Review—Field-Effect Transistor Biosensing: Devices and Clinical Applications

Yu-Cheng Syu; Wei-En Hsu; Chih-Ting Lin

doi:10.1149/2.0291807jss

Ultrasensitive biosensors with emerging micro/nano technologies attract attentions due to potentials of early detection, emergence of precision medicine, genetic diagnosis, and gene sequencing. As different sensing methods are developed, these advancements translate academic biosensing researches into practical in vitro diagnosis (IVD) tools. At the same time, improvements of test strip and instrumentations are undergoing a revolutionary transition from electrochemical/optical to nano-electronic technologies. Therefore, it is worthwhile to take a glance at the electronic-based biosensing devices, which have potential to be one of the major sensing technologies for medical diagnosis applications in near future.

A biosensor, defined by International Union of Pure and Applied Chemistry (IUPAC), is a device that uses specific biochemical reactions mediated by isolated enzymes, immune systems, tissues, organelles or whole cells to detect chemical compounds by electrical, thermal or optical signals.¹ In other words, a biosensor is an analytical tool to monitor dynamics and interactions of biological activities, e.g. DNA hybridization and cell activities. It transfers the monitoring results into electrical signals. A basic form of a biosensor is comprised of three major parts: a bio-recognition element, a transducer, and a signal processing unit. The signal-transducing pathway starts from different kinds of physical quantities, namely charges, mass or photons, changed by the biomolecules. Then transducers sense these physical changes and transfer them into measurable electrical signal (i.e. current or voltage). In the end, the signals are amplified and processed.

Clark and Lyons first introduced the enzyme-electrode biochemical-based biosensor in 1962.² Following this work, a variety of biosensors and sensing mechanisms were reported. In these bio-sensing mechanisms, an interesting sensing approach, i.e. FET-based biosensor, has been proposed and become an emerging field because of fast development in solid-state technologies. Because most of biomolecules carry electrostatic charges and bioactivities involve electrical potential changes, the FET-based biosensor becomes a promising candidate for applications requiring ultra-sensitivity and fast response time. Furthermore, modern complementary metal-oxide semiconductor (CMOS) manufacturing techniques provide advantages of miniaturization, parallel sensing (e.g. sensing arrays), and capabilities to be integrated with electronic circuits and systems. This would be a major advantage for solid-state based biosensors to compete with other bio-sensing mechanism in the future.

To show capabilities of FET-based biosensors, different biomolecular targets were demonstrated in the past decade.^{21–29,31,32,36–44,58–66,72–79} For instance, FET-based biosensors were used to detect nucleotides, amino acid, as well as cells. These targets are typically important biomarkers for clinical diagnosis of diseases, such as cardiac diseases, kidney injury, diabetes, cancers, inflammatory, and infectious diseases. On the other hand, the nucleic acid sensors (i.e. DNA sensors)^{25–29,36–40,58–61,72–76} rely on DNA hybridization for target recognition. Other potential applications are virus or bacteria detection for infectious disease diagnoses,^53,118,119 e.g. AIDS and hepatitis B, as well as other important bio-analytes, such as metabolites.^3,4 With these developments, FET-based biosensors have been addressed to be a good candidate for next generation point-of-care testing (POCT).

POCT is a branch of IVD. It is defined as a medical diagnostic testing at the point where patient care is needed.^5,6 For instance, a glucometer is used at patients' house or bedside, rather than central laboratory in medical centers. To accomplish POCT, its operation should be simple enough for less-trained or even un-trained person. And it dramatically decreases the time to test results, which would provide doctors instant information for correct diagnosis. Due to the advantages of sensitivity, selectivity, miniaturization, light-weight, low-power, and seamless for fitting in electronic read-out systems, the FET-based biosensors become competitive candidates for future POCT applications in contrast to bulky optical-based IVD instruments. In order to achieve fully automatic testing operation, moreover, microfluidic cartridges for sample preparation can be also integrated with FET-based sensing devices.

In this review, the first part focuses on how FET-based biosensors operate as sensing devices. Working principles of four different type of FET-based biosensor are discussed in details, as well as the biomedical applications. The second part of this review addresses disease-based clinical applications and DNA sequencing. Furthermore, we discuss how FET-based biosensor can be designed by CMOS platform and the integration with sample processing and data processing apparatus for clinical sample testing. These technologies by clinical application is categorized, and a brief summary and comparison for FET-based biosensor solutions are provided to each clinical problem.

Principles of FET-Based Biosensing and Solid-liquid Interface

FET-based devices are semiconductor devices comprised of metal-oxide-semiconductor (MOS) structure. When a metal potential (ψ_m) is changed, the electric field induces the band bending of the semiconductor channel accordingly. It results in channel carrier concentration changes, such as accumulation, depletion or invertion. In conventional MOSFET, we intentionally apply gate voltage to invert the channel (i.e. V_G > V_th) and turn on the transistor. Similarly, the gate potential can be given by other factors, such as solution potentials (e.g. pH value) or charge of biomolecules. These factors influence the status of channel carriers and make the current-voltage characteristics shift positively or negatively. Then the characteristic change is presented by the difference of threshold voltage. Therefore, the difference of threshold voltages (ΔV_th) can be calculated as an indicator of sensitivity.

Figure 1a depicts a simplified potential diagram for a general electrolyte-insulator-semiconductor (EIS) model,^7,8 which is an analogous to MOS model with the replacement of metal by electrolyte. As previously discussed, the difference of surface potentials (ΔΨ) majorly contribute to the status of channel changing into inversion. Moreover, the potential difference is originated from the change of charge concentration on surface (Δσ₀) and it can be attributed to the contribution of intrinsic charge from biomolecules or ion released from enzymatic reactions.

This sensing mechanism could be observed more detailed when the solid-liquid interface model is adopted. The electrical double layer (EDL) model⁹ is used to describe the behavior of solid-liquid interface. Due to local coulomb force, ions in solution are separated into three layers: Stern layer, diffuse layer and bulk solution. Stern layer, also known as the stationary layer, forms at the solid surface. Ions in Stern layer are attracted by charges in solid and bound to the surface firmly. Therefore, the electric potential decreases linearly from the interface to the solution. Next to Stern layer, a diffuse layer has ion concentration changing as a function of the distance from the interface. The ion concentration function follows Maxwell-Boltzmann statistics and the electric potential in the diffuse layer decreases exponentially with the distance from the surface of solid-liquid interface.⁷ Finally, the outermost layer is the bulk solution, where coulomb force from interface has no significant impact on this layer. It should be noted that the boundary of diffusion layer and bulk solution is defined by the decay of electric field at the ratio of e⁻¹. The distance from solid-liquid interface to this boundary is called as Deybe length, namely the longest distance where charges of analytes can affect surface characteristic.

Due to the characteristics of ions in these layers, it is able to model these layers by equivalent capacitor model. A simplified circuit model is described as a series connected capacitors around solid-liquid interface as shown in Figure 1b. In the equivalent circuit, the Stern layer and diffuse layer can be modeled as serial capacitors of C_Helm and C_diff. C_Helm comes from previous Helmholtz EDL model, which stands for charge-fixed layer near the interface. C_diff is the capacitance of the diffuse layer. Two different descriptions for the equivalent circuit are raised: voltage source on C_diff^7,10 and tunable C_diff^11,12 due to the change in analytes. In the tunable capacitive model, the difference in analytes is modeled as ion re-distribution in diffusion layer. This model converts ion redistribution to the capacitance change accordingly. In the voltage source model, the value of C_diff is fixed while the electric potential applied on the capacitor is changed. It can be either positive or negative voltage source, depending on the charge types which carried by analytes.

With the voltage source model, it is clear to find out the sensitivity model with analytical method. According to the derivation from Shoorideh and Chui, the operation model of FET-based biosensor can be divided into three stages: 1) A concentration change in analytes causes the charge change near the interface of sensor, namely dσ; 2) The resulted charge change induces the difference in effective gate voltage, namely dV_g,eff; 3) The effective gate voltage difference results in drain current change (i.e. dI), which can be measured from device I-V characteristics. The equation is shown as follows.⁷

Ion sensitive field-effect transistors

Ion sensitive field-effect transistor (ISFET) is an EIS analogous of MOSFET. It is firstly invented in early 1970s to detect pH values in solution.¹³ Based on the previously mentioned theory, biosensing is applicable to the state-of-art of ISFET with complementary bio-probes anchored on the surface of ISFET. Modern ISFET technologies are based on CMOS design essentially. The standardized design platform results in significant progress in ISFET researches from 2000s. In this section, we aim to focus on different designs, CMOS platform and methods to improve ISFET performance. In addition, biomedical samples measured by ISFETs are also provided.

Conventional ISFET

The conventional ISFET is comprised of a MOSFET with the metal gate replaced by a dielectric layer as a sensing membrane. The sensing dielectric is normally consisted of silicon dioxide. The capacitance of the dielectric layer (Cox) determines the capability of overall sensitivity because it directly affects the quantity of induced charges per potential change.⁷ Similar to conventional MOSFET, at the same time, high-k materials are studied as a replacement of sensing membrane^14,26 due to their high capacitance and significantly dense material structure to enhance sensitivity and prevent drifting, respectively. Another configuration of CMOS ISFET is the extended-gate architecture. In this configuration, the gate dielectric is isolated from the solution but it uses gate metal extending from the transistor to the top passivation layer of the device. This feature reduces the steps of post-processing. For instance, ionic etching or chemical etching is required to open up sensing area for a conventional ISFET. However, the extended-gate configuration adds an additional serial capacitor of passivation (e.g. silicon nitride), which limits the sensor capability of overall sensitivity.

Dual-gate ISFET

The dual-gate ISFET (DG-ISFET) employs a back gate on the other side with respect to the active channel. (e.g. silicon substrate or thin-film semiconductor). In previous studies, it is found that the sensitivity of DG-ISFET can go beyond the Nernstian limitation of 59 mV/pH to ∼1 V/pH.^15–19 It is shown that an optimized sensitivity region by top-gate (TG; i.e. fluid-gate, FG) and bottom-gate (BG) matrices.¹⁸ The origin of high pH sensitivity is considered as the capacitance asymmetry between top and bottom gate oxide. In the experiment of sweeping V_BG (i.e. bottom-gate bias) with fixed V_FG (i.e. fluid-gate bias), the sensitivity is defined as the shift of threshold voltage due to pH changes (ΔV_{T, BG}/ΔpH). The conductance change in drain current (I_d) caused by V_BG (i.e. ΔI_{d, BG}) is used to compensate the conductance change caused by pH or concentration differences in analytes on top of the sensor surface (i.e. ΔI_{d, TG}). The relationship between top-gate and bottom-gate is explained in Equation 2,¹⁹ where the ratio of C_tox and C_box determines the degree of sensitivity enhancement. The term α_SN is considered as a constant in the operation region.¹⁹

The DG-ISFET has been realized in CMOS by Huang et al.²⁰ The DG-ISFET device is fabricated on a SOI substrate with high-k material as the sensing dielectric. A significant improvement is reported in signal-to-noise ratio (SNR), drift rate as well as sensitivity. Figure 2 depicts the basic configuration between conventional ISFET and DG-ISFET.

**Figure 2.** The configuration comparison between (a) Conventional ISFET and (b) Double-gate ISFET.

ISFET Bio-applications

Numerous publications have been reported on ISFET biosensing applications. It is used to detected protein concentration,^21–25 DNA detection,^26–29 genetic diagnosis, DNA amplification test³⁰ and cell physiology research.^31,32 Because of the maturity of analog front-end circuit design, it is easy to achieved ISFET array for multiplex sensing and multi-functional sensing. For instance, Maruyama et al.³³ developed a CMOS ISFET array with the active pixel sensor architecture based on 1P1M CMOS process for multiplexing DNA sensing. However, most of the reports did not provide complete sensor characterization. It is difficult to execute a performance comparison within these sensor devices. Recently, the applications of ISFET have been transformed from molecular detection to other emerging applications, such as DNA sequencing.³⁴ There is a significant breakthrough of DNA sequencing by semiconductor technologies. We will discuss it in a later content.

Silicon nanowire biosensors

An ultra-sensitive solid-state biosensor, silicon nanowire (SiNW) is first introduced by Cui et al.³⁵ SiNW has been reported demonstrating the ability of detecting nucleic acid,^36–40 proteins^41–44 and virus⁴⁵ with supreme limit of detection (i.e. LoD). A typical SiNW biosensor consists of a wire-like structure as a three dimensional channel based on crystalline or poly-silicon. The source and drain electrode connect two side of the wire as illustrated in Figure 3.⁴⁹ There are two major types of SiNW sensors: the resistor type as mentioned previously; and the transistor type has an addition bottom-gate for the sensor to perform transistor characteristics.

**Figure 3.** (a) An illustration of silicon nanowire; a wire-like channel connects to source and drain electrodes. (b) The SEM image of a silicon nanowire.⁴⁹

Fabrication process

There are two categories of methods, bottom-up and top-down, to fabricate SiNW-based sensing devices. Bottom-up method can be referred to Cui et al.³⁵ in 2001. The SiNW was grown on the substrate in a chemical vapor deposition (CVD) system via vapor-liquid-solid (VLS) method. Then, the source (S) and drain (D) metals were deposited for connection. On the contrary, top-down fabrication method⁴⁶ forms SiNW by lithography process. For instance, E-beam lithography can be used on a silicon-on-insulator (SOI) substrate for nano-patterning and reaction ion etching (RIE). Then a three-dimensional wire structure can be established. In the three-dimensional wire structure, the sensing region is defined by light doping and S/D contact region defined by heavy doping via ion implementation. With these two fabrication methods, the top-down method becomes popular recently because of its high compatibility to standard CMOS process, which has the potential to be integrated with interface circuit and possibility to be mass produced and cost effective.

Working principle of SiNW

With a relative high doping concentration, the SiNW behaves as a resistor. When the number of charged analyte changes on the sensor surface, the conductance of the resistor changes accordingly. Following previously discussed sensing mechanisms, the charges from analytes affect the conductivity of the nanowire channel due to attractive or repulse electrostatic force. In other words, the degree of depletion within channel changes due to charges from analytes in the solution. An illustration of SiNW's working principle is shown in Figure 4. It is worth to mention that when the resistance decreases (i.e. higher doping concentration), the sensitivity decreases.⁴⁷ Moreover, the unique wire-like geometry provides a very large surface-to-volume ratio compared with ISFET as a planar structure. The effect of high surface-to-volume ratio and geometric effect are shown in.⁴⁸ With the decrease of the thickness, SiNW shows more threshold voltage shifting and conductance change, namely higher sensitivity. Due to the special characteristics, SiNW is considered as a candidate for ultra-sensitive biosensing application.

**Figure 4.** The working principle of n-type silicon nanowire biosensors. (a) Positive charges accumulate on the surface. The electrostatic attraction force to electron carriers results in higher conductance. (b) The original state of SiNW. (c) Negative charges accumulate on the surface. The electrostatic repulsion force to electron carriers results in lower conductance.

The compatibility of CMOS process and its applications

Top-down SiNW demonstrate high potential to be fabricated by CMOS technology. Stern et al.⁴⁹ delivered the result of CMOS compatible SiNW biosensor in 2007, with photolithography and RIE. A CD3 ligand antibody was crosslinked on the SiNW to demonstrate an impressive result, below 100 femtomolar (fM). Another innovative method to fabricate SiNW by CMOS technology used sidewall spacer as the nanowire.^50,51 A poly-silicon gate was formed capped with silicon oxide, followed by poly-silicon deposition and reactive plasma etching. It is interesting to note that the remain part of poly-silicon forms SiNW on each side of the gate by the directional plasma etching. Utilizing this sidewall-based SiNW, a detection of avian influenza DNA with ∼100 fM detection limit was demonstrated.

Huang et al. demonstrated a fully integrated CMOS SiNW biosensor which is fabricated by TSMC 0.35μm 2P4M standard process.⁵² This is the first time a SiNW biosensor realized by a commercialized foundry process. It shows an implemented case that SiNW biosensor inherits CMOS advantages, such as cost effectiveness, mass production, and system-on-chip (SoC) capabilities. The on-chip analog front-end circuit (e.g. amplifier and analog-to-digital converter) reduces the system noise compared with a discrete circuit system. Although the SiNW sensor can be designed via the standard platform, a post etching process is still required to open up the sensing window for further enhance the sensitivity of SiNW sensor. This CMOS SiNW has been used to detect hepatitis B virus DNA,⁵³ cardiac troponin I,⁵⁴ and NT-pro BNP⁵⁵ in clinical sample successfully. It encourages technologies moving forward from ISFET to SiNW in recent years.

Organic FET and grapahene FET biosensor

Organic field-effect transistor (OFET) is an analogous of thin-film transistor (TFT) using organic semiconductor (OSC) as an active material. It can be either p-type or n-type transistor depending on material selection. Different from MOSFET, OFET belongs to TFT family, which operates under accumulation mode.⁵⁶ That means the applied gate voltage controls the channel current in a direct manner. The mobility of OFET is typically 10⁻¹∼10⁻² cm²Vs⁻¹, which is much less than the mobility of crystalline silicon. This difference originates from OSC's material composition, i.e. π bonds and non-covalent bonds are the main carrier transition pathways and result in worse mobility. Despite this drawback, OSC is highly compatible to flexible substrate. The possibility of realizing flexible and wearable electronic devices draws scientists' attentions on OFET in recent years.

Similar to ISFET, OFET is able to be switch from conventional transistor to biosensing transistor. By immersing the OFET in an electrolyte environment, the EDL structure appears immediately. Thus, it is able to control the device by adding a reference electrode as the solution gate electrode. A configuration of OFET biosensor (i.e. electrolyte gate OFET, or EGOFET) is shown in Figure 5a. Based on previous research results, it is reported that 100 nM of glucose by red-ox reaction,⁵⁷ DNA^58–61 and down to pM regime of protein^62–66 can be effectively detected by EGOFET. In addition, EGOFET fabricated on flexible substrate is also demonstrated.⁶⁷ It creates opportunities for the nano-electronic biosensing devices on wearable devices.

**Figure 5.** The illustration of (a) Organic FET biosensor and (b) Graphene FET biosensor.

Graphene FET biosensor

Graphene is one of the most attractive materials since mid-2000s.⁶⁸ Due to its special material properties, such as high electron/hole mobility, transparency, and mechanical strength, etc., graphene is considered as a material for next generation semiconductor devices. Toward a semiconductor device, Graphene FET (GFET) is realized in a TFT structure, shown in Figure 5b. From band structure standpoint, graphene is a zero bandgap material. It can easily generate electrons and holes with positive and negative electric field respectively (i.e. electrical doping). This phenomenon is so-called ambipolar characteristic, in which both positive and negative gate voltage generate drain current. The V_G at the minimum I_ds is called charge-neutral-point (CNP), where the graphene is intrinsic (i.e. non-doped). To operate a GFET, it is normally designed in single (top/bottom) or double-gate configuration.⁶⁹ In a conventional bottom-gate GFET, the active channel is control by the potential from the back side via the deposited gate dielectric. On the other hand, the EDL appears to serve as the top-gate dielectric as a bottom-gate GFET immersed in a solution environment. A voltage change from solution via top-gate is able to generate the ambipolar behavior, too. This creates an environment for researchers to observe the surface potential change due to the disturbance of analytes and proves that GFET can be serves as a biosensing device.

Ohno et al.⁷⁰ introduces an electrolyte-gated GFET biosensor for pH sensing and protein adsorption detection. It is found that the conductance in electrolyte environment is 30 times higher than vacuum. The conductance exhibits a linear relationship when pH increases. Moreover, protein adsorption experiment demonstrates the detection limit to be in the level of picomolar (pM). It indicates the potential of ultra-sensitive biosensing applications. An aptamer-modified GFET is also demonstrated to detect IgE with 47 nM detection limit.⁷¹ In addition, DNA,^72–76 protein specific recognition,^77–79 glucose⁸⁰ and living cell⁸¹ have been shown to prove the capability of GFET biosensor. Moreover, Zheng et al.⁸² proves that chemical vapor deposition (CVD)-grown graphene can be used in the fabrication of GFET biosensor. Compared to exfoliation method, CVD-growth method increases the film quality for large area graphene. This progress took a big step forward for the novel material finally being applied into practical use.

Figure 6 summarizes the critical performance of FET-based biosensors reported in recent years. SiNWs, OFET, and GFET are included in this figure. SiNWs and GFETs achieve femtomolar (fM) limit of detection (LoD) and higher dynamic range. It shows a good potential for ultrasensitive applications. The OFET, on the other hand, has limited LoD and dynamic range due to the material properties. However, the possibilities to integrate with flexible electronics are the biggest strength and value of OFET.

Novel field-effect transistors as biosensor

In addition to the aforementioned FET-based biosensors, new devices emerge to be the potential next-generation biosensors. One category among the various is electrically tunable field-effect devices. Gao et al.⁸³ brings up a novel SiNW tunneling field-effect transistor biosensor detecting CYFRA21-1 protein detection. The use of opposite doping of using p⁺ and n⁺ doping in two side of nanowire creates opportunities for both p-type and n-type channel happens in the same device (i.e. ambipolar behavior). This special design reduces sensor noise and the detection limit improves accordingly. Another strategy to achieve better sensitivity and dynamic range by using multi-gate field-effect design is performed.^84,85 In addition to these structural enhancements, compound semiconductors with high mobility are also raised as emerging FET-based biosensors.⁸⁶ Utilizing different device processes and high mobility characteristics, for example, AlGaN/GaN based biosensors can be used to detect protein-peptide binding affinity⁸⁷ and cardiac biomarkers.⁸⁸ It is interesting to see how to leverage all degree-of-freedom of the geometry and material to increase the performance of devices. This indicates the wider range of analytes' concentration can be detected.

Detection principle of biomolecules and surface modification

In order to detect an existence of specific biomolecules in a target sample, a functionalization process is necessary to be conducted to the device surface. As previously mentioned, the electric double-layer determines the minimum sensing distance—the Debye length. Thus, the target biomolecules must be crosslinked to the surface of the FET-based biosensor in order to induce larger signal.

Since the selectivity of the biomolecule recognition is a major concern of a biosensor development, the functionalization process is always designed to be specific to the target biomarkers. Therefore, scientists leverage advantages of bio-affinity recognition, such as complementary DNA pair, antibody/antigen interaction, and enzyme/substrate interaction, to increase the selectivity of biosensors. The abovementioned methods are able to cover most of the prevalent diseases, such as cardiovascular diseases (protein biomarkers),^54,93,94,98 cancers (protein biomarkers),^{102,104,107,110,113} diabetes (protein and enzyme biomarkers)^118,119 and infectious diseases (nucleotide biomarkers).^53,121,122

Nevertheless, the functionalization process required more complex process than immobilizing the probes only. It is because the functional groups of biomolecules need cross-linkers to form covalent bonds with the sensing film (dielectric layer) of FET-based biosensors. The most commonly used cross-linking process for ISFET and SiNW (i.e. the oxide based sensing dielectric) is the APTES-GA method. APTES is a silane molecule, which is able to bond with sensor dielectric surface, such as SiO₂. This self-assembled organic monolayer, APTES, is formed on the sensor surface and provides a good platform for the second stage linkers by its amine group in the other end. The second stage cross-linker is glutaraldehyde, which is a bifunctional reagent connecting the APTES and bio-probe by imide bonds. There are many other different protocols of surface modification to link different type of sensors. For instance, different materials (e.g. MPTES, GPTES) and electric field assisted modifications.⁸⁹ There is a good review to have an overview of these surface modifications.⁹⁰

Organic FET and graphene FET implement different approaches to conjugate bio-probes. Due to the interface properties, organic semiconductor materials need further treatment on their surfaces to establish functional groups. The plasma-enhanced chemical vapor deposition (PE-CVD) and O₂ plasma have been reported to create –COOH and –OH, respectively.^56,58 The EDC/NHS linker pair is used for –COOH to link peptide nucleic acid (PNA) for DNA sensing and biotin for protein sensing.⁶¹ In addition, the thiol/gold linker pair is used for –OH to link gold nanoparticles, which is further bond to DNA or PNA probes for DNA sensing.⁶⁶ Other novel methods have been proposed, for example, an UV-assisted method is reported to simplify the functionalization process.⁶⁴ Some of the researches apply different configurations to make it possible to functionalize the sensor on the surface of gold electrode.^60,62,63 The modification of graphene is much different from the abovementioned methods. Since it does not exist functional group on the graphene surface, the π—π bond interaction is applied to strengthen the binding force. From many researches, pyrene^71,72 is used as the first stage linker to further link with PNA,⁷² antibodies⁷⁷ and aptamers.^71,78 In addition, gold nanoparticle is another method for graphene surface modification.^75,79

Applications in Various Diseases and DNA Sequencing

With influences of rapid growth of human population, the incidence of many civilized diseases are obviously raising. Therefore, early detection of diseases has become extremely important, particularly in further achieving conveniently personalized precision medical and providing better medical assistance in remote areas. Through various surface-modification methods for biomolecule identification on biosensors, the distinguishable capacity of specific biomarker recognition is being hailed as a milestone in achieving of immediate and accurate disease diagnosis. The following discussion includes some critical diseases and its related detected methods; especially focus on the FET-based biosensors.

Cardiovascular disease (CVDs)/acute myocardial infarction (AMI)

Cardiovascular diseases (CVDs) refer to the disorders of heart or blood vessels, also known as circulatory diseases. It is one of the most common causes of death and aggravated assaults on the health of humankind worldwide. According to statistics from the World Health Organization (WHO),⁹¹ an estimation of 17.7 million people died from CVDs in 2015, representing 31% of all global deaths. Of these deaths, around 7.4 million were due to coronary heart diseases and 6.7 million were due to strokes. Among all of the CVDs, a noteworthy disease, acute myocardial infarction (AMI), occurs when some myocardial blood circulation suddenly interrupted. As a consequence, myocardial damages due to lack of oxygen result in heart failure, arrhythmia, cardiogenic shock and asystole. Nowadays, cardiac troponin I (cTnI), detected through a blood test, is a highly recommended biomarker because it possess extremely predominant sensitivity and specificity for measuring injury to the heart muscle. Therefore, the corresponding cTnI concentration monitoring is particularly important to predict the occurrence of AMI. In recent researches, FET-based biosensors initiate revolutionary development for relevant diagnosing, great enhancing the convenience and realizing the possibility of real-time monitoring at the point of care. The following is a brief introduction of several important developments.

Yen et al.⁵⁴ demonstrated a fully integrated bottom-gate poly-silicon nanowire biosensor system-on-chip (BG-bioSSoC) for cTnI detection in 50% fetal bovin serum (FBS) serum samples. This developed device is implemented by a 0.35 μm 2P4M CMOS commercial-available standard fabrication process. The detection limit of sensor for cTnI is 7.6 pg/ml. This SoC design shows feasibilities for biomolecular analyte detection in serum samples. This demonstrates a potential POCT device for fast clinical diagnosis applications. Kim et al.¹⁰⁰ demonstrated a label-free detection of cTnI with superior sensitivity and high specificity, using silicon nanowire field-effect transistors. A consist of lightly doped honeycomb nanowires and embedded Ag/AgCl pseudo-reference electrode were utilized to provide electrical improvement and increased sensing area. The detection limit of sensor is as low as 5 pg/mL, which is 8 times smaller than the suggested diagnostic threshold cutoff of AMI. Moreover, the fabricated devices demonstrate a good selectivity in cTnI tests. Tuteja et al.⁹⁶ reported a one-step microwave-assisted unscrolling of carbon nanotubes to form functionalized rebar graphene (f-RG) which well characterized on an interdigitated electrode biochip in a FET configuration. The Ab-f-RG electrodes with varying concentrations (1 to 1000 pg/mL) of cTnI prepared in phosphate buffer (20 mM, pH 7.5) were realized. The sensitivity of sensor is 1 pg/mL, exhibiting great promise for its applicability in diagnostics and specificity toward cardiac marker (cTnI). Liu et al.⁹⁵ demonstrated a scalable and facile lithography-free method for fabricating highly uniform and sensitive In₂O₃ nanoribbon biosensor arrays. This approach can detect cTnI for concentrations down to 1 pg/mL, which is much lower than clinically relevant cutoff concentrations (∼40 pg/mL). Moreover, the reusability of the In₂O₃ nanoribbon biosensors had demonstrated with very small variation by applying a regeneration buffer to the used sensor surfaces.

As shown in Table I, it summarizes the limit of detection, linear range, and sensitivity for cTnI biosensors in literatures. The FET-based biosensors show comparable capabilities with other kinds of biosensor types. Different from the others, FET-based biosensors could be integrated into current semiconductor manufacturing process, achieving the strategy of stable production. The portable size also has potential to realize real-time monitoring anytime, anywhere. However, the improvement of immobilization protocol will be a big challenge to reduce the detection limit, including new sensing film development and novel bonding structure approach.

Table I. Comparison of the limit of detection (LoD) for cTnI biosensors.

Types Materials LoD Linear range Sensitivity Ref.

OMC sensor	OMC chip	2 fg/mL	2–10 fg/mL	91777.9 nm/RIU	Zhou et al. (2018)⁹²
Electrochemiluminescent (ECL)	L-Cys @ Ru(dcbpy)₃ ²⁺	0.08 pg/mL	0.25–100 pg/mL	-	Zhou et al. (2014)⁹³
Electrochemical (EC)	ZnO	1 pg/mL	-	-	Shanmugam et al. (2017)⁹⁴
FET-based	In₂O₃ nanoribbon	1 pg/mL	-	-	Liu et al. (2016a)⁹⁵
FET-based	Functionalized rebar graphene (f-RG)	1 pg/mL	10–1000 pg/mL	-	Tuteja et al. (2014)⁹⁶
Electrochemical impedance spectroscopy (EIS)	Au/MHA/TMB/Den	0.28 pg/mL	1–10⁶ pg/mL	-	Akter et al. (2017)⁹⁷
Photoelectrochemical (PEC)	NAC-CdAgTe QDs	1.756 pg/mL	5–20000 pg/mL	-	Tan et al. (2017)⁹⁸
Fluorescence	Fluorophore-coupled streptavidin (FL-SA)	2 pg/mL	10–10000 pg/mL	-	Seo et al. (2016)⁹⁹
FET-based	Si nanowire	5 pg/mL	5–200 pg/mL	-	Kim et al. (2016)¹⁰⁰
FET-based	Si nanowire SoC	7.6 pg/ml	-	-	Yen et al. (2014)⁵⁴
Optical point-of-care immunoassay	UV LED chip / PoC immunoassay	0.22 ng/mL	-	-	Rodenko et al. (2017)¹⁰¹
Surface plasmon resonance (SPR)	HGNPs	1.25 ng/mL	-	-	Wu et al. (2017)¹⁰²

Cancer

Cancer, also known as malignant tumor, refers to the abnormal proliferation of cells. These hyperplastic cells might have potential to invade or spread to other parts of the body. There are over 100 types of cancer known in humans to date. According to statistics from the WHO,¹⁰³ it is the second leading cause of death globally, causing 8.8 million deaths in 2015 (15% of all global deaths). Nearly 1 out of 6 deaths comes from cancer globally. Unfortunately, an estimated number of new cases in the next two decades will increase by 70%. The most common causes of cancer death are comprised of lung, liver, colorectal, stomach, and breast in sequence. In the previous literatures, FET-based biosensors make a lot of effort in this area. By tracking cancer biomarkers, it helps to identify early cancers, forecasting how critical the situation and its respond to treatment. The following is a brief introduction of several label-free detection methods for tumor markers by FET-based biosensors.

Cheng et al.¹⁰⁴ demonstrated the label-free quantitative detection of cytokeratin fragment 21-1 (CYFRA 21-1) and neuron-specific enolase (NSE), two lung cancer tumor markers, in both phosphate-buffered saline (PBS) (pH 7.4) and human serum using FET biosensors. The detection limits for CYFRA 21-1 and NSE using the single- analytic FET biosensors were 1 and 10 ng/mL, respectively. Additionally, the multi-analytic FET biosensor by integrating both anti-body types on the same chip, providing a big step forward the realization of sensor arrays. Hideshima et al. demonstrated a feasible strategy to realize the antibody-modified FET for quantitative measurement of liver tumor marker human alpha-fetoprotein (AFP) in the blood serum with bovine serum albumin (BSA) blocking treatment.¹⁰⁶ In order to establish a suitable testing environment, the evaluation of optimal isoelectric points (around 5) was executed for FET detection under physiological pH. Moreover, for the purpose of effectively minimization of nonspecific adsorption with blood proteins, the antibody-modified surface was covered with 0.1% blocking agent (BSA). Finally, the detection limits for AFP in blood serum reach an adequate level (10 ng/mL) for clinical diagnosis. Bao et al.¹⁰⁹ reported a top–down approach on a silicon-on-insulator (SOI) substrate to fulfill the novel silicon nano-ribbon (Si-NR) FET, which is comparable to the massive fabrication CMOS technology, representing excellent performance of carcinoembryonic antigen (CEA) detection. The Si-NR FET is 120 nm in width and 25 nm in height, with ambipolar electrical characteristics. Moreover, the flowing channel for the CEA solution is provided by microfluidics, integrated on the top of the biosensor. The sensitivity of device is 10 pg/mL and it shows ideal repeatability by 60 devices confirmed and promising for colorectal tumor early diagnosis.

Cancer antigen 125 (CA-125) is the most frequently used for ovarian cancer detection. Majd et al. developed a novel and ultrasensitive aptasensor by the combined of carboxylated multi-walled carbon nanotubes (MWCNTs) and reduced graphene oxide-based field effect transistor (rGO-FET) onto flexible PMMA substrate.¹¹² It exhibited an extremely low detection limit for CA-125 of 5.0 × 10⁻¹⁰U/mL, which successfully applied in real human serum samples with outstanding selectivity in solutions containing other proteins. In addition, it is worth mentioning that the device is also provided with superior mechanical flexibility, which might has the potential to achieve the goal of wearable.

Apolipoprotein A-II (APOA2) protein is recently identified as a significant biomarker that could reflect the increase of occurring probability on bladder cancer. Chen et al. developed a label-free electrical biochip based on an integration of magnetic graphene with long-chain acid groups, anti-APOA2 antibody (Ab-MGLA).¹¹⁵ The developed biochip combined with n-type polycrystalline silicon nanowire field-effect transistor (poly-SiNW-FET). The device exhibited a linear dependence of biomarker concentration in a range between 19.5 pg/mL and 1.95 μg/mL, with a limit of detection (LoD) of 6.7 pg/mL. Moreover, the authors successfully analyzed the urine samples from bladder cancer patients, demonstrating the clinically diagnostic feasibility of its proposed biosensor.

Nowadays, precision medicine is an emerging approach for cancer treatment. In order to promote the well-being of all humankind, an efficiently specific biomarker identification plays a big role in the implement of rapid screening and instant monitoring. FET-based biosensors could offer these essential functions to judge the early state of cancer conveniently. According to Table II, the related information of limit of detection, linear range, and sensitivity among various cancer types are listed for reference. However, in comparison to non-FET-based methods, the limit of recognized capability still has a room for improvement. The enhanced strategy could move toward novel material replacement, surface binding modification, and device structure optimization etc. Therefore, new biomarker detection among various cancer types on FET-based will also be an important direction in near future.

Table II. List of the limit of detection (LoD) for various cancer types.

Types Materials LoD Linear range Sensitivity Ref.

Lung cancer / CYFRA 21-1 (1) / NSE (2)	FET-based	(1) 1 ng/mL (2) 10 ng/mL	-	-	Cheng et al. (2015)¹⁰⁴
Lung cancer / CYFRA 21-1	ZrO₂-rGO (reduced graphene oxide) / Differential pulse voltammetry (DPV)	0.122 ng/mL	2–22 ng/mL	0.756 μA·mL/ng	Kumar et al. (2016)¹⁰⁵
Liver cancer / AFP	FET-based	10 ng/mL	-	-	Hideshima et al. (2012)¹⁰⁶
Liver cancer / AFP	Carbon nanohorns (CNHs) / Electrochemical (EC)	0.33 pg/mL	0.001–60 ng/mL	-	Yang et al. (2014)¹⁰⁷
Liver cancer / AFP	CuS & graphene oxide (GO) / EC	0.5 pg/mL	0.001–10 ng/mL	-	Li et al. (2015)¹⁰⁸
Colorectal cancer / CEA	Si-NR FET-based	10 pg/mL	0.01–10 ng/mL	-	Bao et al. (2017)¹⁰⁹
Colorectal cancer / CEA	PEDOT:PSS & carbon nanotubes (CNTs) / EC	1 ng/mL	2–15 ng/mL	7.8μA·mL/ng·cm²	Kumar et al. (2015)¹¹⁰
Colorectal cancer / CEA	rGO-Thionine-AuNPs / EC	0.65 pg/mL	0.01–300 ng/mL	-	Jia et al. (2014)¹¹¹
Ovarian cancer / CA-125	rGO FET-based	5×10^-10U/mL	1 × 10^-9–1 U/mL	-	Majd et al. (2018)¹¹²
Ovarian cancer / CA-125	Carboxylated g-C₃N₄ / ECL	0.0004 U/mL	0.001–5 U/mL	-	Wu et al. (2016)¹¹³
Ovarian cancer / CA-125	rGO / microfluidic paper-based EC	1 pg/mL	20–70 ng/mL	-	Wu et al. (2013)¹¹⁴
Bladder cancer / APOA2	Si nanowire FET-based	6.7 pg/mL	19.5–1.95×10⁶ pg/mL	-	Chen et al. (2015)¹¹⁵
Prostate, Ovarian and Bladder cancer / hCG	GO / EIS	0.286 pg/mL	0.001–50 ng/mL	-	Teixeira et al. (2014)¹¹⁶
Prostate cancer / PSA	Horseradish peroxidase (HRP) & Multiwall CNTs (MWCNTs) / EC	0.4 pg/mL	1–10000 pg/mL	2.5 ng/Hz	Akter et al. (2012)¹¹⁷

Diabetes

Diabetes, commonly known as diabetes mellitus (DM), is a metabolic disease characterized by long-term blood glucose levels higher than the standard value. There are two main causes of diabetes: the pancreas could not produce enough insulin, or the cells are not insulin-sensitive. According to statistics from the WHO,¹¹⁸ an estimation of 1.6 million deaths were directly caused by diabetes in 2015. Moreover, WHO projects that diabetes will be the seventh leading cause of death in 2030. To diagnose diabetes, glycated hemoglobin (HbA1c) is an efficacious biomarker for long-term monitoring without affected by short-term variations. The level of HbA1c distributes in a certain range from 4% to 6% is regarded as normal in clinical diagnosis, which is measured as the percentage of total hemoglobin (Hb).¹¹⁹ In previous researches, FET-based biosensor also contributes to efforts in this field; it is important for such chronic disease monitoring at home. Some cases will be introduced in following description.

Bian et al. developed a FET-based immunesensor that consists of a FET-based sensor by standard CMOS process and a disposable extended-gate electrode chip by Micro-Electro-Mechanical-Systems (MEMS) technique.¹²⁰ The experimental results showed that it achieved a linear response to HbA1c with the concentration from 4 to 24 μg/ml, and a linear response to Hb with the concentration from 60 to 180 μg/ml. It worth mentioning that one attractive achievement about "direct detection" is demonstrated by Tanaka et al.¹²¹ recently. The direct measurement method is based on sandwich immunoassay that combines anti-Hb and enzyme-labeled anti-HbA1c antibodies. After a series of attempts, it was found that the process of denaturing HbA1c with 0.2% dodecyltrimethylammonium bromide was added to enable it to be accessible to anti-HbA1c antibodies and successfully enhance the sensitivity. Finally, the results showed a good relevance between the certified and experimental values of HbA1c (%) in the clinically range from 5.6% to 10.6%.

AIDS/HIV

AIDS, originally named as "acquired immune deficiency syndrome," is derived from a retrovirus, human immunodeficiency virus (HIV) infection, leading to the destruction of the immune system, and then gradually become the target of many opportunistic disease. Then it finally contributes to a variety of clinical symptoms. According to statistics from the WHO,¹²² 1.0 million people died from HIV-related causes globally in 2016. It is still a global issue under the spotlight.

Ruslinda et al. demonstrated a diamond field-effect transistor (FET) using ribonucleic acid (RNA) aptamers as a sensing element on a solid surface.¹²³ The results show a 91.6 mV potential change in the gate, whereby exist 8 μA shift in the negative direction at a source-drain current in the presence of HIV-1 Tat protein bound to the RNA aptamers. Furthermore, it is the first time to demonstrate reliable use of a real sample of HIV-1 Tat protein by an aptamer-FET, showing outstanding approach to develop relatedly clinical biosensor applications via diamond bio-interfaces. Kwon et al.¹²⁴ reported a large-scale FET-type graphene micropatten (GM) nano-biohybrid-based immunosensor (GMNS) with close-packed carboxylated polypyrrole nanoparticle (CPPyNP) arrays and integration it in a flexible and fluidic system, resulting in a controllable method to detect HIV-2 antibody (Ab). The minimum detectable level for the GMNS to HIV-2 Ab was as low as 1 pM, which was 10 times higher than that (10 pM) of Ag-GM. This is the first experimental demonstration of a large-scale FET-type GMNS with flexible fluidics.

In previous studies, there are plenty of achievements on electrochemical method rather than FET-based biosensors. Despite of this, FET-based sensors potentially gain some benefits including low-cost, high sensitivity, and rapid screening which could bring more excellent consequences to slow down the spread of AIDS.

DNA sequencing

DNA sequencing stands for identifying the nucleotide sequence (i.e. A, T, C, and G) of an unknown DNA strand. Sanger sequencing was first developed with fluorescence labeled nucleotides.¹²⁵ However, it was a time consuming and expensive process till next generation sequencing (NGS) is introduced. NGS, for instance, pyrosequencing by Roche, SOLiD by ABI, and Illumina sequencing provides high throughput and cost effective solutions.¹²⁶ Consequently, it is available for the whole human genome to be decoded in economic approaches. In other words, NGS paves the way of genomic research, genetic disorder, genetic disease and personalized medicine researches. Besides, some of the emerging NGS technologies take the advantages of Moore's Law. The modern CMOS technologies accelerate the development of DNA sequencing to a new horizon due to the huge improvement of transistor numbers and performance in recent years. For instance, ISFET based chemical sequencing^{34,127,129–131} and nanopore-based electrical sequencing^128,129 are considered having potential to be commercialized. It is worth to mention that another advantage of CMOS-based sequencing is the cost. S. Goodwin et al. indicate both the cost of instruments and cost per gigabase (Gb) are below USD $100 for the commercial ISFET sequencer - Ion Proton.¹²⁷ This strength makes ISFET sequencer strongly competitive in the market. In this session, we will discuss the state-of-the-art of ISFET sequencing and review current achievement of highly integrated CMOS ISFET chip for DNA sequencing.

State-of-the-Art: ISFET sequencing

The process of ISFET sequencing could be defined by three steps:^34,130 Firstly, pre-processing cutting DNA into small strands of oligonucleotide. Amplifying the strands and connecting to microbeads. Secondly, sample loading: distributing the microbeads into a microwell array chip, in which an ISFET is on the bottom of each individual well. Finally, reaction and detection: adding each dNTP (i.e. dATP, dTTP, dCTP, dGTP) periodically, DNA polymerization taken place and the chemical change is detected by ISFET. In the last step, DNA polymerases are required in the reagent to generate a complementary strand of the target DNA strand. dNTP works as an indicator added periodically, namely A→T→C→G→A etc. In the DNA polymerization, an orthophosphate and a hydrogen ion are released as by-products in turn resulting in a minor pH change (e.g. 0.02 pH unit per H⁺). As previously mentioned, ISFET is able to detect pH change in solution. As a pH-meter, ISFET is able to monitor the reaction of DNA strand synthesis consequently. With each dNTP adding sequentially, it is easily to realize the sequence of the complementary DNA and in turn figure out the original DNA sequence. Figure 7 depicts the ISFET sequencing flow. Toumazou et al. further discuss the effect of ISFET as a DNA amplification sensor.¹³¹ In this report, ISFET works as a pH-based qPCR (quantitative polymerase chain reaction) and loop-mediated isotherm amplification (LAMP) sensor by real-time monitoring the copy number of each PCR cycle. The results well fit the trend and characteristic curve of fluorescence qPCR and electrophoresis, as shown in Figure 8. It indicates that ISFET could become a strong candidate for DNA synthesis and amplification.

**Figure 7.** An illustration of the working principle of ISFET sequencer.³⁴

**Figure 8.** The performance of ISFET DNA amplification sensor fit the results of (a) electrophoresis and (b) qPCR.¹²⁹

CMOS ISFET sequencers

Researches by standard CMOS technologies have shown extremely potential to become future mainstream DNA sequencers. The highly integrated chips with peripheral circuit achieve high throughput and automation processing. Ion Torrent (now a Thermo-Fisher company) developed an ISFET DNA sequencer in 2011.³⁴ Since it is a commercialized product, a complete characterization is reported.¹³² As we expected, the array size of the sequencer chip significantly increases from 1.2M to 660M when the CMOS process platform switching from 0.35μm to 0.11μm. The throughput also increases from 100 megabases (Mbs) to 1500 Mbs with 99% accuracy accordingly. The read length also increases from 500k bases to 3.5Mbs with advanced process technology, which meets the prediction of Moore's Law. However, the percentage of filled well can be further improved to increase throughput as well as decrease noise. Consequently, Huang et al. demonstrate a dual-modal DNA sequencer chip of both ISFET and photodiode array.¹³³ This dual-modal design prevents the false pH signal from empty wells to be detected by visualizing the existence of microbeads, which are recognized by the integrated CMOS contact imager. Another report discusses the impact of different process node to ISFET performance for DNA sequencing.¹³⁴ 0.35μm, 0.18μm, 0.13μm and 90nm are compared in terms of parasitic capacitance and noise, which are important factors to the signal-to-noise ratio (SNR) of ISFET sequencing.

The area of ISFET sequencing is still growing. Due to the cost advantage, highly parallel processing and potential for personalized application, the ISFET sequencer have become a strong competitor to current optical-based sequencers. These strengths make a brilliant future for the field of ISFET sequencing.

Challenges and outlook of FET-based biosensing

With the epochal evolution of FET-based biosensors, high-throughput biomolecule recognition technology could be realized successfully. It benefits clinical diagnosis but also exists some limitations in applications. Among different kinds of biomarkers, issues of size and concentration of molecules will affect the response limit due to the saturation of effective detection area. Owing to the development trend of nano-technology approaches, biomarkers with smaller size and lower concentration are suitable for FET-based biosensor compared with others detection methods. Moreover, the sensitivity of FET-based biosensor is also a key parameter need to be further enhanced. The unsolved barriers of "Debye length issue" remains an entrenched obstacle. In some types of FET-based biosensor, ion trapping in dielectric membranes will cause drifting characteristic, which drastically decreases the accuracy of detection. To solve this, good resistance to ion aggressiveness will be an important development strategy for material selection. Combined with the challenges in surface modification, such as stability of binding rate maintenance and uniformity of responders, a reproducibility of measurement will be an inevitable ordeal to evaluate that it is reliable enough or not. By the way, due to the operation mechanism of FET-based biosensor, it tends to analyze charged biomolecules. In other words, electroneutral biomolecules relatively suitable for others detection methods, such as fluorescence and electrochemical impedance spectroscopy.

In next generation biomolecular diagnosis, massive and parallel biomedical signal processing will be an emerging path. Some requirements such as biocompatible and wearable turn into focus. To address these factors, the promising candidates of biosensors should be easy to integrate with signal processing systems, even need to implement the system on one single chip. Therefore, in addition to the difficulties mentioned above, impedance matching between sub-systems will become a decisive influence that determine whether the desired signals can be captured with high signal-to-noise ratio or not. Moreover, abilities to control biocompatible and flexible materials also require to be polished. It might be a good idea to integrate heterogeneous materials and structures to bring out new proposal, which could obtain many advantages such as high sensitivity, flexible, and portable simultaneously.

Conclusions

Field-effect transistors are presented as label-free, sensitive and selective electronic biosensors. In this review, the operating mechanism of various types biosensing devices are discussed in details, highlighting several significant achievements on bio-applications. Moreover, compatible CMOS process provides readily available for mass productions and potentials with splendid strategies of analog circuit combination. These features raise the possibility of personalized precision medicine anytime, anywhere. In order to alleviate those seriously epidemic diseases around the world, such as CVDs, cancer, and HIV, the universal of FET-based biosensors would contribute huge help for rapid screening to prevent the spread of pestilence. A series of comparison of the limit of detection (LoD) among various recognition methods (FET-based/non-FET-based) are also summarized in this review. Although it provides tremendous advantages, the capability of FET-based biosensors still has a room for improvement from many perspectives. In summary, FET-based biosensors are ideal for point of care testing (POCT) and would initiate diagnostic revolution for next generation development.

Acknowledgment

The authors would also like to thank Ministry of Science and Technology (MOST) in Taiwan (105-2221-E-002-232-MY3, 104-2628-E-002-014-MY3, and 105-2119-M-002-004) and National Taiwan University (NTU-107L900501) for their financial support.

ORCID

Yu-Cheng Syu 0000-0003-3225-9706

Wei-En Hsu 0000-0003-3240-5486

Chih-Ting Lin 0000-0002-4150-9693

Review—Field-Effect Transistor Biosensing: Devices and Clinical Applications

Article metrics

Author affiliations

Author notes

ORCID iDs

Dates

Abstract