Hierarchically Structured Flexible Electrode on Polyimide for Highly Sensitive and Reliable Biosignal Acquisition

Numerous wearable biomedical devices are developed for the continuous monitoring of personal health or condition. Biosignals acquisition with high sensitivity is important for designing wearable biomedical devices. A sensing electrode between the human body and wearable electronics significantly affects the sensitivity of the sensors. In this study, we fabricated hierarchically structured flexible electrodes on polyimide substrate (HSFE-PI) using micro-casting technique and gold nanoparticles electrodeposition. Polyimides provides robust and outstanding electrical characteristics, and the reliability of HSFE-PI was verified with a cyclic bending test. The integration of hierarchical structures significantly increased the surface area of the electrode by 2.06 times. We applied the HSFE-PI for electromyogram (EMG) and glucose sensing applications and achieved high sensitivity enhancement in both applications. The signal-to-noise ratio (SNR) of measured EMG signals was increased by 2.48 times, and the sensitivity of the glucose detection was increased by 1.42 times compared to the planar counterpart.


I. INTRODUCTION
Wearable biomedical devices have received much attention to monitor individuals' health condition continuously by measuring various biosignals, such as electrocardiogram (ECG), electromyogram (EMG), ion concentration, and blood glucose level, in real time [1]- [6]. In order to design highperformance wearable devices, biosignal acquisitions with high signal-to-noise ratio (SNR) and sufficient sensitivity must be satisfied, as high-quality signals reduce the computational load for signal conditioning and increase the accuracy [7], [8]. A sensing electrode is one of the essential components that affect the sensitivity of biosignal measurements. For example, the electrodes transduce ionic potential inside the human body into an electrical signal or translate the biochemical detection into an electrical signal [9]. Therefore, The associate editor coordinating the review of this manuscript and approving it for publication was Lorenzo Mucchi . an optimal design of the sensing electrode is a key to achieve high-quality measurements.
Numerous previous studies on biosignal sensors report that the enlargement of contact area between the sensing electrodes and measured objects leads to a sensitivity enhancement [10]- [12]. For example, the electrodeposition of nanoparticles on biopotential electrodes can increase the surface area, which improves the SNR in ECG and EMG measurements [7]. Similarly, hierarchical structures with micro-/nano-patterns improves the response time and sensitivity of electrochemical sensors through a large surface area [13]- [17]. Thus, a hierarchically structured flexible electrode is an optimal approach to significantly enhance the performance as the hierarchical structures enlarge the surface area, and the flexibility provides a conformal contact with the curvature of the skin.
Polydimethylsiloxane (PDMS) is widely used as a substrate material for the hierarchically structured flexible electrode because of its low cost and facile fabrication [18]. However, a direct deposition of conducting thin film on PDMS is difficult since a large thermal expansion of PDMS induces irreversible cracks on the surface, thereby significantly reducing the reliability and conductivity [19]- [21]. Transfer of conductive materials, such as silver nanowires, gold nanomeshes, carbon nanotubes, and poly (3,4-ethylenedioxythiophene):poly (styrenesulfonate) (PEDOT:PSS) was suggested to avoid the problem [22]. However, the weak adhesion between conductive layer and polymeric substrate makes the fabrication difficult [23]. There were several approaches to form PDMS electrodes by adding carbon-based nanomaterials [24], [25]; however, the nanomaterials suffer from relatively low conductivity [26], which can significantly degrade the signal quality.
In this study, we demonstrated hierarchically structured flexible electrodes on polyimide substrate (HSFE-PI) and applied the electrode to the biosignals measurements ( Figure 1). Polyimide (PI), a material conventionally used in microelectronics and flexible displays, was utilized as electrode substrate to provide reliable electrode performance. Despite its excellent characteristics, the use of PI for hierarchically structured flexible electrodes was hindered because most PI patterning approaches, such as dry etching or hot embossing, require high cost and complex process compared to soft lithography [27]. We propose a novel approach to fabricate hierarchical structures using a method combining PI casting process using silicon mold and gold nanoparticles electrodeposition, which reduce the cost and complexity of the fabrication process. The hierarchical structure of HSFE-PI resulted in a considerable increase in the surface area of the electrode. EMG signals were measured and analyzed using HSFE-PI and planar electrodes, and glucose sensors were fabricated by integrating glucose oxidase (GOx) on the electrode. Enhancement in SNR and sensitivity from the hierarchical structure was observed for both applications. The high sensitivity and the stable electrical characteristic of HSFE-PI facilitate the design of reliable and accurate wearable biomedical devices.

B. PREPARATION OF SILICON MOLD FOR CASTING PROCESS
A positive photoresist (PR) film (2 µm) was spin-coated on a silicon wafer. The PR was patterned into a hole array (100 × 100) using conventional photolithography. Every hole had an identical diameter of 30 µm, and a center-to-center spacing between each hole was 50 µm. The patterned PR was used as a soft mask and the silicon wafer was etched by deep reactive ion etching (DRIE) with a depth of 40 µm. The PR film was removed by sonication in acetone for 5 mins, and the remaining PR residues were further etched by O 2 plasma PR asher. The silicon mold was rinsed with acetone and isopropanol for 5 mins each before use.

C. FABRICATION OF HIERARCHICALLY STRUCTURED FLEXIBLE ELECTRODE ON POLYIMIDE SUBSTRATE (HSFE-PI)
The fabrication process of HSFE-PI is depicted in Figure 2a. The surface of the silicon mold was treated with ultraviolet ozone (UVO) for 20 mins to obtain a hydrophilic surface. A hydrophilic surface decreases the adhesion between PI VOLUME 10, 2022 FIGURE 2. Fabrication process of HSFE-PI and glucose sensor. (a) UVO surface treatment facilitates the detachment of PI film from the silicon mold. PI varnish is poured on silicon mold and spin coated. Afterward, PI varnish is thermally imidized to form a micro-structured PI film. The film is detached from the mold, and the gold thin film is deposited on the film by e-beam evaporation. Gold nanoparticles (AuNPs) are electrodeposited on the micro-structured electrode to form hierarchical structures. (b) A glucose sensor is fabricated by immobilizing glucose oxidase (GOx) on HSFE-PI. The immobilization process is performed by crosslinking glutaraldehyde (GLA) and bovine serum albumin (BSA). and silicon, thus facilitating a detachment of PI from the mold [28]. After the surface treatment, PI varnish was poured on the silicon mold and placed in a vacuum for 210 secs. The vacuum process is required to fully fill the holes with PI varnish. The sample was then spin-coated at 2000 rpm for 60 secs and soft-baked at 200 • C for 5 mins. This coating process was repeated twice to achieve a PI thickness at 10 µm. The sample was hard-baked at 250 • C for 10 mins and 300 • C for 15 mins, for full imidization. Then, the micro-structured PI film was detached from the mold. Titanium (Ti, 5 nm) and gold (Au, 50 nm) were sequentially deposited on the substrate by an e-beam evaporator. A shadow mask was used to define a circular electrode area with a diameter of 10 mm. AuNPs were electrodeposited on the micro-structured gold electrode to complete a hierarchical structures fabrication. Electrodeposition was performed using an electrochemical analyzer (BioLogic SP-200) and a standard three-electrode setup. The three electrodes were composed of an Ag/AgCl reference electrode (1 M KCl), a platinum plate counter electrode, and the micro-structured gold electrode (working electrode). Before an electrodeposition of AuNPs, the surface of the micro-structured Au electrode was electrochemically cleaned by 12 potential cycling between −0.4 to 1.4 V (vs. Ag/AgCl) at a scan rate of 100 mV/s in 50 mM H 2 SO 4 until a cyclic voltammetry (CV) output became stable. AuNPs were then electrodeposited on the micro-structured electrode by applying −0.4 V (vs. Ag/AgCl) for 300 secs in a solution containing 5 mM HAuCl 4 and 0.5 M KNO 3 . Figure 2b illustrates the fabrication process of glucose sensor. GOx was immobilized onto the electrode surface by GLA and BSA crosslinking. Enzyme loading solution (150 µL), which contained 33 µL GLA solution (25%), 7.2 mg BSA, and 30 Units of GOx, was dropped onto the surface of the electrode. The solution was dried at room temperature for 30 mins so that crosslinking of GLA and BSA is progressed. GOx in the loading solution is fixed inside the crosslinked GLA/BSA matrix, thereby forming a glucose-sensing layer. The glucose sensors were stored at 4 • C when not in use to prevent degradation.

E. SEM CHARACTERIZATION OF HSFE-PI
SEM images of HSFE-PI were obtained using a highresolution FE-SEM (JEOL, JSM-7401F Scanning Electron Microscope) with an accelerating voltage of 5 kV and working distance of 8 mm.

F. ELECTROCHEMICAL CHARACTERIZATION OF ELECTRODES
CV measurements were performed using a BioLogic SP-200 electrochemical analyzer with a standard three-electrode system described above. For a relative surface area analysis, four types of electrodes were compared: 1) planar gold electrode, 2) AuNPs-deposited gold electrode, 3) micro-structured gold electrode, and 4) HSFE-PI. The planar gold electrode was fabricated by depositing Ti/Au (5/50 nm) on a planar PI substrate (10 µm). With additional AuNPs electrodeposition on the planar gold electrode, AuNPs-deposited gold electrode was obtained. The micro-structured gold electrode was fabricated by deposition of Ti/Au (5/50 nm) thin films on the micro-structured PI substrate. The footprint area of the electrodes was identical (78.5 mm 2 ). The definition of footprint area and surface area is shown in Figure 3. Utilizing these electrodes as working electrode, 5 CV cycles were performed for each electrode in a 1 M H 2 SO 4 solution at a scan rate of 50 mV/s (vs. Ag/AgCl).

G. EMG SIGNAL ACQUISITION
EMG signal was measured by using a three-lead configuration [7]. A conventional Ag/AgCl electrode was placed on the right wrist of a subject as a reference electrode. The other two electrodes were fixed on the left bicep through skin adhesive with a distance of 5 cm. Two types of electrodes were compared: planar gold electrode and HSFE-PI. An electrolyte gel was applied between the electrodes and the skin. Each electrode was connected to a semiconductor analyzer (Keysight, B1500A) through a BNC cable, and the potential was continuously measured with a sampling rate of 250 Hz while the subject lifted a 2 kg dumbbell continuously.

H. GLUCOSE CONCENTRATION MEASUREMENT
Solutions with various glucose concentrations were prepared by adding glucose to 0.1 M PBS (pH 7.0). CV was performed using a BioLogic SP-200 electrochemical analyzer with the three-electrode system utilizing the glucose sensor as a working electrode (Figure 4). Cyclic voltammograms of the sensors were obtained in different glucose concentrations at a scan rate of 100 mV/s and a voltage range between −0.2 and 0.8 V (vs. Ag/AgCl).
Amperometric measurements of glucose in PBS with the sensor were conducted using the experimental setup described above. The current data was continuously measured by applying a constant potential of 0.8 V (vs. Ag/AgCl). A small amount of highly-concentrated glucose solution (125 mM) was added at an interval of 100 secs to change the glucose concentration from 100 µM to 350 µM. The noise in the measured data was removed by using the Savitzky-Golay filter.

A. MORPHOLOGY AND SURFACE AREA ENLARGEMENT OF HSFE-PI
Despite its excellent mechanical strength, thermal/chemical stability, and biocompatibility [18], PI was not preferred for hierarchical structure fabrication. This disfavor was due to the difficulties in micro-structuring process of PI. The proposed fabrication approach in this study, combining PI mold casting and AuNPs electrodeposition, produces HSFE-PI with relatively low complexity and fast throughput. The hierarchical structure of HSFE-PI is expected to increase the electrode surface area and electrochemical functionalities.
As shown in Figure 5a, HSFE-PI was successfully fabricated with high yield and fidelity. HSFE-PI is composed of VOLUME 10, 2022 100 × 100 micro-pillars at an interval of 50 µm. Figure 5b shows a magnified SEM image of a single micro-pillar. The diameter and height of the micro-pillar were 30 µm and 40 µm, respectively. As the scallop patterns were formed on the sidewall of silicon mold during the DRIE process, the side of the micro-pillar also contains scallops. The electrodeposited AuNPs with a size of 70-100 nm were observed on the surface of micro-pillars at higher magnification (Figure 5c). AuNPs increased the roughness of electrode surface thereby significantly enlarging the surface area.
In order to study the surface area increment in each method, four types of electrodes were compared: planar gold electrode, AuNPs-deposited gold electrode, micro-structured gold electrode, and HSFE-PI (Figure 6a). The footprint area of all the electrodes was identical to be 78.5 mm 2 . Figure 6b shows the cyclic voltammogram of the electrodes in 1 M H 2 SO 4 solution. The cyclic voltammograms which is consistent with previous studies [29], [30], exhibit a reduction peak of gold oxide near 0.9 V. As the enlargement of contact area between gold and solution increases a reduction current, the larger reduction peak indicates the increased surface area. For quantitative analysis on the surface area, the charge involved in gold reduction was calculated for each electrode by integrating the area under the reduction peak [13]. The relative surface area of the electrodes was calculated by comparing the calculated charge. The relative surface area is summarized in Table 1. Theoretical surface area increment by micro-structures can be calculated by equation (1) [33], assuming the micro-pillar solution (scan rate: 50 mV/s). The reduction peak was observed at 0.9 V for all electrodes. As the charge transferred during gold reduction is proportional to electrode surface area, the charge was quantified by integrating the graph area in the reduction state.
as an ideal cylinder.
where S f is the footprint area of the electrode, L is the contour length, h is the height of cylinder, d is the diameter of cylinder, and N is the number of cylinders. As S f is 78.5 mm 2 , d is 30 µm, N is 10 4 , and h is 40 µm, the theoretical surface area increment is 1.48. The measured surface area of the microstructured electrode was 1.49 times larger than the planar electrode, which is almost identical to the theoretical estimation. The electrodeposition of AuNPs on the planar gold electrode increased the surface area by 1.56 times. Hierarchical structures significantly enlarged the surface area, which was verified by a 2.06 times larger surface area of HSFE-PI compared to the planar electrode. Furthermore, HSFE-PI exhibits superior surface area enhancement compared to the previous studies on skin-electrode ( Table 2).

B. ELECTROCHEMICAL CHARACTERIZATION OF HSFE-PI
The transduction efficiency of electrochemical signals was evaluated for the planar gold electrode and HSFE-PI. Figure 7a exhibits the CV response using solution containing 5 mM K 3 Fe(CN) 6 and 1 M KNO 3 at a scan rate of 200 mV/s. HSFE-PI shows substantially higher peak current compared to the planar electrode in the Fe(CN) 4−/3− 6 redox environment. The anodic peak current (I p ) of HSFE-PI and planar electrode was measured to be 1.99 mA and 1.09 mA, respectively. The electrochemically active surface area (ECSA) of the electrodes was determined using the Randles-Sevcik equation [34]: where n is the number of electrons transferred in the redox event, A is the ECSA of the electrode (cm 2 ), D is the diffusion coefficient of the analyte in bulk solution (cm 2 /s), C is the concentration of the analyte (mol/cm 3 ), and ν is the scan rate (V/s). As the Fe(CN) 6 redox environment is a representative system that exhibits a heterogeneous one-electron transfer, n equals 1 [35]. The diffusion coefficient of Fe(CN) 3− 6 in solution consisted of 5 mM K 3 Fe(CN) 6 and 1 M KNO 3 is 5.0 × 10 −6 cm 2 /s [34]. The ECSA of the planar electrode was 80 mm 2 , which is almost identical to the footprint area of 78.5 mm 2 . The ECSA of HSFE-PI was extracted to be 147 mm 2 , which was 1.86 times larger than its footprint area. As shown in Figure 7b, the peak currents have a linear relationship with the square root of the scan rate. This result suggests that the electrochemical reaction of Fe(CN) 3−/4− 6 on the surface of the electrode was a diffusion-controlled process [36]. Thus, the use of HSFE-PI can bring substantial enhancement of sensitivity to electrochemical sensors.

C. MECHANICAL STABILITY OF HSFE-PI
As most wearable devices are used in various non-planar conditions, the reliability of flexible electrodes is an important parameter. A cyclic bending test was performed to confirm the reliability of HSFE-PI (Figure 8). HSFE-PI was repeatedly bent for 10,000 cycles with a bending radius of 5 mm using R-bending system (Sciencetown, Korea). The sheet resistance of HSFE-PI was measured by the 4-point probe method in every 2,000 cycles using setup described in [37]. HSFE-PI exhibited stable sheet resistance throughout the cyclic bending test, which implies that HSFE-PI can retain its outstanding electrical performance while forming a conformal contact with the skin. VOLUME 10, 2022 FIGURE 8. Sheet resistance of HSFE-PI during cyclic bending test with a bending radius of 5 mm. Inset shows experimental setup for the cyclic bending test. HSFE-PI exhibits stable sheet resistance at 10,000 cycles.

D. EMG MEASUREMENTS
EMG signals were measured on the biceps using planar electrodes and HSFE-PIs (Figure 9a). Two signal electrodes were placed on the left bicep with a 5 cm distance, as shown in Figure 9b, and a conventional Ag/AgCl electrode was placed on the right wrist as a reference. As the electrode substrate was thin and flexible, they formed a conformal contact with curvilinear human skin. Figure 9c shows EMG signals measured by the planar electrodes and HSFE-PIs. The amplitude of EMG signal measured by HSFE-PI is much larger than the planar electrode due to the larger surface area, which leads to the lower contact resistance. To calculate the SNR of EMG, the signal during muscle contraction (S S ) and the noise signal without any muscle contraction (S n ) were measured for the same time period. Then, the SNR was calculated by equation (3) [38].
The SNR of planar electrode and HSFE-PI were 5.1 and 12.7, respectively. The HSFE-PI exhibited 2.5 times higher SNR compared to the planar electrode. Comparison of our HSFE-PI electrode with previously reported skin electrodes is summarized in Table 3. The SNR values were normalized by area for a fair comparison, and our HSFE-PI exhibits high sensitivity.

E. GLUCOSE SENSOR CHARACTERIZATION
The glucose sensor was fabricated by immobilizing GOx on the surface of electrodes through GLA/BSA crosslinking method. The detailed glucose sensing mechanism is described by the following equations [41]: (c) Measured EMG signal using the electrodes. Amplitude of the acquired EMG signal using HSFE-PI was larger than the planar gold electrode, which resulted in higher SNR. SNR of the EMG signal for the planar electrode and HSFE-PI was 5.1 and 12.7, respectively.  Immobilized GOx converts glucose into gluconic acid and produces H 2 O 2 (3 and 4). When the produced H 2 O 2 diffuses to the electrode surface, it is oxidized, and an electron transfer occurs (5), which generates current. Cyclic voltammograms of glucose sensors fabricated on the planar gold electrode and HSFE-PI were measured in a solution of 2 mM glucose in PBS (Figure 10a). The HSFE-PI glucose sensor shows a higher current level than that of the planar electrode, which implies higher electrochemical sensitivity. The anodic peak current of the HSFE-PI glucose sensor was 1.42 times higher compared to the planar electrode sensor. The higher sensitivity achieved by HSFE-PI is attributed to its large ECSA as it facilitates an oxidation of the H 2 O 2 . As shown in Figure 10b, HSFE-PI glucose sensor can distinguish different glucose concentrations. Figure 11 shows an amperometric response of the HSFE-PI glucose sensor. The output current was increased as glucose was repeatedly added. The sensitivity of the sensor was calculated by using a two-point method [42], [43], and it was 18.9 µA·mM −1 · cm −2 . Comparison of our HSFE-PI glucose sensors with previous studies is summarized in Table 4. Our HSFE-PI glucose sensor exhibits high sensitivity with excellent flexibility, which demonstrates the potential of HSFE-PI to the wearable electrochemical sensors.

IV. CONCLUSION
In this study, HSFE-PI was fabricated by electrodeposition of AuNPs on a micro-structured electrode. The microstructured electrode was prepared by the PI mold casting process. HSFE-PI exhibited a high surface area, which was 2.06 times larger than the planar electrode. Also, the reliability of HSFE-PI was confirmed by the cyclic bending test. The effects of hierarchical structures were confirmed by both biopotential and electrochemical measurements. The hierarchical structures increased the SNR of EMG by 2.5 times and the sensitivity of the glucose sensor by 1.42 times. The fabricated HSFE-PI glucose sensor exhibited high sensitivity and flexibility. Therefore, HSFE-PI is suitable for the nextgeneration wearable biomedical applications.