A Review of Electrochemical Electrodes and Readout Interface Designs for Biosensors

Electrochemical detection is widely used in biosensing fields, such as medical diagnosis and health monitoring due to its real-time response and high accuracy. Both passive and active electrodes and the corresponding readout circuits have been continuously improved over the past decades. This article summarizes the redox reaction method, state-of-the-art electrode materials, and readout circuits based on the passive three-electrode. The redox-current-based readout circuits are widely used and developed toward multichannel high precision and low power consumption. In terms of active electrodes, this article reviews the development of field-effect transistors (FETs)-based electrochemical detection and readout circuits. In the past decade, the development of organic electrochemical transistors (OECTs) has also enabled more precise electrochemical detection.


I. INTRODUCTION
E LECTROCHEMICAL detection is widely used in medical diagnosis and health monitoring due to its realtime response and high accuracy. Accurate and real-time electrochemical continuous detection is required for drug guidance, surgical intervention plans, and postoperative monitoring [1], [2]. In addition, electrochemical detection is also widely used for the daily health detection, such as sweat analysis [3], [4] and blood monitoring [5], [6], [7], [8].
An electrochemical detector consists of an electrochemical interface and a readout interface. The electrochemical interface acts as a transducer to convert chemical specification into electronic indicators, such as changes in potential, current, or impedance. The readout interface is used to bias the current and potential required by the electrochemical interface and to readout the converted electrical indicators.
The principles of a typical electrochemical interface include redox reactions and target adsorption [9]. For redox-reactionbased electrochemical interface, the oxidation and reduction reactions occur each at two electrodes, which are called the working electrode (WE) and the counter electrode (CE), respectively. With a fixed current bias, the WE's potential is proportional to the concentration of the analyte, which can be readout directly by readout circuits. The CE is usually an inert electrode that provides a reference. However, limited by current control requirements, the potential readout circuit suffers from a low sensitivity, which limits its application potentials [10]. In recent years, current readout circuits typically utilize a three-electrode system to detect the Faradaic currents generated by electron exchange during redox reactions. In addition to the WE and CE where redox reactions take place, a reference electrode (RE) is added to provide a stable and constant potential. The designs of the three-electrode system have been continuously optimized for lower cost, faster speed, and higher accuracy [11], [12], [13], [14], [15].
On the other hand, the impedance of an adsorption-based transducer changes by absorbing or binding target molecules to the transducer. The readout circuits for applying voltage and finding current are widely used as the impedance readout interface. The adsorption-based transducer has been used in transistor designs which turns the concentration into the changes of the transistor's threshold voltage or channel conductivity. A typical representative is the ion-selective field-effect transistor (ISFET). Benefiting from the high integration levels guaranteed by the process, readout circuits of ISFET are becoming fully integrated systems [16]. ISFETbased sensing is widely used for ion imaging [17], [18] and DNA sequencing [19], [20], [21]. In the past decade, breakthroughs in organic electrochemical transistors (OECTs) have also enabled electrochemical detection with higher precision [22], [23], [24]. However, certain challenges of these electrochemical sensors still remain unanswered. From a technological point of view, the main challenges are related to the enhancement of their sensing performance (robustness, selectivity, sensitivity, reliability, etc.) and the reduction of their dimensions and power consumption. Although nanomaterials provide decent electrochemical and mechanical properties [25], they still suffer from electrical instability since it is difficult to control the number, size, and shape of their nanostructures [26]. Efforts should also be made for optimizing the flexibility and stretchability of wearable electronics to provide intimate integration to soft tissues and overcome potential inflammatory response [27]. In addition, a more reliable electrode system with its corresponding readout interface and packaging strategy will improve the long-term stability of these devices in liquid environments. In the future, advanced material strategies and circuit designs will hold great promise for improving the performance of these electrochemical systems.
The remainder of this article is organized as follows. Section II introduces the redox-reaction-based electrochemical interface and the corresponding three-electrode-based readout interface circuits. Section III introduces the development of ISFETs and OECTs and the follow-up of the readout interface circuits. Section IV concludes the entire work.

A. REDOX MEASUREMENT PRINCIPLES
As shown in Fig. 1(a), when controlling the redox current, I redox , one can readout the potential V cell , which allows the calculation of species concentrations. The voltage between WE and RE is denoted as V cell , which is related to the analyte concentrations and is described by the Nernst equation [29]  where V 0 cell is the standard potential of a half-reaction, R is the universal gas constant, T is the absolute temperature, and n is the number of electrons involved in the half-reaction and F is Faraday's constant.
[Red] equals the activity of the reduced species.
[Oxi] equals the activity of the oxidized species. However, the interference of the charging current of the electrode-solution interface double-layer capacitor C we is difficult to be filtered out, which results in low sensitivity [28]. Fig. 1(b) shows the experimental arrangement for currentreadout experiments. When the bias circuit at RE has a large impedance, the current flowing through R re is almost 0. The RE provides a stable and constant potential compared to the WE and avoids the deterioration of the sensing performance. The current generated by redox reactions is directly proportional to the concentration of the analyte. The correlated current flow can be described by the Cottrell equation as shown in the following [30]: where I redox is the current at time t. n is the number of electrons involved in the reaction. F is Faraday's constant, A is the geometric area of the WE, c is the concentration of the oxidized species, and D is the diffusion coefficient of the oxidized species. In general, amperometric and voltammetric techniques are the most common methods used to trace the variation of the redox current. Amperometric sensors depend on the continuously current measurement stimulated by a stable potential controlled of V cell . As one of the most commonly used type of electrochemical sensors, amperometric sensors exhibit rapid robustness, excellent sensitivity, and low detection limit, which is the smallest amount of the analyte that can be reliably detected by the device [31]. However, interfering species that have lower oxidation potentials than the applied voltage could be oxidized during the measurement, causing the amperometric sensors to suffer from poor selectivity [9].
In voltammetric sensors, electrochemical characteristics of the analyte are collected by measuring the current over a controlled variation of the applied voltages [32]. For example, the potential at which the target analyte begins to be oxidized or reduced can be recorded via voltage sweeping when the resultant current shows a peak or trough due to the mass transport effect. Different time-voltage relationships result in different voltage shapes including linear sweep voltammetry (LSV), cyclic voltammetry (CV), differential pulse voltammetry (DPV), and square-wave voltammetry (SWV). In LSV, voltage is scanned from a lower limit to an upper limit in only one direction (either positive or negative) with a preset sweeping rate. CV changes the sweep direction at a fixed rate between two values to generate cycles, in which case the target analyte may undergo multiple redox cycles. Therefore, a reversible electrochemical process recorded by CV has a well-defined feature. CV is a powerful electrochemical technique widely used for investigating the redox processes and relevant characteristics (e.g., redox potentials and electrochemical rate constant) involved in the electrochemical reaction of interests [33]. DPV can be used to obtain accurate and reproducible values of the redox potential with higher selectivity and sensitivity. In the DPV experiment, the amplitude of the current is measured in a forward pulse and a reverse pulse, and the current difference between the forward and reverse pulses is recorded. Therefore, the influence of capacitor current is minimized by the subtraction of current [34]. SWV derives the I-V characteristics from the application of the potential in the form of rectangular waves. SWV can achieve better capacitor current suppression for higher sensitivity while requiring less time per scan than DPV [35].

B. ELECTRODE MATERIALS
As redox reactions only occur at the surface of the WE, materials have a significant impact on the performance of the electrochemical analysis. Typically, these materials should exhibit ideal electrical conductivity, chemical inertness, mechanical properties, and low background current resulting from redox reaction under the applied potential. Noble metals (Au, Pt, Ag, Pd, etc.), and carbon-based materials (glassy carbon, carbon nanotubes, graphene, carbon nanofibers, etc.), metal-oxides (ZnO, WO 3 , TiO 2 , CeO 2 , etc.) are the main materials used for WEs [50], [51]. Other 2-D nanomaterials, such as black phosphorus (BP), MXenes, and transition metal dichalcogenides (TMDCs) which have high electroactive surface area, have also shown significant improvement in performance of sensors [52], [53]. In addition, some efforts have been made on the modification of the WEs to achieve fast electron transfer kinetics, increased sensitivity and selectivity and stable electrode surfaces [54]. Inert conducting materials (e.g., platinum or graphite) are often utilized as CEs and the normal hydrogen electrode (NHE), saturated calomel electrode (SCE), and saturated silver/silver chloride electrode (Ag/AgCl) are most commonly used for REs.
Based on these materials, electrochemical sensors have wide applications in health care, disease diagnosis, environmental monitoring, food evaluation, and drug delivery by identifying or detecting biomarkers of interest (e.g., dopamine, nitric oxide, glucose, lactate, etc.). Table 1 presents the list of recent electrochemical sensors, materials, target analytes, and the detection range. However, it still remains a challenge to accurately monitor these biomarkers in biological systems since their concentrations significantly vary in different organs or tissues [55] and most of them exhibit a short half-life time ranging from milliseconds to minutes.

III. THREE-ELECTRODE BASED READOUT INTERFACE
The readout interface used for voltage (V cell ) biasing and current (I redox ) readout in a three-electrode system is called a potentiostat. Typical three-electrode potentiostats include a resistance feedback TIA (R-TIA), a capacitance feedback TIA (C-TIA), and the current mirror (CM) circuits. R-TIAs and C-TIAs circuits convert read redox currents to voltages and ADCs are used for conversion from analog voltages to digital signals. As for the CM circuits, after mirroring the current to be measured, the current is converted into frequency or other electrical parameters. The TDCs are used to quantize time signals into digital signals.

A. RESISTIVE FEEDBACK CURRENT READOUT CIRCUITS
The most popular resistive feedback potentiostat is a singleended circuit, as shown in Fig. 2. The electrochemical reaction potential V cell = V WE − V RE . The potential of WE, V WE , is connected to the virtual ground, while the potential of RE, V RE , follows the applied voltage V IN . Thus, the value of V cell changes with V IN . The output voltage V out is proportional to the redox current I redox and the feedback resistance. The readout of small currents requires a high conversion gain for the R-TIA. For example, an input current of 1 pA requires a 1-G resistor to reach 1-mV output, which is difficult to be implemented in standard complementary metal-oxide-semiconductor (CMOS).
Potentiostats need to be able to provide output voltages of different magnitudes for different chemical analytes. However, the operation of a single-ended potentiostat is limited by the swing of the control amplifier for lowsupply voltage chip applications and can be affected by common-mode interference. In order to adapt to various scenarios and solve the problem of small output swing of single-ended potentiostats in low-voltage applications, Ghodsevali et al. [56] and Martin et al. [57] proposed the potentiostat with a fully differential structure, as shown in Fig. 3. R sense is used as a current sampling resistor, and R sym keeps the circuit structure symmetrical. This structure nearly doubles the output swing of the voltage-controlled amplifier, as shown in This allows the potentiostat to accommodate a larger dynamic output range of the sensor. In addition, the fully differential structure can better suppress common mode noise. This structure is difficult to apply to the detection of small currents. R sense should not be too large to reduce the voltage drop. Besides, R sense and R sym are directly connected in series in the WE and CE loops, and their thermal noise introduces a large input error.
In order to integrate large resistors on-chip, a pseudoresistor can be considered. The traditional pseudo-resistor structure consists of two back-to-back MOS transistors as diodes. When they work in the weak inversion region and there is no dc current, a high equivalent resistance value can be achieved. However, the PVT (process, voltage, and temperature) robustness of traditional pseudo-resistors is poor. Besides, when the dc current is nonzero, the noise performance is worse than that of normal resistors. Therefore, Djekic et al. [58] proposed an improved pseudoresistive feedback TIA, as shown in Fig. 4. Djekic et al. [58] described a method of connecting pseudo-resistance elements in series to reduce mismatch and improve the linearity by reducing the voltage drop over the pseudo-resistance unit. At the same time, the series connection of pseudo-resistor elements forces it to work in the linear region to reduce the spurious noise. The silicon on insulator (SOI) technology reduces parasitic effects and alleviate stability problems. The transimpedance value can be adjusted between 1 M and 1 G with corresponding bandwidths from 8 kHz to 2 MHz. The 1-M transimpedance change is less than 10% over a temperature range from −40 • C to 125 • C.

B. CAPACITIVE FEEDBACK CURRENT READOUT CIRCUITS
In order to realize an integrated low-noise potentiostat, another common practice is to use C-TIA to form an integrator-differentiator structure. Compared with the R-TIA with limited resistance size, the C-TIA potentiostat is not limited by resistance thermal noise, and the input equivalent noise can be lower [8], [59]. But the leakage current at the input node will cause the integrator to saturate. There are two main ways to solve this problem. One is to connect a switch in parallel with the feedback capacitor to periodically reset the feedback capacitor. The second is to add a dc feedback loop to extract the input low-frequency current. This allows the integrator to discharge continuously and avoid saturation. VOLUME 3, 2023 79 Authorized licensed use limited to the terms of the applicable license agreement with IEEE. Restrictions apply.

FIGURE 6. CDS technique (adapted from [60]).
As shown in Fig. 5, the switch-based structure cannot process the input signal when it is periodically reset. The dc offset noise usually causes problems such as saturation of the readout circuit. In order to eliminate the offset error, a correlated double sampling (CDS) circuit has been applied to the capacitive feedback readout circuit [60]. CDS circuits can also reduce 1/f noise. CDS removes low-frequency noise by sampling the input twice and recording the difference between the two samples. Fig. 6 shows an example CDS circuit where two sampling actions are implemented by a set of switches controlled by two clocks with opposite phases (ϕ1 and ϕ2).
The ability of this structure to handle small currents at high frequencies is limited. Dai et al. [61] proposed a selftimed switched capacitor reset network, as shown in Fig. 7. Two complementary control signals ϕ1 and ϕ2 are used alternately for C i1 and C d1 or C i2 and C d2 . So, there is always one capacitor in the working state and the other in the reset state. The system is self-timed and does not require an additional reset clock. The reset signal comes from the comparison of the output voltage of the integrator with the set threshold voltage. Both the integrator and the differentiator are reset in a charge-conserving manner, which minimizes the reset transient and recovery time. Furthermore, this configuration has inherent instant switching between voltage mode and frequency mode, which increases the dynamic range of the input current. The disadvantage is that at the moment of reset, the output voltage will have a small pulse due to mismatch.
Ferrari et al. [62] proposed a dc feedback loop based on the integrator-differentiator structure, as shown in Fig. 8. The dc gain of the dc feedback loop is where A is the gain of the operational amplifier. At low frequency, the feedback is strong enough that I DC flows into R dc . At high frequency, the feedback fails and dc response of the output node is zero. Therefore, integrator saturation is avoided and the output dynamic range is increased. A large resistance on the order of 1 M -1 G is required to ensure that the dominant pole of the dc feedback loop is small enough. Ferrari et al. [62] proposed an accurate current reducer based on a transconductor, as shown by the dashed box in Fig. 8. Since the dc feedback loop does not affect the signal of the desired frequency, the linearity of the resistance on this path is sufficient.

C. CURRENT-MIRROR BASED READOUT CIRCUITS
A CM-based potentiostat is another common current sensing structure, as shown in Fig. 9. This structure does not directly convert current to voltage. It mirrors the redox current and converts the current into a time or frequency signal through a modulator [13], [63]. However, the noise performance of the CM readout circuit is usually limited by the optional

FIGURE 9. Current-mirror-based readout circuits (adapted from [13]).
amplifier of the voltage bias circuit. Chopping techniques can be used to reduce bias circuit noise [4]. A current compensation circuit can be used to achieve bidirectional current sensing and output range extension [64]. In order to reduce the influence of the CM's channel length modulation effects, the large channel length or an enhancement CM topology can be used, such as a cascode or gain-boosted structure.

D. CASE STUDY OF PASSIVE-ELECTRODE-BASED READOUT INTERFACE
A summary of the performance of passive-electrode-based readout interfaces is shown in Table 2. For resistance feedback TIA(R-TIAs), since the large resistance is difficult to integrate on the chip, the thermal noise of the resistance limits the improvement in detection accuracy. An equivalent large resistance is achieved by using the pseudo-resistor, which can be used to detect small currents. However, as mentioned earlier, its PVT characteristics are not stable, and the high resistance makes it difficult to achieve a large bandwidth. Capacitive feedback TIAs (C-TIAs) can achieve low noise and large bandwidth, but requires additional circuits to avoid integrator saturation. The design of the TIA-based potentiostat is a compromise between the performance and bandwidth. The CM potentiostat features low power consumption, small area, and a large dynamic range of detection current, making it suitable for multichannel applications. However, its bandwidth is relatively low and the sensitivity is relatively poor.

IV. ACTIVE ELECTRODES INTERFACE
In the passive electrode, the Faradaic current of the redox reaction is modified by the charging current of the double layer capacitor. As shown in (2), the amplitude of the redox current is proportional to the area of the WEs, what prevent scaling down the area. Active electrodes can change this issue using adsorptionbased transducers. In the past 60 years, carrier-based ion-selective electrodes have been extensively studied and combined with the field-effect transistors. The response time of the ISFET is short. In addition, ISFETs can be fabricated using CMOS technology, opening up new opportunities for the development of compact system solution [68]. The first integrated multispecies electrodes, called extended gate chemically sensitive field effects (EGFETs), were introduced in 1983 [69]. Organic FETs (OFETs) represent a further development in materials. The easy functionalization of the electrodes is guaranteed by synthetic chemical modification. In the past decade, an OECT demonstrated its advantage using aqueous media. OECT uses hydrated, ion-permeable conducting polymers that change the OECT's conductivity through reversible ion exchange with electrolytes. High SNR is guaranteed due to its extremely high transconductance. In recent years, the main principle of OECT-based electrochemical sensors is to amplify redox voltage into current by utilizing the high transconductance.

A. ISFET AND READOUT CIRCUITS
The first ISFET was introduced by Bergveld in 1972. The schematic diagram and equivalent circuit of the ISFETs are as shown in Fig. 10. A typical ISFET consists of three terminals (source, drain, and gate) and an RE. Compared with three-electrode system, ISFETs exhibit a higher sensitivity with a smaller size. Only one RE is required to achieve a biochemical measurement. The working principle of ISFETs

FIGURE 10. (a) Structure and (b) symbol of ISFET (adapted from [16]).
is based on the variation of the conduction of the channel adjusted by the potential on the RE. A positive bias applied on the RE will push the cations away from the gate surface, which results in a negative charge accumulation in the insolation layer. The charge redistribution on the gate will influence the threshold voltage (Vth) of ISFETs, leading to a change of the current I DS . This process can be modulated by the surface reactions where the species of interest interact with the gate electrode.
The material studies on the ISFET most focus on the gate materials. High-k materials (including Si 3 N 4 , AL 2 O 3 , HfO 2 , ZrO 2 , etc.) are most common used as gate materials. By redesigning the structure of the device (e.g., extended gate structure, the Fin-gate structure, and stacking layers of gates), the performance of the ISFETs can be further improved. In addition, the modification of the gate electrode by introducing a microstructure on the gate or functionalizing with nanomaterials can improve the response time, sensitivity, and selectivity of the devices [80]. In addition, the functionalization of the ISFET gate gives more possibilities of the device to detect nonlabelled analytes, such as glucose, DNA, enzymes, protein biomarkers, and cell-related secretions or metabolism [80], [81].
For ISFETs, the threshold voltage V th can be expressed as [82]  where E ref is the constant potential of the reference electrode. is a chemical parameter, which is a function of the pH of the solution. χ sol is the surface dipole potential of the solvent and therefore has a constant value. Si is the work function of silicon, Q ox is due to the charge accumulated in the oxide, Q ss is the charge at the oxide-silicon interface due to process defects, Q B is the depletion region charge, and φ f is the Fermi potential. Table 3 summarizes the materials and properties of recent works on ISFETs and readout circuits. The readout circuit is critical for the drift and temperature issues of electrochemical sensing using ISFET. As shown in Fig. 11, a typical constant-voltage and constant-current (CVCC) test circuit generally biases the ISFET in the strong inversion region to ensure the linear relationship between I d and V GS . The I d and V DS of the ISFET are guaranteed to be fixed by I 1 and V 1 , and V S will be read out as V out when the pH changes. The change of source voltage V S is proportional to the pH. CVCC circuits are immune to capacitive splitting caused by the passivation capacitors. However, the stability issues and difficulty in achieving inherent compensation limits the overall performance [16].
The current mode readout circuit has been widely used in recent years [83]. As shown in Fig. 12, with proper biasing, both MOS and ISFET in the circuit are biased in the weak inversion region. There is an exponential relationship between I d and V GS . Therefore, it can be deduced that the output current I out is proportional to the H + concentration as  where n is the subthreshold slope parameter. γ is a grouping of pH-independent chemical potentials and U T is the thermal voltage [83]. A combination of potentiostats which is used for threeelectrode systems can also be multiplexed with ISFETs as shown in Fig. 13 [84]. The voltage V D is biased by the potentiostat 1. The voltages V G and V S are biased by potentiostat 2. The current I DS is readout by potentiostat 2.

B. OECT AND READOUT CIRCUITS
The first OECT was reported by White et al. [101] in the mid-1980s. As shown in Fig. 14, metal electrodes (a source, a drain, and a gate electrode) and an organic semiconductor film lying between the source and drain electrodes represent a typical structure of OECTs. Different from conventional FETs, OECTs rely on ions injection or extraction from the electrolyte into the organic films, leading to changes of its doping state and conductivity.
As a typical material for OECTs, the reversible doping process of conducting polymer poly(3, 4ethylenedioxythiophene) doped with poly(styrene sulfonate) (PEDOT:PSS) can be described as [102] PEDOT + : PSS − + M + ↔ PEDOT 0 + M + : PSS − + h + (7) OECTs based on PEDOT:PSS operate in a depletion mode. During the operation, a positive gate bias pushes positive cations (M+) from the electrolyte into the PEDOT:PSS layer and the hole density in the PEDOT+ is reduced followed by the decrease of the channel current [103]. Compared with ISFET in which the physical thickness of the channel does not tune the performance of the device, volumetric gating gives OECTs a much larger transconductance than that of ISFETs up to a certain frequency range (up to kilohertz) [104]. When serving as an electrochemical sensor, the redox reactions on the gate electrode cause an increase in the potential at the electrolyte/channel interface which pumps extra positive cations (M+) from the electrolyte into the PEDOT:PSS channel [105]. The whole process can be described by the following equation based on Bernard's model [106]: where V g−eff is the change of the effective gate voltage generated by the redox reactions; γ is the ratio between the electrolyte/channel capacitance and the gate/electrolyte capacitance; k is the Boltzmann constant, T is the absolute temperature, q is the charge of an electron, and (analyte) is the concentration of species of interest; and C is a constant. Table 4 summarizes active materials and key electrical parameters of recent works of OECTs. Apart from PEDOT:PSS, other conjugated polymers can also be used to fabricate the channel of OECTs. These conjugated polymers, as well as PEDOT:PSS, all exhibit attractive electrical properties with simple fabrication process. For typical OECTs, the potential range of both the drain-source voltage (V ds ) and the gate-source voltage (V gs ) is −1 to +1 V. The drainsource current (I ds ) is around few mA to dozens of nA. Generally, OECTs exhibit a transconductance up to dozens of mS during operation at low frequencies [104].
OECT has been used for various electrochemical detections that can be performed using traditional three electrodes configuration [22], [23], [24]. The readout circuits of the three-electrode systems are also being used for OECTs. This requires adding a voltage bias and properly adjusting the range of the readout current. As shown in Fig. 15, Tian et al. [24] used DAC 2 to configure the voltage V g .

FIGURE 15. Readout circuit for OECT in [24].
Due to the high SNR of OECT, the circuit design has not been very challenging yet. However, the nonlinearity of the readout current and the power consumption caused by the high-current I DS are worth improving. Similar to ISFETs, OECTs will also develop toward miniaturized and arrayed applications, which also require adapted readout circuits.

V. CONCLUSION
For decades, electrochemical detection based on redox reactions has been widely used. This article analyzed the challenges in the electrochemical interface and summarized the response time required for common analytes and the corresponding concentration ranges for different electrode materials, which help the design of readout circuits to achieve tradeoffs between accuracy, bandwidth, and power consumption. In recent years, the development of active electrodes has solved many difficulties of passive electrodes in the improvement of temporal and spatial resolution. This article also reviewed and analyzed the key performances of ISFETs and OECTs published in recent years. The readout circuits of ISFETs require multichannel and adaptive nonideal compensation. There is no published OECT readout circuit, but it can be expected that the circuit needs to be optimized for power consumption while maintaining a high signal-to-noise ratio.