layers in green, SU-8 photoresist in orange, and PECVD oxide in blue. (**a**) ARROW layers enable liquid-core waveguiding, and (**b**) the low-pressure chamber (pit) and trench features are etched before (**c**) SU-8 photoresist forms the fluid volume. (**d**) A pedestal helps to improve yield and isolate the optics and (**e**) PECVD oxide forms the walls. (**f**) The liquid-core and solid-core waveguides are etched and (**g**) cladding PECVD oxide protects it all. (**h**) Finally, acid removes the sacrificial SU-8 photoresist, leaving the channels hollow.

The critical features of a completed device are seen from the top down in Figure 7, showing the fluid focusing junction, as well as the liquid-core channel being intersected by solid-core waveguides. Note how the solid-core features in Figure 7a rest on a pedestal wider than themselves, while the pedestal of the liquid-core channels is not visible from the top, because it only lies directly underneath these features. The pedestal strengthens the hollow channels as it moves the crevice of the oxide all the way to the etched substrate surface. Figure 7b,c show the top of the focusing junction oxide as well as the pit and trench features inside the focusing junction.

**Figure 7.** Optical three-dimensional (3D) profilometer images of a completed device center show (**a**) the liquid-core focusing junction and channel and the intersecting solid-core waveguides as well as (**b**) the top of the focusing junction oxide and (**c**) the inside of the focusing junction showing the etch features.

#### **3. Results**

#### *3.1. Experimental Methods*

The sample and buffer fluid flow from left to right and hydrodynamic focusing occurs in the fluid junction shortly before the optical excitation region as seen from top down in Figure 8, which correlates to the center of the chip layout found in Figure 3. It is at this excitation region where the liquid-core waveguide fluid channel and the solid-core MMI waveguide intersect and the light-matter interaction occurs by multi-spot excitation (see Figure 8b), which enables the direct determination of the particles' flow speed. The solid-core waveguides are oriented perpendicular to the edge of the chip, with the excitation waveguides vertical and the collection waveguide horizontal. Note how the liquid-core waveguide guides fluorescent signal emission to the collection waveguide, which is coupled to the liquid-core and guides the signal to the chip edge, where it can be collected and analyzed.

**Figure 8.** (**a**) Top down schematic of the center of the optofluidic chip. Fluid flows from left to right in the dark blue liquid-core channels, passing through the focusing junction before interacting with the light delivered by the light blue solid-core waveguides. (**b**) Photograph of on-chip multi-mode interference excitation waveguide (MMI) multi-spot excitation scheme using fluorescent dye and DI water.

Three-dimensional focusing was confirmed and the diffusion of the focused sample stream in buffer fluid was characterized in a previously published design similar to this one [31]. The present design shares liquid-core waveguide dimensions (12 μm × 6 μm cross-section), functional method (buffer fluid focusing), and operational flow rates (~3 cm/s), as well as modeling parameters, fabrication processes and materials, and testing methods. Good agreement between predicted and actual streamlines found in the previous design can thus translate to the current design. Moreover, the inertial focusing effects are found to be negligible by calculating the maximum Reynolds number to be less than one. The low diffusivity of the fluorescent beads in conjunction with the predicted laminar flow stream, previous experimentation, and calculated low Reynolds number make us confident that particles remain in hydrodynamically focused streams. Focusing is indicated by an analysis of signal distributions rather than repeating previous experiments.

The previous experiments were performed with Cy5 fluorescent dye and water which were drawn through the device with negative pressure and illuminated with a 633 nm wavelength laser coupled to the chip through a single mode optical fiber (Newport, Irvine, CA, USA, FS-V). The photonic signal passed through a penta-bandpass optical filter (Semrock, Rochester, NY, USA, FF01-440/521/607/694/809-25) before it was detected by a Time-Correlated Single Photon Counting (TCSPC) avalanche photodiode (APD) (PicoQuant, Berlin, Germany, TimeHarp 260 Nano) and analyzed using our event detector scripts developed in Python and available on github (San Francisco, CA, USA) (github.com/vganjali/EventDetector). It utilizes continuous wavelet transform (CWT) with custom made wavelets to enhance the signal in both time and scale domains, followed by the threshold

method to localized events accurately in time and scale. Scale information is designed to match with the Δt value that was related to subsequent bright spots for a given multi-peak signal (representing the multi-spot illumination pattern shown in Figure 8b) and it is used as a means to determine the velocity of the detected fluorescent beads [35]. If using diffusible samples such as dye is desired, design constraints should be considered. This includes locating the focusing junction nearer the excitation region and flowing the sample fluid at the highest possible velocity (negative vacuum pressure) to mitigate the lateral spreading of the focused stream.

The same apparatus, as shown in Figure 9, was used for evaluating the present design, but here we used 200 nm dark red fluorescent beads (FluoSpheresTM Carboxylate-Modified Microspheres from InvitrogenTM, Carlsbad, CA, USA) diluted to 107/mL concentration to elicit discrete signal peaks for analysis. The laser light is coupled to the multi-mode interference excitation waveguide (MMI), which was designed to generate a 75 μm long multi-spot illumination pattern at the intersecting liquid-core waveguide, as seen in Figure 8b.

**Figure 9.** The test system includes the optofluidic chip connected to a vacuum for generating negative pressure to pull sample and buffer fluid through. A 633 nm laser is coupled to the excitation waveguide through an optical fiber to elicit fluorescent photon emission which is collected and detected by an avalanche photodiode (APD) for analysis.

The lateral spreading of molecules in a focused stream is approximately proportional to the square root of diffusivity. The beads have a reported diffusivity of about 2 <sup>×</sup> <sup>10</sup>−<sup>8</sup> cm2/s, two orders of magnitude lower than the dye [36,37]. Two-hundred nm beads should exhibit little lateral spreading due to diffusion and be useful in characterizing the detection enhancement of the 3DHDF design. This means that the beads are representative of molecules with similar or even higher diffusivity, such as DNA, which are of interest in optofluidic interrogation.

#### *3.2. Fluorescent Signal Coe*ffi*cients of Variance*

The effect of 3DHDF on fluorescent signal quality was measured by first filling all of the fluid channels with a fluorescent bead solution as a control test. In this case, there is no effective focusing of a particle stream because the entire cross section of a fluid channel is filled with particles. Measured fluorescence for this case was then compared to the case, where the sample fluid channel was filled with solution containing beads and the buffer channels with water. In both cases, fluid was drawn through the optofluidic channels with negative pressure at the outlet and illuminated with multiple spots of 633 nm laser light through the solid-core MMI waveguide. An avalanche photodiode at the chip edge collected the fluorescent signal captured orthogonally in the liquid-core waveguide and a computer reported the intensity in counts per 0.1 ms, as observed in the traces in Figure 10. The control intensity trace in Figure 10a represents a uniformly bead filled fluid channel and Figure 10b represents a focused bead stream that was sheathed by buffer fluid.

**Figure 10.** Signal intensity traces in counts per 0.1 ms were collected for (**a**) unfocused operation over 150 s and (**b**) focused operation over 300 s. (**c**–**f**) Histogram distributions for signal intensity and fluid velocity show more consistent signal with three-dimensional hydrodynamic focusing (3DHDF).

These plots show the measured optical intensity versus time, with intensity spikes representing fluorescing beads passing through an excitation point. When comparing the control trace to the experiment trace we see an almost sevenfold decrease in events captured per time. This is likely due to the buffer fluid in the 3DHDF case occupying a large fraction of the channel volume and reducing the number of beads passing through the channel per time. Figure 10c–f show histogram distributions of both signal intensity and fluid velocity for the control and 3DHDF cases. For both the intensity and velocity distributions, we see a narrowing due to hydrodynamic focusing, which is expected.

An improvement in detection performance can be represented by the coefficient of variance (CV) for the measured signal. This is calculated by dividing the standard deviation of the signal intensity by the mean of the distribution. A smaller CV represents a narrowing of the distribution or an increase in mean. They both indicate enhanced detection of optofluidic sensors promised by HDF. The data found in Figure 10c–f were used to generate CV values for characterizing detection enhancement. The resulting signal intensity CVs are 0.12 and 0.08 for the control and experiment, respectively, a decrease of 33%, which is one way to quantify the increase in signal uniformity that is afforded by 3DHDF. The calculated CV for the fluid velocity distributions are 0.09 and 0.07 for the control and experiment respectively, a decrease of 22%, again with 3DHDF showing improved uniformity. The data were obtained while using −15 inHg vacuum-pressure-driven flow to overcome the reservoir effects while maintaining high signal intensity, as will be discussed shortly.

#### *3.3. Pressure Dependent Detection Enhancement*

We wanted to test how the mean and standard deviation of signal traces that were collected at various vacuum pressures would relate. Increasing the vacuum pressure results in a greater pulling force on the fluid through the channels, and thus higher fluid velocities and a lower chance of detecting all emitted photons. This is related to the time resolution and dead time of the detector. Fluorescence happens on the nanosecond scale, which means the beads should have enough time to be excited and emit fluorescent signal, but the detector is unable to detect all of the received photons in the short time interval that the fluorophore spends in the excitation volume. The result is lower signal intensities. We changed the pressure in increments of 5 from −5 inHg to −25 inHg as well as −28 inHg, the limit of the vacuum system. Figure 11a shows the normalized mean and standard deviation values of the signal intensity and fluid velocity. As expected, the velocity increased and the intensity decreased with

increasing vacuum pressure magnitude. The mean fluid velocity ranged from approximately 1 cm/s to 5.4 cm/s.

**Figure 11.** (**a**) With increasing vacuum pressure, the mean (μ) and standard deviation (σ) of the velocity increase while those of the intensity decrease (**b**) The coefficient of variance (CV) decreases with increasing vacuum pressure for both signal intensity and particle velocity before flattening out. Back pressure widens the focused sample stream at low operational pressures but is dominated at higher pressures.

Figure 11b plots the coefficient of variance for intensity and velocity for each vacuum pressure. The trend observed is a result of reservoir effects. This is seen as higher CV values at low operational vacuum pressures that drop with increasing vacuum pressure before flattening out. The sample channel has higher fluidic resistance than the buffer channels, due to the sample channel being narrower than the buffer channels. This means the buffer channels consumed reservoir fluid volume faster than the sample channel, which results in an increase in back pressure on the sample fluid due to the greater height of fluid in the column [33]. During experimentation, an unbalance of volumetric flow ratio occurred, resulting in a less tightly focused sample stream (higher CV). The reservoir effect is dominant at low pressures, but it becomes negligible at higher vacuum pressures, being seen as the flattening of the CV line. The flat portion matches expectations as the device is designed to operate over a range of velocities with negligible focusing change.

We concluded that a minimum of −15 inHg vacuum pressure should be used to overcome the reservoir effects that were observed in the testing of the device. This finding, in conjunction with the signal intensity data found in Figure 11a, led us to perform device characterization at −15 inHg vacuum pressure (as shown in Figure 10) to overcome the observed reservoir effects while maintaining the greatest signal intensity. Negative pressure driven flow schemes used for optofluidic detection should similarly balance signal variance with signal intensity to achieve optimal detection schemes.

We compared the device yield of the silicon-based microchips in over 300 total devices. The present design showed an average of 45% yield when compared to the average 29% yield of the previous design, a 1.56 times improvement in device yield. We attribute this yield improvement to the robust design while using a single-layer of photoresist to form the fluid channels.

#### **4. Conclusions**

A 10-μm scale, single layer 3D hydrodynamic focusing design was implemented, which improved device yield by 1.56 times over previous work. The low-pressure chamber, or pit, is the key feature that induces 3D focusing using buffer fluid that is delivered by negative pressure-driven flow, controlling the position of target particles in the fluid channel during fluorescent signal generation. This mitigates the disadvantageous natural phenomena that are present in microfluidic channels by narrowing signal intensity and flow velocity distributions and constraining target particles to highest collection efficiency regions. It does this while using just four fluid reservoirs. The concept, modeling, fabrication, and

testing were outlined. Fluorescent beads representing molecules of interest were analyzed to find 33% more consistent signal with focusing. The reservoir effects were observed at low operational vacuum pressures and an optimal detection scheme was used to maximize signal consistency while maintaining high signal intensity. Signal intensity and flow velocity coefficients of variance matched expectations by decreasing between the control and the experiment, indicating successful focusing and detection enhancement. Further improvement and optimization in signal intensity can be achieved by aligning focused particle streams with the optical excitation mode center and with collection modes in the liquid-core waveguide. This could be achieved with the current geometry while using independent fluid inlet controls to flow additional buffer fluid volume. A gravity driven back-pressure method would be ineffective at higher operational vacuum pressures as the effect becomes dominated. Alternatively, the pit approach could be redesigned to efficiently operate with negative pressure-driven flow.

**Author Contributions:** Conceptualization, E.S.H., J.G.W., and A.R.H.; methodology, E.S.H.; validation, E.S.H., V.G., and H.S.; formal analysis, E.S.H., V.G., and H.S.; investigation, E.S.H. and V.G.; resources, A.R.H. and H.S.; data curation, V.G.; writing—original draft preparation, E.S.H.; writing—review and editing, E.S.H.; visualization, E.S.H.; supervision, A.R.H.; project administration, A.R.H. and H.S.; funding acquisition, A.R.H. and H.S. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by the National Institutes of Health, grant numbers 1R01AI116989 and 1R01EB028608, and by the National Science Foundation, grant number CBET-1703058. The APC was funded by the National Institutes of Health, grant number 1R01AI116989.

**Conflicts of Interest:** A.R.H. and H.S. have a financial interest in Fluxus Inc., which commercializes optofluidic technology.

#### **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/).
