Modular Droplet‐Based Fluidics for Large Volume Libraries of Individual Multiparametric Codes in Lab‐On‐Chip Systems

Droplet‐based lab‐on‐a‐chip systems offer vast possibilities in the manipulation, guidance, tracking, and labeling of individual droplet‐based bioreactors. One of the targeted application scenarios is in drug discovery where millions of unique codes are required, which is out of reach for current technologies. Here, a concept for the realization of multiparametric codes, where information is stored in distinct physical and chemical parameters, is proposed and validated. Exemplarily, the focus is on the use of impedance and magnetic sensing by monitoring ionic concentration as well as magnetic content per droplet and droplet volume. Codes based on aqueous ferrofluid droplets are prepared using a tubing‐based millifluidic setup and consist of up to six droplets of different combinations of volumes and magnetic concentration. It is demonstrated that a droplet chain of three single droplets of different volumes with nine different magnetic nanoparticle concentrations accompanied with four different ionic concentrations per droplet offers up to 3 million unique codes. The developed fluidic platform can be readily extended to other types of sensors including optical ones to boost the coding capacity even further.


Introduction
[8] By combining two immiscible liquid phases in appropriate microfluidic geometries (like T-, Y-, or Flow-Focusing junctions), it is possible to realize isolated liquid environments.This technology, also known as dropletbased microfluidics, gives rise to numerous advantages to single-flow microfluidic solutions regarding the throughput (as every droplet can be regarded as a single experiment), automatization potential using droplet splitting, fusion, and storage at low-volumes, faster reaction speeds and minimizing contamination or impurity processes since droplets are encapsulated in the carrier liquid. [9,10]Due to these characteristics, droplet microfluidics is applied in medicine and biology in lab-on-a-chip (LOC) or micrototal analysis systems (TAS). [11]The current technology enables the precise formation and manipulation of droplets, namely their sorting, [12,13] trapping, [14] and fusion [15,16] at high rates.However, an easy and reliable tracking method is crucial for any running experiment since each droplet needs to be clearly identified and tracked during assays, which can last from minutes up to days and weeks.Multiple approaches were developed for barcoding and real-time tracking of droplets including droplet chain size coding, [17] DNA-based barcodes, [18,19] or fluorescent molecules. [20]To avoid influencing the droplet environment via, e.g., the introduction of fluorescent or DNA markers, one strategy depicts a label consisting of a droplet code represented by a chain of several droplets flowing next to a specific analyte droplet.Labeling the individual droplets in the chain can be achieved via, e.g., fluorescence.However, this strategy needs very precise and expensive spectrometers able to detect various fluorophores at various concentrations, especially in the context of the number of sample numbers for current biotechnological and chemical state-of-the-art processes, i.e., up to several million samples in high-throughput screening (HTS) assays in drug discovery. [21]n alternative route relies on the utilization of magnetic labels using several concentrations of ferrofluid in droplets.In this regard, the combination of magnetism and droplet-based microfluidics, also known as droplet-based micro-magnetofluidics, offers the capability to measure and characterize tiny volumes of magnetically active suspensions.[24] Due to their biocompatibility as well as the absence of the magnetic moment without external magnetic fields, suspensions of superparamagnetic nanoparticles like Fe 3 O 4 are typically applied in the field of biosensorics and lab-on-a-chip approaches. [25]Among different types of magnetic field sensors utilized in droplet-based microfluidic setups (e.g., giant magnetoresistance (GMR) and anisotropic magnetoresistive effect sensors (AMR) [26,27] ), planar Hall effect (PHE) sensors offer advantages due to their linear response and high sensitivity to small magnetic fields of ferrofluid droplets. [28]n this context, nanoliter droplets with different concentrations of ferrofluids were detected and characterized in a contactless manner using PHE sensors. [28]MR sensors were successfully used to detect nanoliter ferrofluid droplet chains holding binary [29] to quaternary [30,31] codes.Despite having an infinite number of codes in theory, this technology meets practical challenges for the preparation of large libraries of codes at the level of 1 million codes.Indeed, 20 binary coded droplets are needed for 1 million codes, which result in a rather long tubing to handle this number of samples leading to the high hydrodynamic resistance in fluidic channels.In detail, 20 droplets each 0.5 mm long require 10 mm per code.For 1 million codes, the length of the tube should be at least 10 km, which can hardly be handled by a fluidic pumping system.Whereas the use of quaternary codes allows to reduce the amount of droplets per code twice for the same coding capacity, this does not solve the issue with the hydrodynamic resistance.On the other hand, the necessity to detect more signal levels per droplet (four instead of two) puts stringent requirements on magnetoresistive sensors, which should have a sufficiently high signal-to-noise ratio (SNR) and assure sufficiently large safety margins for the discrimination of multilevel code within one droplet.
Hence, at present, there is no technology, which can practically offer millions of unique codes in fluidics.Although magnetic coding could be a viable option, the current approach, which is focused only on the optimization of the sensor performance, seems to be insufficient.Indeed, the typically undertaken approach is to optimize magnetic field sensors and integrate them in a rather simplistic fluidic circuit.In contrast, biologists, chemists, and medical doctors approach the problem from a different side.Namely, this community applies advanced fluidics, which provides a high degree of flexibility and control of droplets (e.g., volume, spacing, and chain sequence) and then integrates appropriate sensors to meet the requirements of the detection.In particular, in biology and chemistry, the complexification of the microfluidic system enabled impressive miniaturized applications in droplets like molecular analysis of DNA and proteins, diagnosis of infections and cancer, as well as investigations in cell-based assays with high flexibility and high-throughput analysis. [9,12]Ultimately, we follow this methodology of complexification of the fluidic circuitry and combine it with ultra-sensitive sensorics to allow the practical generation and characterization of multiparametric droplet codes.
Here, we propose a new concept of multiparametric coding in droplet-based fluidics where we rely not solely on the optimization of the sensing unit but start with the targeted optimization of the microfluidic circuit and its complexification.Our concept benefits from the interplay of magnetic and impedance sensors and advanced droplet-based microfluidics to enable multidigit (formation of n droplets of different volume in the droplet chain) and multiparametric (m different physical entities per droplet enabling not only magnetic but also impedimetric detection) (Figure 1).The code length n (i.e., n is number of droplets per code) is controlled by the fluidic system since it defines the number of droplets in the chain.The definition of detectable parameters per droplet, or multiparametricity, is shared between the microfluidic part since it defines the droplet's parameters like volume or concentration of ferrofluid per droplet, as well as the sensor system, which detects these entities.We address one of the challenges of micro-magnetofluidics in the creation of extended libraries of ferrofluid droplets.Namely, instead of typically used extraction of droplets to separate codes, [29] we realize a new approach based on cleaving off the subsequent droplet with a single droplet.In this regard, we demonstrate a microfluidic system, which allows the generation of neighboring ferrofluid droplets of different volume and containing ferrofluid at various concentrations.We investigate the magnetization magnitude of the droplets for identification of the optimal detection environment and analyze all possible combinations of the two-droplet code.Finally, the impact of added multiparametric encoded information of the code by variation of the droplet volume(s) is explored.This fluidic platform is readily extendable to design droplet chains or higher complexity and involves different sensing properties, which we exemplify by implementing impedimetric detection of ionic concentrations.This allows us to realize over 3 million unique codes.Our investigations offer important insights into the fabrication of multiparametric droplet codes for functional fluidic applications including large-scale drug screening.

Results and Discussion
The ultimate goal of guaranteeing millions of unique codes in a practical way becomes reachable by increasing both parameters n and m (Figure 1).For instance, utilizing four droplets per code with four distinct parameters and three different values of each parameter allows to obtain over 43 million (3 16 ) unique codes.Considering the practical utility of the proposed concept, it seems insightful to work with three droplets per code (i.e., n = 3).Indeed, even in the case of millifluidics with 0.5 mm channel diameter, the length of a three-droplet code is ≈1.5 mm, which allows it to accommodate 1 million codes in a tubing with a length of 1.5 km.Filling of a L = 1.5 km long circular tubing with inner radius, R, of 250 μm at a flow rate (Q = 1 mL h −1 or 2.78 × 10 −10 m 3 s −1 ) with water at 20 °C (viscosity  = 0.001 Pa s) can be achieved using conventional microfluidic pumps since the pressure drop is at the level of ΔP = Q 8  L  R 4 = 1.8 Bar.For .By a modular combination of the two adjustable parameters in the droplet chain, a multiparametric code is generated.B) While multidigit n is defined by the fluidic system, multiparametricity m (defined by the sensor system) is given by the number of physical properties, which can be distinguished per droplet, e.g., different ferrofluid concentrations, pH values, ionic concentrations or optical density.Every modularly attached droplet in the droplet chain will result in n+1 multidigitally, while every new sensing of a droplet parameter increases multiparametricity m.Eventually, machine learning algorithms are needed to interpret these data points.C) Potentially, the code can be extended to triple, quadruple, and sextuple coding, as shown in photographs in the bottom row (scale bar: 1 mm).Furthermore, via an increase of m, the number of codes can be increased to over 43 million (3 16 ) (see table ).
comparison, when using binary coding, the needed pressure increases by a factor of ≈10, since the tubing has to have a minimum length of 10 km to host all unique codes.This required pressure is out of reach for commercial microfluidic pumps.Hence, the proposed concept of multiparametric coding is about an order of magnitude improvement over binary codes.To demonstrate the principle, first, we will focus on the specific case with n = 2 (two droplets per code) and m = 2 (volume and magnetic detection of droplets).For this, we modularly extend single droplets with a second droplet, both independently controllable in their volume and ferrofluid concentrations.Further, spacer structures were introduced to achieve robust code formation in the tubing (Figure 2).Furthermore, we demonstrate that the fluidic platform is readily extendable to arbitrary n and m to meet the requirements of large code libraries in fluidics.

The Setup
Ferrofluid droplets are created using a Y-junction with 500 μm inner diameter (Figure 2A, left panel) and guided to a second Y-junction introducing a second stream of carrier oil (Figure 2A, middle panel).This structure increased the inter-droplet distance needed for a correct droplet code formation downstream, where the droplet stream was brought in contact with a single-phase independent second ferrofluid stream.Droplet pairs are fabricated by cleaving off the second ferrofluid stream by passing single droplets, creating a two-droplet code in close proximity (Figure 2A, right panel).Downstream of the coding geometry, the tube containing the droplet code was guided over the PHE sensor, and the transversal resistance change was recorded.
The proposed microfluidic setup can be extended in a modular manner by adding further Y-junctions to realize codes with three or more droplets (Figure 1; Figure S1, Supporting Information).Modular extension of the droplet codes comes at the cost of a more involved droplet flow sensing.Specifically, the stable code generation in the current fluidic circuit requires precise identification of the speed of droplets.We note that even for single droplet generating microfluidic structures, the droplet velocity does not correlate linearly with the flow rate but depends on several parameters like their volume, viscosities of the droplet phases and carrier phases, geometry of the channel cross-sections, interfacial tensions, temperature and concentrations of surfactants. [32,33]Thus, especially for various ferrofluid concentrations in droplets used in codes (and thus different viscosities), spacing geometries, and increased tubing lengths for measurement, the speed of created droplets is hard to predict.Consequently, a real-time measurement of the droplet speed For all experiments on the code generation and analysis, electrode pair #3 (and #4 for speed measurements) was used.C) Typical signal appearance after baseline subtraction of a code containing droplets with a 2 mg cm −3 ferrofluid, followed by a droplet with 1 mg cm −3 ferrofluid.D) Speed measurement of droplets at various flow rates.For this measurement, electrode pairs #3 and #4, indicated in panel (B), were probed in parallel.The droplet speed was calculated as a ratio of the distance between electrode pairs #3 and #4 and the time difference corresponding to the detection events (signal peaks) measured using these electrodes (inset).The blue dashed curve shows a guide to the eye.
is crucial to ensure the reliability of the code generation.To tackle this challenge, the utilized PHE sensor is designed to have multiple measurement electrode pairs.It depicts 6 measuring electrode pairs, allowing higher flexibility in measurement applications, i.e., measurement of droplet speeds (Figure 2B).We note that another obvious advantage of having multiple electrode pairs is the possibility to have redundancy in case one of the electrode pairs fails during operation.The PHE sensor's geometry is shaped as an ellipse with two lateral electrodes for current supply.For the measurement, the PHE sensor is fixed on a printed circuit board (PCB) plate and micro-electrodes are further bonded on copper electrode pads on the PCB and downstream to coaxial connectors.The connection scheme and dimensions of the sensor can be seen in Figure S2 (Supporting Information).Additionally, studies were carried out utilizing electrode pairs 3 and 4 at a distance of 1.5 mm (Figure 2B) for measuring the droplet speed (Figure 2D; Figure S3, Supporting Information).For droplets of larger length, e.g., longer than 1.5 mm, other pairs of electrodes with a larger separation distance can be used.
For reliable droplet code generation, precise knowledge of the droplet's speed is mandatory.The flow rate of each subsequent ferrofluid stream, which participates in the code generation (i.e., 2nd ferrofluid stream in this work or potential further streams for three or more droplet chains) has to be carefully selected based on the speed of the previous droplet or droplet chains.Furthermore, a non-linear dependence was observed for the droplet speed and combined flow rates of the microfluidic systems, which can be explained by the complex interactions affecting the droplet speed as well as the deformability of the tubing system (Figure 2D).The ferrofluid used in this work consists of an aqueous suspension of 10 nm Fe 3 O 4 nanoparticles.While the magnetic moment of each nanoparticle fluctuates and averages to zero over time without any surrounding magnetic field, they align to applied magnetic field lines thereby increasing their magnetic stray field.We utilize this behavior of a superparamagnetic solution to boost the sensor's ability to detect the droplets by increasing the magnitude of the stray field of each ferrofluid droplet.The raw data describe the resistance alteration of the transversal resistance of the PHE sensor upon the presence of magnetic fields induced by individual droplets.Here, the change of the resistance depends on the concentration of magnetic nanoparticles per droplet.In total, three different concentrations (1, 2, and 3 mg cm −3 ) of ferrofluid were chosen for the droplet code generation.Prior to droplet code formation, the PHE sensor was calibrated for the individual concentrations at various magnitudes of magnetization of the droplets using Helmholtz coils, which expose droplets to a magnetic field of 1-5 mT.For this, droplets with various concentrations of ferrofluid were guided over the PHE sensor, and the transversal resistance R xy was recorded.This resistance increases depending on the strength of the external magnetic field.The signal strength was found to be at 0.41 mΩ for 3 mg cm −3 , 0.31 mΩ for 2 mg cm −3 , and 0.17 mΩ for 1 mg cm −3 when droplets with volume of 100 nL were exposed to the external magnetizing field of 5 mT.Furthermore, the sensor signal is found to be linearly proportional to the magnitude of the external magnetic field (Figure S4, Supporting Information).In Figure 2C, typical data can be seen for a droplet code consisting of a 2 mg cm −3 droplet, followed by a 1 mg cm −3 droplet.For these experiments, the flow was set to 0.1 mL h −1 .Two clear plateau signal levels are observed at different signal levels, resembling the respective concentration of the ferrofluid inside the droplets.
To elucidate the effect of external droplet magnetization and to find the optimal measurement regime, a droplet code consisting of a droplet with 2 mg cm −3 concentration of ferrofluid followed by a droplet with 1 mg cm −3 concentration of ferrofluid was measured at various external magnetic fields.A setup of Helmholtz coils is used to magnetize the droplets in the out-of-plane direction with respect to the PHE sensor plane.This configuration of the applied magnetic field helps to minimize the influence of the external magnetic field on the magnetic state of the sensor.Signal magnitudes for passing droplet codes are greatly increased by increasing the magnetic field from 1 to 5 mT.Showing plateau signal level values of 0.33 mΩ for 2 mg cm −3 ferrofluid concentration at a magnetic field of 5 mT, it drops to 0.09 mΩ at 1 mT.(Figure 3A, ΔResistance 1 ).This behavior is in agreement with the reference measurements of individual droplets (Figure 2C; Figure S4, Supporting Information).Similar behavior is observed for the second droplet carrying the ferrofluid at the concentration of 1 mg cm −3 (ΔResistance 2 ).The dependency of the plateau signal level shows linear behavior in the range of measured magnetic field strength for both droplet heights, ΔResistance 1 and ΔResistance 2 (Figure 3B).Furthermore, analysis of the background noise of the whole system was carried out to understand improved signal strengths toward moving ferrofluid droplets.For this, the background signal was measured in the presence of various magnetic fields, followed by the calculation of the signal-to-noise ratio (SNR) for each field using SNR = 20 log(RMS/STD) with RMS as root mean square and STD as standard deviation.Results indicate that SNR increases for higher magnetic fields (Figure 3B, green data points).
In addition to the overall signal strength, an important indicator for the sensor's optimal measurement regime is the difference between the two plateau signal levels of the individual droplets.The difference between ΔResistance 1 and ΔResistance 2 , labeled as ΔResistance sens was calculated for each magnetic field strength (Figure 3C).The difference of the plateau signal levels increases with an increase of the magnetic field strength.As before, SNR values of the plateau signals were calculated and subtracted to demonstrate improved sensitivity in elevated magnetic fields.Thus, it is advantageous to increase the bias field and all following measurements were carried out in the applied magnetic field of 5 mT.The identification of the peak or plateau signal level was achieved via a peak finding algorithm using the 2nd derivative of the data (Figure S5, Supporting Information) and the mean is calculated using 40 individual droplet pairs.

The Droplet Codes
In the spirit of exploring the suitability of the conjunction of the droplet-based micro-magnetofluidics and PHE sensorics for (de-)coding applications, all possible combinations of droplet codes consisting of two droplets, using three ferrofluid concentrations per droplet, were created.This includes the combinations of different concentrations as well as droplet codes with the same concentration of ferrofluid, i.e., 1-1, 2-2, and 3-3.Here, the number corresponds to the concentration of ferrofluid in a droplet, i.e., "1-1" stands for a droplet with ferrofluid concentration of 1 mg cm −3 by another droplet with the concentration of 1 mg cm −3 .Furthermore, the first number represents the concentration of ferrofluid used for the creation of the initial droplet sequence.The second number corresponds to the concentration of the 2nd ferrofluid flow.During code fabrication, the second ferrofluid stream is cleaved off, so that it flows in front of the first ferrofluid.Thus, the droplet order is mirrored from its description, i.e., a 1-3 code depicts a 3 mg cm −3 droplet followed by a 1 mg cm −3 droplet.In total, nine different combinations of droplet codes were fabricated (Figure 4A) and guided over the PHE sensor directly after their formation.Without droplets in the vicinity of the PHE sensor, no signal variation is observed, as the distance between droplet codes is located between 2 and 3 mm, which results in a stable baseline at ≈3.25-4.15mΩ depending on the magnitude of the external magnetic field (Figure S4, Supporting Information, panel A).As soon as a droplet code is flowing over the sensing area, an increase in the transversal resistance is observed.Here, the resistance change depends on the concentration of the ferrofluid present in the droplet and it allows clear identification and decoding of the droplet order.Representative data are shown in Figure 4B for the codes 1-3, 2-3, and 3-3 measured in an external magnetic field of 5 mT.For the droplet combination 1-3, a clear peak plateau signal level is observed for both concentrations, i.e., 33 mΩ for 3 mg cm −3 and 0.22 mΩ for 1 mg cm −3 (top panel, arrows).Similar behavior is observed in the data for the droplet codes 2-3.In the case of droplet codes 3-3, two plateau signal levels are observed at the same signal level.This valley is thought to be caused by a lower magnitude of stray fields between the droplets, which drops with 1/d 3 (d: distance between droplets).By repeating this procedure, a map for all nine various concentrations was created, which can be used for decoding the respective droplet code.Potentially, this data set can be further utilized in the training of machine learning algorithms for the identification of the droplet codes.Here, all nine combinations of droplet codes were guided over the PHE sensor, and the two-droplet code iden-tifying parameters, ΔResistance 1 and ΔResistance 2 , were calculated.The results can be seen in Figure 4C, where ΔResistance 1 and ΔResistance 2 are plotted.

Two-Parametric Droplet Codes
Unlike electrochemical sensing, when the chemical composition of liquids is measured directly, the detection of magnetic stray fields using magnetoresistive sensors [34,35] is not spatially restricted to the analyzed liquid volume.It spreads in 3D space and it is to be expected that specific signals of individual droplets in a droplet chain will be influenced by the volume of droplet(s), distance between droplets, and concentration of ferrofluid in a droplet.Benefiting from this effect, multiparametric coding of the droplet chain becomes possible.Indeed, the droplet volume adds extra information to the code without alteration of the order of the ferrofluid concentrations.This expands the decoding information to a new dimension and helps to extend the information encoded in the droplet chain.While plateau signal level values for droplet pairs show clear data clouds in the scatter plot (Figure 4D), these values can be affected by the droplet volume, since PHE sensors rely on the detection of magnetic stray fields emerging from the individual droplets.This behavior can be seen in Figure 5A, where the droplet code 2-1 for two droplet volume configurations is depicted.Here, a lower left shoulder signal for the droplet with a concentration of 1 mg cm −3 can be seen if the two droplets possess the same volumes (2-1, Figure 5A, orange data, and inset).While doubling the flow rate of the liquid stream with the ferrofluid concentration of 1 mg cm −3 , and thus the droplet volume of the first droplet, a higher shoulder plateau signal level for the bigger droplet is observed (2-1(l), Figure 5A, black data, and inset).Meanwhile, no elongated plateau signal level was observed for the bigger droplet volume compared to the smaller volume due to the increased flow rate of the bigger droplet volume, equalizing the time of the droplets spent over the sensor.Quantitative comparison of these two configurations using ΔResistance 1 and ΔResistance 2 of 2-1 and 2-1(l) reveals distinguishable data clouds in the scatter plot without alteration of the droplet sequence (Figure 5B, black (doubled droplet volume of the droplet with the ferrofluid concentration of 1 mg cm −3 ).A higher plateau signal level of the 1 mg cm −3 droplets in the 2-1(l) configuration was found.B) Scatter plot of ΔResistance 1 and ΔResistance 2 for four different configurations of the droplet code 2-1.Distinct data clouds for these two configurations could be found without changing the ferrofluid content per droplet and the droplet order.Further manipulation of the scatter clouds was achieved via halving the droplet volume for 1 and 2 mg cm −3 droplets.C) Impact of the degree of droplet's magnetization (different strengths of the magnetizing fields) on the signal profile for various droplet volumes.For this, the time-dependent 1-1 code at equal volumes (green data) is compared with a 3-1 code with a bigger front-flowing droplet at 1 mg cm −3 (black data).The difference in the peak signal level of this droplet is pronounced in higher magnetic fields (ΔS 1 ).D) Dependence of ΔS 1 and ΔS 2 on the external magnetic field.While ΔS 2 scales linearly with increasing magnetic fields, ΔS 1 saturates at 3 mT.Each data point represents the mean and standard deviation of 30 droplets.The dashed curves show a guide to the eye.and orange data).Furthermore, two additional droplet code configurations for 2-1 chains were fabricated, i.e., half volume of the droplet with ferrofluid concentration of 1 and 2 mg cm −3 (Figure 5B, green and brown data, respectively).A decrease in the droplet volume leads to a smaller signal of the sensor, both for the two ferrofluid concentrations.This enables an additional dimension for multiparametric coding, where information encoded in the volume of droplets is used at the same footing as the ferrofluid concentration in a droplet.
In this manner, via the incorporation of new parameters (here, volume) alterations in the codes, the flexibility and capability of the coding performance can be increased.For this increase, only the flow rates of the individual ferrofluids have to be altered while the fluidic circuit and the sensor remain the same.Using only two volume alterations (same volume and double volume) increases the nine combinations from the same volume droplets (3 2 ) to 36 possible combinations (3 2 x 2 2 ).Furthermore, by modular extension of the fluidic circuit to incorporate a third droplet with three different volumes for all droplets in the code, the coding capacity can be increased to 729 (3 (3 + 3) ), representing the same number of combinations if nine individual concentrations would be used.
To elucidate the influence of the external magnetic field on the decoding sensitivity of the sensor on the volume of droplets, droplet codes of the combinations 1-1 and 3-1 were measured in magnetic fields of 1-5 mT.Furthermore, the droplet volume of the 1 mg cm −3 droplets in code 3-1 was doubled, compared to the volumes of 1-1, and the signal differences ΔS 1 (1 mg cm −3 vs 1 mg cm −3 ) and ΔS 2 (3 mg cm −3 vs 1 mg cm −3 ) were calculated for the measurements taken at different magnetic fields (Figure 5C).Here, ΔS 2 linearly increases with the magnetic field strength from 52 Ω at 1 mT to 182 Ω at 5 mT, resulting in a pronounced decoding sensitivity at stronger fields (Figure 5D, top).Furthermore, although ΔS 1 increases with the magnitude of magnetic fields from 5 Ω at 1 mT to 23 Ω at 5 mT, it saturates at a field of ≈3 mT.This finding is in agreement with previous studies, where PHE sensors were demonstrated to be insensitive to the volume of droplets at low external magnetic fields. [28]Thus, to utilize multiparametric analysis of passing ferrofluid droplets relying on PHE sensors, an appropriate degree of ferrofluid magnetization in an external magnetic field has to be present.For our setup, this field is ≈3 mT.

Crossing the Million Unique Codes
In previous sections, we described the methodology of creating droplet codes and performing 1-and 2-parametric detection using magnetic field sensors.Yet, using only magnetic field sensing with a limited number of distinct values of magnetic parameters (i.e., the concentration of magnetic nanoparticles) makes it challenging to achieve large code libraries in the range of 1 million codes to meet requirements for sample sizes used in state-of-theart (bio-)medical assays and protocols.To address this challenge, in this section, we demonstrate that the developed fluidic platform can be extended to accommodate more droplets per code, more values per sensed parameter, and include other parameters for sensing (i.e., ionic concentration).
We incorporate a third droplet in the droplet chain by adding a third Y-junction downstream of the second droplet-generation structure.A typical data pattern can be seen in Figure 6A for a droplet code 3-1-2, while peak analysis results in a data representation in three-dimensional (3D) space including peak values of all three droplets (Figure 6F, top panel).To further increase the number of unique codes, we increase the number of distinct concentrations of magnetic nanoparticles per droplet from 3 to 9.This is possible owing to the high sensitivity and stability of the PHE sensor, which allows for distinguishing also smaller concentration differences (Figure 6B).Furthermore, we extend the utilized droplet volumes to work with droplets of four distinct volumes.Figure 6C demonstrates the signal for the droplet code 1-1-1 with alterations of droplet volume in the first droplet.While the black data set depicts no alterations in the volumes (with default droplet volume of 100 nL, black data set), a decrease of the volume in the first droplet to 75 nL results in a lower shoulder signal (green data set).In contrast, higher volumes (125 nL (yellow) and 150 nL (orange)) lead to a higher shoulder signal (see further data sets in Figure S6, Supporting Information).The data analysis moves to the 4D space, introducing the droplet volume in the fourth dimension (Figure 6F, bottom panel).
The fluidic setup is further extended to perform impedancebased ionic sensing downstream of the PHE sensor (Figure 6D).In the first droplet of the 1-1-1 chain, we introduce NaCl solution with four different salt concentrations ranging from 10 to 500 mm.Based on the salt concentration alteration in the first droplet, a non-magnetic parameter could be added to the analysis, contributing to the increase in the number of unique codes.The ionic concentration is measured using a millifluidic impedance analyzer, which utilizes an inductive coil to assess the properties of fluids in the droplet chain. [36]onsidering the number of detected parameters and distinct values per parameter, the amount of unique codes that can be produced and detected using the reported setup is almost 3 million (2 985 984 = 9 3 4 3 4 3 ).

Conclusion
In this work, we propose a new concept of coding and decoding using droplet-based fluidics, allowing to generate over 1 million unique codes in a practically realizable fluidic circuit.With the use of multiparametric detection of droplet codes, consisting of two adjacent nanoliter ferrofluid droplets, the practical use of this kind of coding vastly increases, compared to single parameter detection per droplet.In this, the tubing length needed to host 1 million unique codes can be reduced by one order of magnitude compared to binary codes, making it possible to realize this fluidic circuit.To further increase its potential, the concept can be extended to a higher number of droplets per code and multiparametric (number of detectable droplet properties) analysis of droplets (e.g., optical and electrochemical detection in addition to magnetic field sensing).For validation of this concept, we created codes with various concentrations of ferrofluid ranging from 1 to 3 mg cm −3 in droplet volumes ranging between 70 and 150 nL.Using a droplet code of two droplets of the same volume with three concentrations, all nine combinations were fabricated and guided over the PHE sensor while recording the transversal resistance R xy .Based on the numerical quantification of this effect, a map of various combinations was achieved for decoding the individual codes.In addition, real-time control of droplet speed in this complex fluidic setup was achieved by multiplexed data collection utilizing multiple contact pairs on the PHE sensor.Moreover, additional multiparametric analysis and additional dimension of stored droplet code information were achieved via variation of the droplet volume.Here, different configurations of the droplet volumes allowed further information content per droplet code without alteration of the droplet order and concentration of ferrofluid per droplet.This variation of droplet volumes of individual droplets within the code increased the capacity of the code library, i.e., increasing the number of combinations from 9 to 36 codes in a two-droplet code.
By exploring this modular approach toward functional dropletbased fluidics, we extend the number of droplet codes up to nearly 3 million by monitoring three-droplet chains with nine magnetic concentrations, four different volumes, and four ionic concentrations.Further optimizations and down-scaling can be achieved by the transfer of this setup to miniaturized PDMSbased microfluidic channel systems for higher control of droplet flow and higher sensitivity to the droplet content by minimizing distances of sensors to fluids and reducing overall droplet volumes.These complex codes with multiple volumes, concentrations, and total number of droplets will require computational help to interpret the multidigit and multiparametric data patterns for each code.For this, computational techniques like machine learning (ML) algorithms (e.g., supervised discriminant analysis [29] or unsupervised k-means clustering [6] ) demonstrated high potential in these kinds of data patterns in LOC and TAS systems.

Experimental Section
Fabrication and Characterization of PHE Sensors: Al 2 O 3 (60 nm)/Ta(5 nm)/permalloy(200 nm)/Ta(5 nm) films were sputtered on (100)-oriented Si wafers in an ion beam-sputtering system (Nanoquest I, Intlvac). [37]Elliptical shapes having dimensions of 5 mm in length and 0.625 mm in width were patterned by photo-lithography.Using the lift-off process, gold electrical contact pads of 300 nm thickness were fabricated.Detailed magnetic and electronic noise characterizations of all six voltage pairs were performed on the fabricated PHE sensors.This single-domain PHE sensor had an average effective anisotropy field of ≈10 Oe.The equivalent magnetic noise of all the voltage pairs was better than 300 pT √Hz −1 at 50 Hz for ≈150 mA excitation in a three-layer shielded condition.Detailed noise and magnetic measurement procedures can be found elsewhere. [34,38]roplet Fabrication and Delivery: Millifluidic droplets were fabricated utilizing commercial Y-junctions (Techlab GmbH, Germany) with 500 μm channel diameter.An increase in the inter-droplet distance and fabrication of the droplet code was carried out in transparent Y-junctions with a channel diameter of 800 μm (HZDR Innovation GmbH, Germany). [39]iquids were transported within FEP-based tubings with an inner diameter of 500 μm and an outer diameter of 1.6 mm (Dolomite Microfluidics, UK).The aqueous ferrofluid droplet phase was utilized in three different concentrations: 1, 2, and 3 mg cm −3 .For this, ferrofluid stock solution (EMG700, Ferrotec, USA) was initially diluted with DI water in a ratio of 1:4 and further diluted to the desired concentrations.Prior to every liquid change, the respective syringes were carefully cleaned using isopropanol and deionized water and all fluidic components were flushed with HFE oil.It is noted that in this work ferrofluid was used, which is based on 10 nm-sized Fe 3 O 4 nanoparticles.These samples reveal superparamagnetic behavior at room temperature with a bifurcation point at the zero field cooling-field cooling plot at ≈270 K.The structural and magnetic properties of this commercially available ferrofluid were well characterized.For a comprehensive magnetic characterization, a recent paper by Sadat et al. was referred to. [40]tabilization of the ferrofluid droplets was ensured by probing the permanent carrier HFE oil phase (Novec 7500, 3M Corp., USA) with 1.5% surfactant w/w content (Pico-Surf, Sphere Fluidics Limited, UK).Liquid transportation was executed actively utilizing 10 mL syringes (VWR International, Germany) controlled by a syringe-based pump (nemeSYS 290N, cetoni GmbH, Germany).For droplet generation, flow rates of HFE oil and ferrofluid were set to 0.5 mL h −1 .Initially, the ferrofluid droplet chain and HFE oil with 1.5% surfactant (0.4 mL h −1 ) were guided to the second Y-junction to increase the distance between droplets.Following, the tube containing the droplets and the second ferrofluid phase (0.25-0.4 mL h −1 , depending on the ferrofluid concentration) were connected to the 3rd Yconnector to generate the droplet code.
The tubing, containing the droplet code, was aligned over the sensor ellipse and fixed using a self-made holder.The bottom of the tube, touching the sensor, was additionally sanded to reduce the distance between the droplet codes and the sensor.The measurement of the droplet code was carried out either at the given sum of the flow rates or by slowly pushing premade droplet codes at a speed of 0.1 mL h −1 using HFE oil from the syringe for increasing the inter-droplet distance.
Data Capture and Analysis: The PHE sensor was integrated on a PCB and the electrodes of the PHE sensor were bonded to the electrode pads of the PCB.The pads were joined to coaxial adapters and connected to a resistance tensormeter [41,42] (HZDR Innovation GmbH, Germany), allowing a direct 4-probe measurement with direct data readout.Utilizing a drive voltage amplitude of 15 V, the transverse resistance of the PHE sensor was collected.The current amplitude was set at 60 mA through the sensor.Raw signal of the PHE sensor resistance signal was captured from the resistance tensormeter at 30 ms time resolution.Data analysis and representation were carried out in OriginLab's Origin 2019.Typical data pattern representation (e.g., Figures 1C, 3A, and 4B) included baseline subtraction using the reference curve of the respective smoothed data over 3000 data points.Droplet peak heights were calculated using Origins 2019 automatic peakfinder function.
Impedance-Based Detection of Droplet Codes: For impedance sensing, 1 mL of solution with different concentrations of ferrofluid with solid sodium chloride (S7653, Merck KGaA, Germany) was mixed to achieve concentrations of 10, 50, 100, and 500 mm.A chain of codes containing three droplets was generated and the tubing was guided through a coil of the impedance analyzer, which was located downstream of the PHE sensor.Doing so, next to magnetic information stored in the concentration of magnetic content of individual droplets, chemical information of the droplets was detected additionally.The operation principle of the fluidic impedance analyzer is described elsewhere. [36]tatistical Analysis: Statistical analysis of droplet speed (Figure 1D) and droplet height signal (Figure 1B,C) was calculated using at least 20 mean values of droplet peak values including their standard deviation (mean ± STD).Scatter plot representation in 2D (e.g., Figures 4C and 5B) and 3D (Figure 6F) were limited to 40 data points for each presented droplet code to maintain the readability of the plots.

Figure 1 .
Figure 1.Presentation of the concept of coding application in droplet-based microfluidics.A) Conceptual illustration of the strategy: Individual code droplets contain two parameters, i.e., various ferrofluid concentrations (parameter c: 1-3 mg cm −3) and volumes (parameter V: 75-150 nL).By a modular combination of the two adjustable parameters in the droplet chain, a multiparametric code is generated.B) While multidigit n is defined by the fluidic system, multiparametricity m (defined by the sensor system) is given by the number of physical properties, which can be distinguished per droplet, e.g., different ferrofluid concentrations, pH values, ionic concentrations or optical density.Every modularly attached droplet in the droplet chain will result in n+1 multidigitally, while every new sensing of a droplet parameter increases multiparametricity m.Eventually, machine learning algorithms are needed to interpret these data points.C) Potentially, the code can be extended to triple, quadruple, and sextuple coding, as shown in photographs in the bottom row (scale bar: 1 mm).Furthermore, via an increase of m, the number of codes can be increased to over 43 million(3 16  ) (see table).

Figure 2 .
Figure 2. A) Microfluidic Y-junctions for droplet code generation.Left: The ferrofluid phase (top) and carrier oil phase (left) are pumped to the Y-junction, resulting in ferrofluid droplets in the exit tube (right).Middle: After increasing the inter-droplet distance, droplet codes were fabricated using a Y-junction (Right).Depending on the ferrofluid concentrations (three levels per droplet), up to nine codes could be generated.The scale bar is 1 mm.B) Scanning electron microscopy (SEM) image of the PHE sensor.Different components of the sensor are highlighted with false colors.The sensor layout has six electrode pairs for collecting the transversal resistance R xy .For all experiments on the code generation and analysis, electrode pair #3 (and #4 for speed measurements) was used.C) Typical signal appearance after baseline subtraction of a code containing droplets with a 2 mg cm −3 ferrofluid, followed by a droplet with 1 mg cm −3 ferrofluid.D) Speed measurement of droplets at various flow rates.For this measurement, electrode pairs #3 and #4, indicated in panel (B), were probed in parallel.The droplet speed was calculated as a ratio of the distance between electrode pairs #3 and #4 and the time difference corresponding to the detection events (signal peaks) measured using these electrodes (inset).The blue dashed curve shows a guide to the eye.

Figure 3 .
Figure 3. A) Close-up of the typical data appearance for the droplet code consisting of a droplet with a ferrofluid concentration of 2 mg cm −3 , followed by a droplet with a concentration of 1 mg cm −3 .Transversal resistance measurement of the PHE sensor reveals different plateau signal levels of the droplets in the code, namely droplets with 2 and 1 mg cm −3 .The magnitude of the signals measured by the device scales with the magnetization of ferrofluid droplets provided by the external magnetic field.B) Plateau signal level ΔResistance 1 (first peak, x-axis) versus ΔResistance 2 (second peak, y-axis).Linear relationship between these two values is found for increasing magnetic fields.Each data point represents the mean value of 40 individual droplet codes.C) Influence of the external magnetic field on the difference of plateau signal levels of ΔResistance 1 and ΔResistance 2 .The difference ΔResistance sens scales with the degree of magnetization of the droplets.The dashed and solid curves in panels (B and C) show a guide to the eye.

Figure 4 .
Figure 4. A) Collection of the analyzed droplet codes.In total, all nine various combinations were analyzed consisting of three different ferrofluid concentrations.B) Exemplary appearance of the droplet code signal.Based on the alteration of the transversal resistance depending on the ferrofluid concentrations, droplet chains are decoded based on the analysis of various plateau signal levels for different magnetic content in the respective droplet.C) Scatter plot of nine different droplet codes consisting of droplets with the same volumes.Depending on the ferrofluid concentrations in the droplet code, distinguishable scatter clouds could be observed.

Figure 5 .
Figure 5. Influence of the droplet volume.A) Comparison of the droplet code 2-1 in two configurations: 2-1 (same volume for both droplets) and 2-1(l)(doubled droplet volume of the droplet with the ferrofluid concentration of 1 mg cm −3 ).A higher plateau signal level of the 1 mg cm −3 droplets in the 2-1(l) configuration was found.B) Scatter plot of ΔResistance 1 and ΔResistance 2 for four different configurations of the droplet code 2-1.Distinct data clouds for these two configurations could be found without changing the ferrofluid content per droplet and the droplet order.Further manipulation of the scatter clouds was achieved via halving the droplet volume for 1 and 2 mg cm −3 droplets.C) Impact of the degree of droplet's magnetization (different strengths of the magnetizing fields) on the signal profile for various droplet volumes.For this, the time-dependent 1-1 code at equal volumes (green data) is compared with a 3-1 code with a bigger front-flowing droplet at 1 mg cm −3 (black data).The difference in the peak signal level of this droplet is pronounced in higher magnetic fields (ΔS 1 ).D) Dependence of ΔS 1 and ΔS 2 on the external magnetic field.While ΔS 2 scales linearly with increasing magnetic fields, ΔS 1 saturates at 3 mT.Each data point represents the mean and standard deviation of 30 droplets.The dashed curves show a guide to the eye.

Figure 6 .
Figure 6.Crossing the million unique codes.A) Time-dependent typical data pattern for the code 3-1-2.B) The mean value of the peak signal of 20 single ferrofluid droplets with magnetic concentrations from 0.05 to 4 mg cm −3 .C) Depiction of the time-dependent signal of the droplet code 1-1-1 with various volumes of the first droplet.Depending on the volume change of the first droplet from the default volume (100 nL, black data), shoulder formation can be observed for various tested volumes (green: 75 nL, yellow: 125 nL, and orange: 150 nL).D) Modification of the ionic content (10-500 mm) in the first droplet in the droplet code 1-1-1 with default volume.Depending on the concentration, impedance-based analysis downstream of the PHE sensor shows various significant signal appearances.E) Nomenclature of the droplet code library, where 1st droplet depicts a droplet code of 1-1-1 with no volume or ionic content alterations, while the last droplet code represents a droplet code of 3-3-3 with the same parameters.F) Depiction of the 4D space of a parameter set of 2. Top: 3D space of the droplet magnetic content; Bottom: Addition of the new dimension by the introduction of the droplet volume (1-1-1 vs 1-1-1l (l: large droplet of 150 nL)).