## Introduction

Increased demand for clean water requires efficient strategies to purify water. Multiple water purification technologies have been developed, including physical processes or biochemical processes such as filtration, agglomeration, and desalination1,2,3. Among these, photocatalytic water purification is especially promising, given that photocatalysts are environmentally friendly materials that degrade or mineralize various pollutants with light being the only energy input. When a semiconductor catalyst absorbs photons with an energy equal to or greater than the band gap of the catalyst, photoexcited electron-hole pairs are generated. The reactions of electron-hole pairs with surrounding oxygen and water molecules form reactive oxygen species such as superoxide and hydroxyl radicals that could decompose organic chemicals and bioparticles into innocuous products such as CO2 and H2O4,5,6,7.

The performance of water purification using photocatalysts is mainly dependent on photocatalytic efficiency and mass transfer efficiency. The former relates to light absorption and charge transport in photocatalysts, whereas the latter reflects the contact between the catalyst and the reactants, in addition to the removal of the products. To significantly enhance the photocatalytic efficiency, photocatalysts have been fabricated in various nanoscale geometric structures to provide large surface-to-volume ratios in the form of nanorods, nanotubes, and nanofibers using hydrothermal processes, sol-gel processes, or electrospinning8,9,10,11,12,13,14,15. However, the exchange with the catalyst is still limited as reactants and reaction products remain on the photocatalysts, which prevents further improvement of the water purification performance. To facilitate mass transfer efficiency, an optofluidic microreactor has been proposed, which combines microfluidics and photocatalysis16,17,18,19,20. While the solution is passed through the microfluidics channel, the photocatalyst causes a catalytic reaction to decompose various chemicals when exposed to light (Fig. 1a). The light and fluid in the optofluidic microreactor are characterized by a short molecular diffusion path, an intrinsically large surface-to-volume ratio, uniform illumination, and efficient light harvesting21,22,23,24,25. The light-harvesting can be further improved using nanostructures of photocatalysts that scatter light, or by modifying the base substrates to increase the optical path and thereby increase the number of electron-hole pairs that are generated due to multiple interactions with the photocatalyst26,27. In order to further enhance the performance of the optofluidic microreactor, which is dependent almost exclusively on diffusion, an innovative alternative is required to simultaneously increase the photocatalytic and mass transport efficiency.

In microfluidic channels, which generally are in the laminar flow regime due to low Reynolds numbers, efficient fluid mixing requires a chaotic flow produced by micromixers. A micromixer is a device based on mechanical microparts used to enhance the overall mixing capacity of the system by increasing the contact area between the reactants28,29. Based on the underlying principle, micromixers can be divided into two categories: active and passive. Active mixers use an external energy source to force the mixing of fluids inside the microchannel by electric, magnetic, or ultrasonic means. Passive mixers do not require external energy other than a pressure difference to drive the flow where the mixing is performed. In particular, mixing and transport is achieved by chaotic advection generated by complex microchannel geometries, such as multidimensional structures, patterned groove structures, and split-and-recombination structures29,30,31,32,33. The patterning of the surfaces of the micromixer using specially designed microstructures is a common strategy for enhancing the mixing performance of a passive micromixer34,35,36.

In this study, a planar photocatalytic microreactor that possesses hierarchical structures both at the micro- and nanoscale is presented for water purification using solar energy (Fig. 1b–f). Rhodamine 6 G (Rh6G) solution was used as a representative organic dye pollutant, which is highly corrosive and may have serious ecological effects on the aquatic environment37,38. When the Rh6G is degraded with a photocatalytic reaction, it is mineralized to carbon dioxide (CO2) and water (H2O), which are less toxic organic by-products that are environmentally benign. To maximize the water purification performance, the benefits of the nanostructured TiO2 photocatalyst and the micromixer design of the channel were combined. The micromixer design with asymmetric herringbone-shaped grooves was applied to the microchannel to significantly speed up the mixing of the fluids in the low Reynolds number regime by inducing chaotic advection. The influence of the reactor’s substrate on which photocatalysts were deposited, the structures of photocatalytic materials, the flow rate and the microscale geometry on the reaction efficiency were systematically investigated. Nanostructured TiO2 photocatalysts were fabricated by a physical vapor shadow growth technique, also known as glancing angle deposition (GLAD), which can produce highly porous nanostructured films on substrates. A numerical simulation of the flow inside the channel was carried out, allowing conclusions to be drawn about the hydrodynamic mechanism by which the reaction takes place in different cases. The result is an optimized design in which the nanostructures of the photocatalyst improve the water purification efficiency, while the herringbone micromixer further improves the reaction by creating a vortex along the reactor and thereby aids the molecular transport to and from the surface of the photocatalyst. Lastly, to demonstrate the capability of the reactor, more persistent contaminants, sulfamethoxazole, and polystyrene microplastics were successfully degraded. This study contributes to the systematic identification of the underlying physical mechanisms of the hierarchical design combining the nanoscale design of photocatalysts and the microscale design of mixer systems. In addition, the results of this study can contribute to providing solutions for more efficient water purification systems. The planar microreactor may also be of use in the general use of photocatalysts for energy applications as well as chemical and bioprocessing.

## Results and discussion

The effects of the photocatalyst structures and the substrate materials on which the photocatalysts were deposited were studied using a planar photocatalytic microreactor that has the photocatalysts on one side and a flat microchannel on the other side. The effects of flow mixing on the reaction efficiency were further investigated using a microreactor possessing a microchannel with herringbone grooves on top of the photocatalysts.

### Enhancement of photocatalytic efficiency by helix nanostructure

To study the effects of the photocatalyst structure on water purification performance, two different types of substrates were prepared, one with a dense TiO2 film (TiO2 DF) and another with TiO2 nanohelices (TiO2 NHs). A TiO2 DF and the TiO2 NHs with respective thicknesses of 400 and 450 nm were formed on the Si substrate (Fig. 1d, e and Supplementary Figs. 1, 3). The TiO2 NHs exhibited a porosity higher than 60% with a very high surface-to-volume ratio, as estimated from a linear effective medium approximation of the refractive index value of the film (Supplementary Fig. 2)39. In addition, the charge transport efficiency of the TiO2 NHs is high, because of the near-single crystalline property, as confirmed in previous studies40,41. XRD and UV-Vis absorption spectra show that the TiO2 in this study has anatase structures with an optical band gap of around 3.25 eV (Supplementary Figs. 4, 5). The XPS data reveals the two strong peaks from the TiO2 NHs at around 465.5 and 459.8 eV, which are attributed to Ti 2p1/2 and Ti 2p3/2, and the peaks of O 1 s are located at about 531.0 eV, whose energy is equal to the O 1 s electron binding energy for TiO2 (Supplementary Fig. 6).

In a stationary state without flow for 90 min, the TiO2 NHs demonstrated the highest degradation (34.03 ± 1.32%), followed by the DF surface (22.66 ± 1.27%), whereas almost nothing was degraded without TiO2 (Fig. 2a). Although TiO2 NHs have a significantly larger surface area and fast charge transport properties, their degradation is only higher by a factor of ~1.5; thus, it is likely that the mass transport is the limiting factor for the degradation in the stationary state. Accordingly, an optofluidic microreactor was introduced to test this hypothesis and to improve the mass transfer by applying convection flow and thus increase the photocatalytic efficiencies.

We fabricated two optofluidic microreactors using an unstructured TiO2 DF and TiO2 NH substrates as reaction surfaces, respectively, and compared their water purification capabilities. The microreactors had a flat and rectangular reaction chamber consisting of TiO2 DF or NHs substrates on one side and a flat PDMS microfluidic slab on the other side. The Rh6G solution was introduced into the TiO2 dense film-based microreactor (DF reactor) and TiO2 nanohelix-based microreactor (NH reactor) at various flow rates in the range of 45–630 μL/min. When the solution passed through the DF reactor under UV illumination, Rh6G molecules in the solution were decomposed by photocatalysis at the bottom surface. Under these conditions, convection dominates as opposed to diffusion, considering the Péclet number Pe = $$\frac{{u}\mathcal{h}}{D}$$ where u, $${\mathcal{h}}$$, and D are the flow velocity, channel height, and diffusion coefficient, respectively, which is a measure of the relative importance of convection between diffusion, of the system, ranged from 180–2500 when the diffusion coefficient of Rh6G in water was taken as $$2.8 \times 10^{ - 6}{{{\mathrm{cm}}}}^2/{{{\mathrm{s}}}}$$42. Given that the reaction of the photocatalyst occurs on the surface of the reactor, the reaction rate is relatively low in the case of a DF reactor with a small surface area. During the passage through the reactor, approximately 30% of the Rh6G was degraded at a low flow rate of 45 μL/min, and with an increase in the flow rate to 200 μl/min and higher, less than 10% of the Rh6G was decomposed (Fig. 2b). There was a decrease in the degradation with an increase in the flow rate, given that the residence time in the reactor was reduced accordingly. The reaction rate constant k, which refers to the purification capacity per unit time, was a low 0.26 ± 0.07 min−1 and almost constant, regardless of the flow rate (Fig. 2c). The TiO2 nanohelix-based microreactor (NH reactor) demonstrated significantly improved water purification capacity than that of the DF reactor. At a low flow rate of 45 μl/min, almost all the dye molecules were degraded with 90% of the average degradation; and at a very high flow rate of 600 μL/min, the average decomposition efficiency was maintained at 21.6%. The reaction rate constant ranged from 2.10–3.42 min−1 with an increase in the flow rate from 45–630 μL/min. Compared with a DF reactor, the reaction rate constant of the NH reactor clearly exhibited a value that was higher by a factor of 13.23 on average. Please note that the reaction rate constant increased with an increase in the flow rate.

This result confirms that nanostructures of TiO2 cause a remarkable improvement in the purification capacity in the photocatalytic reactor, and the convective flow resolves the mass transfer deficiency. The reaction capacity is determined mainly by two factors: the intrinsic photocatalytic efficiency and the external mass transfer efficiency of Rh6G from the liquid to the TiO2 surface43. The intrinsic photocatalytic efficiency depends on the material properties and geometry of the TiO2, and it is independent of the flow conditions. However, if the intrinsic photocatalytic efficiency is high, and critical reactants such as Rh6G, oxygen, and water necessary for the reaction are not exchanged quickly enough; the reaction efficiency is lowered1. This is indicated by the results, in that the reaction rate constant of the NH reactor was higher than that of the DF reactor under convection flow by a factor of approximately 13; whereas, in the stationary state, the degradation of the NHs was only 1.5 times higher. As is seen in Fig. 2c, the photocatalytic reaction rate constant of the flat surface, which has a lower reaction rate constant, to begin with, does not improve with the flow, whereas the TiO2 nanohelix surface with a high reaction rate constant is significantly improved as the reactants are replenished by the flow.

### Enhancement of photocatalytic efficiency by base substrates

The light-harvesting efficiency is a critical parameter for determining water purification performance. In microreactors, light is typically transmitted through a glass substrate, which causes some losses of solar energy. A dual-film optofluidic microreactor was proposed to enhance light harvesting by re-utilizing transmitted light at the bottom10. An alternative method involves the doubling of the optical path using reflective metallic substrates. It should be noted that TiO2 photocatalysts can be coated on such metallic substrates; however, these may influence the phase and microstructure of the TiO2, particularly for dense films, due to thermal stresses in the annealing process44.

In this study, TiO2 NHs were deposited on substrates, which are less influenced by thermal stress, given that each NH had a contact area of less than 50 nm with the substrate. We prepared four types of optofluidic flat microreactors with different base substrates, namely, glass (Gl) substrate, Silicon (Si) substrate, Aluminum coated substrate (Al substrate), and Platinum coated substrate (Pt substrate), on which the TiO2 NHs were deposited; and thereafter investigated their effects on the water purification capacity at different flow rates under UV light. As shown in Fig. 3a, the NH reactors with Pt, Al, and Si substrates exhibited significantly higher performances than those with the Gl substrate. This is because a portion of the light is reflected from Pt, Al, and Si substrates to interact again with the NHs; whereas, most of the light is transmitted through or absorbed by the Gl substrates45,46. It should be noted that NH reactors with Pt and Si substrates exhibited higher efficiencies than those with Al substrates. Given that Al exhibits a significantly higher reflectivity, especially in the UV region, this was not expected. This is apparently due to the oxidation and degradation of Al in the measurement environment and confirmed by the re-usability test results. It should be noted that TiO2 NHs deposited on Pt, Si, and Gl substrates can be reused multiple times, even after several days (Fig. 3b). However, in the case of Al substrate, the efficiency abruptly decreased upon re-use on the day after a single use. At the initial usage, 83.8% of the dye solution was decomposed at a low flow rate, and 35.1% at a high flow rate. However, in the subsequent usage, it decreased to 34.2 and 9.5% at low and high flow rates, respectively. Therefore, Al is not a good candidate for water purification applications due to its oxidation.

There are very few studies that have tried to increase the light-harvesting efficiency of the optofluidic microreactor. Li et al. introduced a dual-film optofluidic microreactor with transparent glass as a base substrate, and a double-sided TiO2 nanorod reactor showed improved performance compared to a single-sided reactor10. However, there are few reports of using base materials other than glass, and metal-based substrates in this study showed much-improved efficiency than glass substrates. Among those, the Pt-coated substrates demonstrated the highest performance with a reaction rate constant that was higher than that of the Si-coated substrate by ~2.5. Given that Pt and Si have similar UV reflectance values in the range of 300–400 nm, as employed for the proposed water degradation system45,46, the Pt or Pt/TiO2 NH interface could have an influence on water degradation. In particular, Pt can also function as a catalyst, thus resulting in a significantly higher photocatalytic efficiency. Furthermore, the internal photoemission, which describes the optically induced charge carrier transition at the interface, can play a role47. It is expected that a Schottky junction is formed at the interface of Pt and TiO2 NHs. Considering the work function of Pt (5.8 eV) and the electron affinity of TiO2 (4.0 eV), this results in a Schottky barrier height of approximately 1.8 eV48,49. Thus, photoexcited electrons from Pt can overcome the Schottky barrier under UV illumination and contribute to photogenerated charges in TiO250.

### Enhancement of mass transfer efficiency by herringbone micromixer

To further enhance the mass transfer efficiency, a herringbone micromixer was used in the reactors51. The water purification performances of the DF and NH reactors with the mixer were compared by measuring the Rh6G degradation capabilities of the reactors at various flow rates under UV-light illumination. In the case of the DF reactor, the degradation and reaction rate constant of the DF reactors with and without the mixer were almost identical, which suggests that the application of the mixer did not have a significant influence on the purification capacity. In the NH reactor, however, the mixer increased the degradation and reaction rate constant by 12 ± 5% and 17 ± 5% on average, respectively (Fig. 4). At a low flow rate of 45 μL/min, the NH reactor with the mixer degraded 88.9% of the Rh6G molecules in the solution, which was 2% more efficient than in the absence of the mixer. At a higher flow rate of 630 μL/min, 25.3% of the molecules were degraded, which was 17% higher than that without the mixer. This increasing trend was clearly observed in the reaction rate constant, by which the mixer increased the reaction rate constant by 6% at a low flow rate of 45 μL/min, and 20% at a high flow rate of 630 μL/min. Thus, with an increase in the flow rate, there was a greater increase in the water purification performance of the herringbone mixer.

To determine the mechanism by which the herringbone mixer improved the water purification capacity, a finite element-based numerical simulation was performed to visualize the fluid flow behavior in the reactor. The Rh6G degradation within the DF reactor and NH reactor was simulated based on the same geometry, where the reaction rate constant of each surface was set differently, thus reflecting the intrinsic reaction rate. To simplify the problem, the entire microreactor was divided into smaller segments that represented the periodic structure of the entire herringbone mixer design. Thus, the segmented model of the herringbone mixer had a width of 1 mm and a length of 7.1 mm, which were 1/15 and 1/4 of the width and length of the entire channel, respectively. The model without the mixer had a rectangular channel with a width of 1 mm, length of 7.1 mm, and height of 100 μm. The Navier–Stokes equation for incompressible flow was used to simulate the convective flow-through the microchannel. The Reynolds numbers ($$Re = \rho Uh/\mu$$, where ρ, U, and μ are the density, velocity, and dynamic viscosity of the liquid, respectively, and h is the channel height) obtained in the study ranged from 0.038–0.5, thus indicating that the flow was in the laminar regime.

$$\rho \nabla \cdot {{{\mathbf{u}}}} = 0,$$
(1)
$$\rho \frac{{\partial {{{\mathbf{u}}}}}}{{\partial t}} + \rho \left( {{{{\mathbf{u}}}} \cdot \nabla } \right){{{\mathbf{u}}}} = \nabla \cdot \left[ { - p{{{\mathbf{I}}}} + \mu (\nabla {{{\mathbf{u}}}} + \left( {\nabla {{{\mathbf{u}}}}} \right)^T)} \right]$$
(2)

where u and I are the flow velocity vector and unit vector, respectively, p is the pressure, and t is the time. The density (=997 kg/m3) and viscosity (=1 mPa∙s) were set the same as those of water at 25 °C. The fully developed flow was supplied to the inlet at a flow rate (Q) of 45–630 μL/min towards the outlet (p = 0) on the opposite side. A no-slip boundary condition (u = 0) was applied to the upper and lower walls of the channel, and periodic conditions ($${{{\mathbf{u}}}}_1 = {{{\mathbf{u}}}}_2$$ and $$p_1 = p_2$$, where subscripts 1 and 2 indicate each side wall) were applied to both sides of the channel.

The change in the concentration of the solution passing through the microchannel was calculated using the following convection–diffusion equation:

$$\frac{{\partial c}}{{\partial t}} = \nabla \cdot \left( {D\nabla c} \right) - \nabla \cdot \left( {{{{\mathbf{u}}}}c} \right) + R$$
(3)

where c is the concentration of the solution, D is the diffusion coefficient of Rh6G in water ($$= 2.8 \times 10^{ - 6}$$ cm2/s)42, and R describes the sinks or sources of the concentration. The initial concentration was set to 0, and the liquid with $$c_0 = 1$$ was supplied at the inlet and removed through the opposite outlet ($${{{\mathbf{n}}}} \cdot D\nabla c = 0$$, where n is a normal vector for the outlet). The flux (J) was zero on the upper wall (n ∙ J = 0), and periodic boundary conditions were applied to the side walls ($$- {{{\mathbf{n}}}}_1 \cdot {{{\mathbf{J}}}}_1 = {{{\mathbf{n}}}}_2 \cdot {{{\mathbf{J}}}}_2$$). On the bottom surface, the concentration decreased at the following Nth order rate by the following chemical reaction with TiO2:

$$- {{{\mathbf{n}}}} \cdot D\nabla c = k \cdot c^N$$
(4)

where k denotes the catalyst activity. Moreover, k and N were used as fitting parameters to match the numerical results with the experimental results. Given that the entire microreactor channel was longer than the segment used for the numerical model by a factor of four, four iterations were repeated using the velocity field and concentration distribution at the outlet as the boundary conditions at the inlet of the subsequent iteration. The average concentration at the outlet of the previous iteration was used to calculate the degradation and the reaction rate constant.

The degradation was computed according to various reaction rates by changing the fitting parameters N and k in Eq. (4) to compare the numerical and experimental results (Fig. 4a). The empirical parameters were obtained from the simulation, as follows: N = 1.2, $$k = - 8 \times 10^{ - 7}$$ (for the DF reactor), and $$- 1.05 \times 10^{ - 5}$$ (for the NH reactor). The abovementioned parameters clarify the experimental results for the cases without and with the mixer, respectively, where the NH reactor had a reaction rate constant k that was higher than that of the DF reactor by a factor of ~13 (Fig. 2c). The mixer of the NH reactor increased the reaction rate constant by 22% at a low flow rate of 45 μL/min, and by 36% at a high flow rate of 630 μL/min. Moreover, there was almost no change in the value of the DF reactor in the cases with and without the mixer, thus indicating a similar trend to that of the experimental results (Fig. 4b).

The flow behavior of the fluid in the reactor helps clarify the influence of the herringbone mixer on the reaction efficiency, which strongly depends on the mass transfer of reactants from the solution to the photocatalyst surface. As illustrated in Fig. 5a, as there was no flow along the Z axis inside the reactor without the mixer, the transport of Rh6G from the upper region to the bottom surface was primarily dependent on diffusion. Relatively weak diffusion when compared with convection in microreactors, as indicated by the high Péclet number (180 < Pe <2500), results in a steep concentration gradient along the Z axis of the channel (Fig. 5c, d; without mixer cases with bold lines). The lowered concentration near the TiO2 bottom surface, which represents the deficiency of the reactants, could limit the reaction rate and reduce the Rh6G degradation. Moreover, the herringbone mixer generated four counter-rotating vortices along the flow direction in a laminar regime, with a primary large-scale vortex along the main channel, followed by a small-scale secondary vortex on the channel wall at the end (Fig. 5b). Given that the vortices could mix Rh6G from the upper to the bottom of the reactor, the mixer flattened the concentration gradient along the Z axis (Fig. 5c, d; mixer cases with dotted lines) and increased the reaction rate accordingly by eliminating the deficiency problem.

In the DF reactor, the degradation and the reaction rate constant changed slightly, even with the addition of the mixer, whereas the NH reactor further enhanced the purification by the mixer (Fig. 4a). In the DF reactor with a low intrinsic reaction rate, the concentration gradient of the reactants throughout the reactor was low, where a high concentration of 90% or more reactants was present at the bottom of the reactor, leaving no space for the mixer to effectively improve the reaction performance (Fig. 5d). If the active reaction of the catalytic surface leads to a shortage of the reactants, as indicated by the steep concentration gradient of the NH reactor with the mixer, the vortex generated by the herringbone mixer significantly improves the reaction performance by increasing the available reactants to the catalyst and removing the reacted products from the catalyst.

### Decomposition of various aquatic pollutants

To show the capability of our device to degrade various water pollutants, photocatalytic degradations of cationic dye Methyl orange (MO) and more persistent contaminants, sulfamethoxazole (SMX) and polystyrene (PS) microplastic were investigated52,53,54,55.

The MO solution was introduced into the TiO2 nanohelix-based microreactor (NH reactor) at various flow rates in the range of 45–630 μL/min under UV illumination. During the passage through the reactor, approximately 30% of the MO was degraded at a low flow rate of 45 μL/min, and with an increase in the flow rate to 630 μL/min, less than 5% of the MO was decomposed (Fig. 6a and Supplementary Fig. 8). The reaction rate constant, k, was a 0.27 ± 0.06 min−1 in average (Supplementary Fig. 8C). Although the degradation efficiency was lower than that of the anionic dye Rh6G, it showed a relatively high degradation potential. SMX is an antibiotic that has been consistently detected in wastewater and is necessary to be removed in order to avoid an increase in antibiotic resistance in humans. In the absence of flow, SMX was hardly decomposed with TiO2 NHs. Next, the SMX solution was introduced into the TiO2 NH reactor at various flow rates in the range of 9–180 μL/min under UV illumination. With the NH reactor, approximately 28% of the SMX was degraded at a low flow rate of 9 μL/min, and with an increase in the flow rate above 45 μL/min, less than 5% of the SMX was decomposed (Fig. 6b). Lastly, since the existing sewage treatment system cannot catch all microplastics, they can remain in the water for a long time, be absorbed by aquatic organisms, travel through the food chain, and finally act as a toxic substance to humans. In this vein, we demonstrated our present system is suitable for the degradation of microplastic, especially polystyrene (PS) microparticle. To study the performance of the PS particle degradation process, the photodegradation of standard polystyrene latex suspension, which was employed as the microplastic model, with 3.79 ± 0.23 × 108 particles/mL in Deionized water was examined. The PS particle solution was introduced into the microreactor with a flow rate of 2.25 μL/min by a syringe pump, and the photocatalytic reaction was carried out under UV-light illumination. The degraded PS particle solution was collected from the outlet of the microreactor, and their morphologies are observed by SEM. As a result, PS particles were clearly degraded in NH reactor, where the average size of the particle decreased from 106.03 ± 8.79 nm to 84.06 ± 17.19 nm (Fig. 6c, d).

Compared to the previous studies, where the degradation of the SMX or PS particle took several hours to days, our system showed relatively fast degradation, where the residence time of SMX and PS particles in the reactor were only around 5 and 20 min with a flow rate of 9 and 2.25 μL/min, respectively54,56,57,58. These results suggest that our system performs very well in removing water pollutants.

A hierarchical photocatalytic microfluidic reactor was designed and characterized for the photodegradation of Rh6G under UV irradiation. Table 1 summarizes the reaction rate constants for all types of microreactors at low and high flow rates. This study attempts to maximize water purification efficiency by simultaneously improving photocatalytic efficiency and mass transfer efficiency by combining the benefits of a nanostructured photocatalyst, light harvesting by base substrates, and a micromixer. The experimental results confirmed that the nanostructures of the nanohelix-based reactor enhanced the water purification capacity almost by 13 times under conditions of convective flow compared to the dense film-based reactor. The light-reflective metallic substrates also increased the purification capacity compared to the Gl substrate by enhancing the light-harvesting efficiency. In addition, the herringbone micromixer applied to the nanohelix reactor further contributed to the water purification performance by up to 20% by generating a vortex along the reactor, which was qualitatively confirmed by a numerical simulation. We additionally confirmed that our reactor, with this improved purification performance, can decompose organic matters as well as more persistent contaminants by demonstrating the photocatalytic degradation of the cationic dye methyl orange, antibiotic sulfamethoxazole, and polystyrene microplastics. What would normally take hours or days for pollutants to decompose was accomplished in minutes with our reactor system. By systematically demonstrating the mechanisms governing the hierarchical design of nanoscale photocatalysts and micromixer systems, the findings of this study contribute to the smart design of an optofluidic microreactor with potential applications in the fields of energy, environmental science, and biomedicine.

In order for this solar-based water treatment system to be used in the field, some hurdles have to be overcome. First, the amount of water that can be purified needs to be increased, possibly by connecting the reactor systems in parallel or in series. In addition, using a photocatalyst that reacts not only with UV-light but also more efficiently across the entire solar spectrum will increase the utility of the system. Also, even if the reactor could be reused several times in this study, nanoscale photocatalysts need to be manufactured more robustly in the future to guarantee product stability. In natural water, there are various elements widely present, such as anions and humic acids, which could potentially retard the photocatalytic reaction by inhibiting the formation of free radicals, reducing light transmittance, or adsorbing to active sites of the photocatalyst. Therefore, in the future engineering design, if TiO2 is to be applied for the treatment of wastewater in nature, the retardation effects of various substances in the natural environment should be concerned.

## Methods

### The TiO2 substrate preparation and characterization

Four base substrates were used: glass (Gl), silicon (Si), platinum (Pt), and aluminum (Al). For Gl substrates, microscope slides with a thickness of 1 mm were cleaned with acetone and isopropanol. The Si substrate was the wafer (boron doped, <100> orientation), cleaned in piranha solution (a mixture of three parts of concentrated H2SO4 and one part of 30% H2O2) for 30 min. For the metallic substrates, the piranha-cleaned Si wafers were first coated with a 5 nm thick Ti adhesion layer and then with a 20 nm thick Pt or Al layer (>99.99% purity) by e-beam evaporation. Thereafter, TiO2 nanostructures were fabricated using the GLAD technique. In GLAD, the obliquely incident TiO2 vapor flux produces small random islands of deposited TiO2 material. Behind these clusters, the subsequent vapor flux cannot reach due to a shadowing effect59. This causes the film to grow by means of slanted nanorods with high porosity. Furthermore, a film of nanohelices could be produced by rotating the substrate during growth and by controlling the substrate rotation speed59. In this study, TiO2 nanohelices (NHs) were fabricated with a substrate tilt angle of 87° at a deposition rate of 2 Å/s and a substrate rotation speed of 0.28 °/s. A TiO2 dense film (DF), as a reference sample, was also fabricated with a vapor flux incident along the surface normal (tilt angle of 0°) and a deposition rate of 2 Å/s. All films were annealed at 500 °C for 2 h in ambient air to induce crystallization into the anatase phase.

X-ray photoelectron spectroscopy (XPS, PHI X-tool, ULVAC-PHI, Inc.) with Al (mono) X-rays (25.1 W) was used to detect the chemical combination state of TiO2 NHs. Combined with the X-ray energy dispersive spectroscopy (EDX), scanning electron microscopy (SEM, Quanta 250 FEG; FEI Ltd.) was utilized to analyze the basic material compositions and visualize the morphologies of the NHs (Supplementary Figs. 13). The crystal structure of the NHs was analyzed by X-ray diffraction (XRD). XRD was carried out using a Rigaku D/ MAX2500 with Cu Ka radiation (Supplementary Fig. 4). The UV-Vis measurements were performed in a Varian Cary 4000 UV-Vis spectrometer. UV-Vis spectra was measured using TiO2 NHs on a double-side-polished sapphire substrate that is transparent in the region of the TiO2 band edge (Supplementary Fig. 5).

### Microreactor fabrication

The microreactor had a rectangular reaction chamber consisting of a TiO2 substrate on one side and a flat or herringbone-structured polydimethylsiloxane (PDMS) on the other. (Fig. 1). The herringbone grooves with alternating permutation arrangements were designed to induce a chaotic flow for enhanced flow mixing. The master mold for the PDMS reactor was fabricated from acrylic material using a milling machine. To prototype the mold, a PDMS elastomer base and curing agent mixed at a ratio of 10:1 were cast on the molds and incubated in an oven at 75 °C for 2 h. The cured PDMS slab was peeled off, and holes with a diameter of 0.5 mm were punched at the inlet and outlet of the slab. After plasma treatment on the TiO2 substrate and the PDMS slab, they were precisely aligned and bonded. Finally, tubes were connected to the inlet and outlet of the reactor, the inlet side tube was connected to a syringe pump for fluid injection.

The volumetric dimensions of the flat microreactor chamber were 1.5 cm (width) × 3.0 cm (length) × 100 μm (height) = 45 μL, and those of the herringbone reactor chamber were the same as the dimensions of the planar area. Moreover, the height was adjusted to ensure that the volume was the same as in the flat microreactor. The height of the herringbone groove was 50 μm, and the gap between the neighboring grooves was 100 μm. The angle between the longitudinal axis of the channel and the herringbone structure was 45°. The three-dimensional (3D) geometry and dimensions of the microfluidic chip are illustrated in Fig. 1. At the inlet and outlet, tree-branch-shaped microchannels were adopted to ensure a uniform flow over the entire reaction chamber.

### Experimental setup for photocatalytic degradation

To study the performance of the water purification process, the photodegradation of 0.05 M Rhodamine 6 G (Rh6G, Sigma-Aldrich) dye solution, which was employed as the representative organic pollutant model, was examined under UV-light irradiation. For a reaction without flow, the TiO2 dense film (DF) and TiO2 nanohelix (NH) substrates covered with 3 mL of Rh6G solution were placed in a petri dish and irradiated with ultraviolet (UV) light (Altis uvcure, South Korea, 365 nm) for 90 min. As a comparison, only 3 mL of Rh6G solution without TiO2 was also prepared and irradiated with the same UV-light. For the optofluidic reaction, Rh6G solution was added to the microreactor, and the photocatalytic reaction was carried out under UV-light illumination (310–390 nm, UVICO, Rapp OptoElectronic GmbH). The Rh6G solution was introduced into the device by a syringe pump (Harvard Apparatus, PHD 2000) and decomposed by photocatalysis while passing through the microreactor (Fig. 1a). The degraded Rh6G solutions were collected from the outlet of the microreactor, and their absorption spectra were analyzed using a UV–visible spectrophotometer (Sunrise, TECAN, Germany). The degradation of Rh6G was evaluated by monitoring the absorbance change of various Rh6G solutions at 664 nm before and after degradation. The degradation (%) and reaction rate constant (min−1) are defined as $$(1 - \frac{{{{\mathrm{c}}}}}{{c_0}}) \times 100$$ and $$- {{{\mathrm{ln}}}}(\frac{{{{\mathrm{c}}}}}{{c_0}})/t$$; where c, $$c_0$$, and t refer to the current and initial concentrations of the Rh6G solution and the effective reaction time required for all the solution to flow from the inlet to the outlet, respectively. The degradation of Rh6G within the microreactor channel was simulated by computational fluid dynamic modeling using COMSOL Multiphysics 5.4 (COMSOL Inc., Stockholm, Sweden). The detailed numerical experiment conditions are given in sections corresponding to the Figs. 4 and 5.

To study the capability of the herringbone microreactor, photocatalytic degradation of various aquatic pollutants, 0.05 M Methyl orange (MO, Sigma-Aldrich) cationic dye, 30 $$\mu {{{\mathrm{M}}}}$$ Sulfamethoxazole (SMX, Sigma-Aldrich), and standard polystyrene latex suspension (Malvern, USA, 0.1 μm), were tested. The degraded MO solutions were analyzed with their absorption spectra using a UV-Vis spectrophotometer. For the analysis of the degraded SMX solutions, a 1260 Infinity II Prep- high-performance liquid chromatography (HPLC) System (Agilent Technologies, Santa Clara, CA, USA) and an EC-C18 column (4.6×150 mm, Agilent Technologies) were used. The detailed HPLC conditions were as follows: 60:40 0.1% phosphoric acid/acetonitrile (v/v) mobile phase, 10 μL injection volume, 25 °C temperature, and 0.8 mL/min flow rate. The SMX peak at a retention time of 2.89 min was detected with a UV/Vis detector at 270 nm, and the height of the peak was measured for SMX analysis. The degradation of the PS microparticles was analyzed by measuring the average diameter of the particles based on SEM images.

### Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.