Flexible Sensor and Readout Circuitry for Continuous Ion Sensing in Sweat

In this letter, we present a custom-designed and flexible readout circuitry for the characterization of flexible and planar electrolyte-gated carbon nanotube field-effect transistor-based biosensors for ammonium detection in sweat. We employed spray-deposited semiconducting carbon nanotubes as active material, and functionalized the devices with previously synthesized ion-selective membrane, based on the nonactin ionophore. In the design of the readout circuitry, we focused on enabling low-power operation with a single coin-cell battery with compact wireless data transmission. To maximize the bendability of the flexible PCB, particular attention was taken in layout routing as well as in selecting small-sized packages for the commercial integrated circuits and all-in-one systems such as programmable Bluetooth system-on-chip (i.e., ST microelectronics BlueNRG SoC). We recorded a high sensitivity of 4.516 $\mu$A/decade for sweat ammonium level analysis between 0.1 and 100 mM, which fully covers the relevant range of interest in the context of sport monitoring. The characterization was carried out with the introduced front-end readout, and the results were benchmarked with a “gold standard” instrument, showing good and reliable performance of the developed fully flexible bioelectronic system.


I. INTRODUCTION
Wearable technology has demonstrated a tremendous breakthrough in the last decade by enabling continuous monitoring of the physiological status [1], [2]. In particular, sweat analysis has opened new pathways in understanding the body's biochemistry, as it represents an optimal candidate to replace the conventional and invasive blood biosensing [3]. Sweat, in fact, is easily accessible from the skin, and enables the extraction of a significant amount of information, being rich in biological indicators [4], [5].
Among the various biological indicators present in sweat, ammonium (NH + 4 ) has recently been recognized as an important parameter in the context of sport science [6], as an up-and-coming key biomarker for the metabolic breakdown of proteins, especially in the physiological concentrations (0.12-2.17 mM) [7]. The most widespread methodology for the biosensing of NH + 4 in sweat is based on electrochemical sensors [8], [9]. Electrolyte-gated field-effect transistors (EG-FETs) represent a promising alternative sensing platform [10], [11]. In particular, carbon nanotube (CNT)-based EG-FETs (EG-CNTFETs) enable the exploitation of the favorable properties of CNTs for the fabrication of highly sensitive biosensing platforms [12], [13]. Moreover, EG-CNTFETs can be easily integrated onto flexible polymeric foils including, among the others, polyimide (PI) and polyethylene terephthalate (PET) [14], [15]. Such flexible substrates enable the integration of sensors in wearable platforms of arbitrary shape, with the possibility of meeting the requirements (e.g., low-cost, large area, flexibility, and conformability) needed to achieve proper market penetration [16].
Nevertheless, current integration schemes present limitations that hinder the seamless and reliable connection between the sensing element and the readout circuitry. Traditional fabrication processes, such as wire bonding or flip-chip bonding, may not be compatible with flexible substrates (e.g., because of the high temperatures required), restricting the types of sensors and readout electronics that can be employed. This may lead to challenges such as signal loss, noise, and poor sensor performance [17]. In this work, we have addressed This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/ this issue from the system perspective. First, we fabricated flexible and planar EG-CNTFETs on a PI substrate, functionalized with an ion-selective membrane, for NH + 4 detection in the concentration range of 0.1-100 mM. Afterward, we fabricated a fully flexible coin-cell battery-powered readout circuitry by employing a small-sized, i.e., 40 mm × 45 mm PI substrate [see Fig. 1(a)]. To maximize the flexibility of the circuit, particular attention was given to minimizing the complexity of the hardware and the number of components. Later, the proposed fully flexible readout circuitry was first validated by characterizing the EG-CNTFET-based NH + 4 sensors, and then, benchmarked with a benchtop semiconductor parameter analyzer, which is a "gold standard" reference point. The results showed good sensitivity of the sensors in the physiological range of NH + 4 in sweat during physical activity, and that the flexible readout circuitry is fully able to replace the bulky benchtop instrumentation, paving the way for the integration of the sensors in wearable platforms.

A. EG-CNTFETs Fabrication and Functionalization
The layout of the EG-CNTFETs consisted of interdigitated electrodes for source and drain and a planar, U-shaped gate electrode [see Fig. 1 [18]. 95% semiconducting CNTs (Merck, catalog number 773735) were employed as active layer. Five devices (N = 5) were fabricated by following the procedure optimized in [19]. The devices were functionalized with an NH + 4 -selective membrane, based on the nonactin ionophore [8]. 10 µL of the membrane cocktail was drop-casted on the active area of the devices (8.85 mm 2 ), and the devices were stored for 12h at 4 • C [8], [15]. They were then stored at room temperature, immersed in a 1 mM NH + 4 solution in artificial sweat to achieve conditioning of the ion-selective membrane. The artificial sweat was prepared by mixing 8 mM KCl, 30 mM NaCl, 0.084 mM creatinine, 0.17 mM glucose, 0.059 mM uric acid, and 0.01 mM ascorbic acid in deionized (DI) H 2 O, with a resistivity of 18.2 M cm [9]. Before and after each measurement, the devices were gently washed with fresh artificial sweat.

B. Readout Circuitry Design
As illustrated in the hardware schematic depicted in Fig. 2, the proposed circuitry comprises two main blocks: front-end readout and power management circuitries. During all the phases of the development (i.e., circuit design, choice of commercial components, and layout routing), the focus was on minimizing the number of components, and on placing them in such a fashion that, when bending the PCB, the parts of the board with physical components remained stationary, while the board itself flexed along predefined horizontal and vertical lines. This strategy allowed us to create a flexible circuit without compromising its functionality.
The first stage of the readout block uses a high-speed (90 MHz Gain Bandwidth), low-noise, precision trans-impedance amplifier (OPA380AID, Texas Instruments, USA) to convert the sensed current into voltage and also to bias the EG-CNTFET-based sensor's source. In its feedback loop sits a digital potentiometer (AD5122A, Analog Devices, USA), whose value is set with the microcontroller (BLUENRG-M2SP, STMicroelectronics, Switzerland) connected via an I 2 C serial communication bus. Next, the sensed voltage is digitized with a 24-bit low-power analog to digital converter (ADC, ADS1251U, Texas Instruments, USA). The high 19-bit effective resolution is critical to increasing the accuracy of the measurement. The digitized data are continuously fed to the microcontroller via Serial Peripheral Interface (SPI). Finally, the embedded Bluetooth low-energy (BLE) module transmits the sensed information to the custom-designed user interface for the data postprocessing.
The power management block is responsible for charging the 3.7 V lithium-ion coin cell battery, regulating, and generating the required stable bias voltages for the operation of the transistor-based sensor. The battery charge management controller (MCP73871, Microchip Technology, USA) comes with internal load-sharing capability, which enables the charging of the battery while maintaining its operation for a prolonged measurement duration. An efficient (96 %) synchronous boost converter (TPS61032, Texas Instruments, USA) was used to generate the regulated 5 V supply. Regulated 5 V was chosen to be acquired as input to the 3.3 V dc-dc buck down converter instead of the battery's initial 3.7 V (which drops down to ∼3.0 V upon battery usage), to avoid misoperation of the converter. Later, the 5 V is converted to 3.3 V with a low-dropout voltage regular (TPS75333, Texas Instruments, USA) for the micro-controller's operation. Similarly, the regulated 5 V is converted to 2.048 V and 1.235 V to generate reference voltages for biasing the EG-CTNFET-based sensor's source and gate, V S and V G , respectively. The gate voltage reference is generated by a band-gap reference connected externally to a 36 k resistor (LM385-1.2, Texas Instruments, USA), since the sensor can tolerate the typical 1% to 2% tolerance of band-gap voltage references in its gate. However, the drain-source voltage is required to be ultrastable to minimize measurement errors due to readout noise. Therefore, a precision voltage reference with an accuracy of sub 1% is implemented to produce the source bias voltage (LM4132, Texas Instruments, USA). Subsequently, a factory-fixed 10:1 resistor divider network with a

C. Readout Circuitry Validation-EG-CNTFET-Based NH + 4 Sensors Characterization
To verify the performance of the proposed flexible readout circuitry when flexed, a static bending test was carried out. The circuitry was wrapped around cylinders of different radii (i.e., 13, 20, 28, 35, 50, 80 mm, see the inset of Fig. 3), and the V G , V D , and V S voltages were measured continuously with a digital multimeter (UT161E, Uni-Trend Technology, China) for 90 s.
A semiconductor device parameter analyzer (SDA) (B1500A, Keysight Technologies, USA) was used as a standard reference instrument for the benchmark validation of the proposed custom-designed circuitry. We utilized a standard probe station to interface the flexible sensor with both the custom-designed circuitry and SDA. This approach ensured that any potential nonideality associated with contact and cable connections affected both units equally, thus maintaining comparable recordings for both setups. The drain-source current I DS was recorded in real-time while keeping the biasing fixed at V GS = −0.81 V and V DS = −0.186 V, to be perfectly in parallel with the voltages applied by the on-board reference voltage employed in the flexible circuitry. Each device was tested with the SDA, and then, with the developed circuitry, with three different values for the digital potentiometer (R FB = 10 k , 39 k , and 82 k ). The source, drain, and gate electrodes were covered with 200 µL of artificial sweat, and the I DS was continuously monitored for 70 min to let the device stabilize [20], [21]. Without interrupting the measurement, the EG-CNTFETs were then tested for the detection of five different NH + 4 concentrations: 0.01, 0.1, 1, 10, 100 mM. An interval of 10 min was employed between two consecutive additions. At the end of each test, the devices were washed with fresh artificial sweat, and between the different tests, they were stored immersed in the conditioning solution for 24 h.

III. RESULTS AND DISCUSSION
The results of the bending test confirmed how the readout circuitry performances are uninfluenced by the bending of the substrate. The results for the smallest tested bending radius (i.e., 13 mm) are reported in Fig. 3. The voltages produced by the circuitry perfectly overlap between the flat condition and the bent condition. Analogous results were collected for the other bending radii.
To correctly compare the measurements carried out with the proposed flexible circuit with the ones taken from the SDA, we estimated the I DS by where V TIA is the output voltage of the trans-impedance amplifier (TIA) and V S the voltage at the source electrode of the EG-CNTFET. From (1), it is straightforward that saturation of the TIA can occur increasing the value of R FB . In this respect, the value of R FB sets the tradeoff between sensitivity and the dynamic linear detection range of the measurement (i.e., of the TIA). By taking into account the highest value of R FB tested, i.e., R FB = 82 k , it could be calculated a theoretical upper limit for the |I DS | of 30 µA. In fact, we observed saturation of the TIA for 4 out of the 5 tested devices, when testing R FB = 82 k . For R FB = 39 k and R FB = 10 k , such issue was not observed.
The corrected response of one of the tested EG-CNTFET-based biosensors (i.e., I * DS = I DS after the baseline correction [18]) to the different NH + 4 concentrations and R FB values is depicted in Fig. 4. For R FB = 10 k and R FB = 39 k , the calibration curves showed very good overlap with the measurement taken with the SDA, hence proving the good operation of the proposed circuitry. As mentioned before, the measurements taken with R FB = 82 k suffered from the saturation of the TIA, hence the calibration curve did not show the expected behavior and moved away from the measurement taken with the SDA. For all the different testing conditions, the average values of the sensitivities for the five tested devices are reported in Table 1. All the tested devices showed selective detection in a complex electrolyte such as artificial sweat, demonstrating the robustness of the EG-CNTFET-based sensing platform. The values of sensitivities were obtained from the fitting of the I * DS over the logarithm of the NH + 4 concentrations. One of the advantages of the proposed approach resides in its digital readout: with a sufficient amount of calibration curves, the microcontroller can be programmed to associate the electrical reading to the actual concentration through look-up tables or statistical inference. The measurements taken with R FB = 10 k proved to be the closest to the ones taken with the SDA, followed by R FB = 39 k . The difference in the value of the relative standard deviation (re-STD) can be addressed to the fact that the devices were kept stored in artificial sweat between the tests [22], [23].

IV. CONCLUSION
In this letter, we presented the design and validation of a flexible readout circuitry for the characterization of flexible and planar EG-CNTFET-based biosensors for NH + 4 detection. The biosensors were characterized using artificial sweat (a closed mimic of actual human perspiration) as electrolyte, for the detection of NH + 4 in the range from 0.01 to 100 mM. This experiment served as a proof-of-concept to test the functionality of the proposed readout circuitry: further characterization of the EG-CNTFET-based NH +