Comparison between Field Effect Transistors and Bipolar Junction Transistors as Transducers in Electrochemical Sensors

Field effect transistors (FET) have been widely used as transducers in electrochemical sensors for over 40 years. In this report, a FET transducer is compared with the recently proposed bipolar junction transistor (BJT) transducer. Measurements are performed on two chloride electrochemical sensors that are identical in all details except for the transducer device type. Comparative measurements show that the transducer choice significantly impacts the electrochemical sensor characteristics. Signal to noise ratio is 20 to 2 times greater for the BJT sensor. Sensitivity is also enhanced: BJT sensing signal changes by 10 times per pCl, whereas the FET signal changes by 8 or less times. Also, sensor calibration curves are impacted by the transducer choice. Unlike a FET sensor, the calibration curve of the BJT sensor is independent of applied voltages. Hence, a BJT sensor can make quantitative sensing measurements with minimal calibration requirements, an important characteristic for mobile sensing applications. As a demonstration for mobile applications, these BJT sensors are further investigated by measuring chloride levels in artificial human sweat for potential cystic fibrosis diagnostic use. In summary, the BJT device is demonstrated to be a superior transducer in comparison to a FET in an electrochemical sensor.

The application of field effect transistors (FET) as transducers in electrochemical sensors was first described in 1970 by Bergveld 1 . Since then, field effect transistor (FET) based electrochemical sensors [2][3][4][5][6][7][8][9][10][11] have been extensively investigated due to their enhanced sensitivity, resolution, low power, portability and fabrication compatibility with silicon processing technology. Over the years, these electrochemical sensors have been demonstrated to detect both ions and biomolecules [2][3][4][5][6][7][8][9] , and have several sensing applications, including DNA sequencing 12,13 and mobile diagnostics 14,15 . Recently, bipolar junction transistor (BJT) device [16][17][18] was proposed as a transducer in an electrochemical sensor. In this study, we compare a FET transducer with the recently proposed BJT transducer. The objective is to evaluate the impact of the transducer type on the performance metrics of an electrochemical sensor.
An electrochemical sensor has two basic components as shown in Fig. 1: the sensing surface (or receptor) and the transducer. The sensing surface interacts with the target analyte and the transducer converts this interaction into a readable electronic signal. The sensor performance characteristics depend on both the components. The sensor selectivity and affinity towards the target analyte depends solely on the sensing surface because the analyte interacts only at the sensing surface as illustrated in Fig. 1. Other performance metrics such as sensitivity, resolution, and calibration depend on both components. Since the present objective is to compare the performance of a FET transducer with that of a BJT transducer in an electrochemical sensor, it is important to ensure that the sensing surface does not impact the evaluation. If the sensing surface and target analyte solution are kept the same, the sensing surface selectivity and target analyte affinity would therefore remain the same; in that case the observed differences in the electrochemical sensor performance metrics could then be attributed to transducers. In other words, electrochemical sensors that are identical in all details except for the transducer device type (FET versus BJT) must be used to ensure that the transducer comparison is accurate.
In the present study, electrochemical sensors with FET and BJT as transducers are compared. Both the sensors have the same chloride ion sensitive silver chloride sensing surface. Measurements show that the recently proposed BJT transducer is superior in comparison to the widely used FET transducer: both sensitivity and signal to noise ratio (SNR) are significantly higher for BJT sensors. More importantly, SNR, sensitivity and calibration curves are independent of measurement voltages for the BJT sensor. This implies that optimal measurements can be made over the entire sensing range with minimal calibration requirements which is a significant advantage, particularly for mobile diagnostic applications. In contrast, sensitivity, resolution and calibration curves are dependent on the applied gate voltage for a FET sensor. Consequently, for the FET sensor, detailed calibration curves are required to obtain accurate quantitative sensing results and also use voltages have to be carefully chosen to ensure maximum sensitivity and resolution. As a demonstration for mobile diagnostic applications, the BJT sensor is further investigated for diagnosing cystic fibrosis which requires accurate measurement of chloride ion in sweat [19][20][21] . Sensing measurements are shown to be repeatable and reversible in artificial human sweat. In summary, the BJT is shown to be a superior transducer option for an electrochemical sensor.

Results and Discussion
In this section, electrochemical sensors with FET and BJT devices as transducers are compared. These electrochemical sensors are identical except for the transducer device type to ensure that the transducer comparison is accurate. For the comparison purpose, we have chosen chloride ion sensitive silver chloride sensing surface. Also, these electrochemical sensors are further investigated by for potential cystic fibrosis diagnostic applications. Figure 2(A) shows the schematic of an electrochemical sensor with bipolar junction transistor (BJT) as the transducer. The sensor consists of a NPN BJT device with its base connected to a silver chloride coated silver (AgCl/Ag) wire that forms the chloride sensitive sensing surface, in contact with the aqueous solution. A miniaturized reference electrode is also immersed in the solution. The BJT device is fabricated using standard silicon processing technology as discussed elsewhere 22 . Post fabrication, the BJT base is connected to an AgCl/Ag wire; the AgCl/Ag preparation is described in the Methods section. The BJT emitter area is 2.5 × 10 −7 cm 2 (25 μ m 2 ) and the sensing surface area is ~0.04 cm 2 . The voltages applied at the emitter, collector and reference electrode are V E , V C and V B , respectively. The collector current (I C ) is the sensing signal and sensing measurements are made with V C = V B = 0 V and V E is either varied or set at a fixed value.

Electrochemical sensors with FET and BJT as transducers.
The equation for the sensing signal I C is briefly reviewed 18 : where, T is the temperature in Kelvin, k is the Boltzmann constant, I o is a BJT device constant, and q is the electronic charge and V BE = (V B − V E ). SS is the sub-threshold swing that is defined as the change in V BE corresponding to a decade change in I C , therefore, SS = 2.3 kT/q from equation 1. In contrast, the sensing surface potential ψ s depends solely on the sensing surface charge density associated with bound analyte, and is therefore a characteristic of the sensing surface/analyte interaction. Figure 2(B) shows the schematic of an electrochemical sensor with a FET device as the transducer. The sensor has an N-type FET device with its gate connected to AgCl/Ag wire with AgCl forming the sensing surface in contact with the solution. A reference electrode is also immersed in the solution. The sensing surface is attached to the gate post FET fabrication. The silver chloride sensing surface, reference electrode and FET device area (= 25 μ m 2 ) are the same as those for the BJT sensor. The FET device consists of a dual gate SiO 2 /HfO 2 gate dielectric layer with TiN as the gate. SiO 2 thickness is 1.5 nm and is grown by thermal oxidation at 900 °C, and HfO 2 physical thickness is 2.0 nm and is deposited by atomic layer deposition (ALD) at 400 °C. The equivalent oxide thickness (EOT) of the dual layer stack is ~1.8 nm as determined by the accumulation capacitance. TiN thickness is 50 nm and is sputter deposited at 300 °C. More fabrication details of the HfO 2 FET are given elsewhere 23,24 . These SiO 2 /HfO 2 FETs are advantageous for the transducer application due their significantly lower gate currents in comparison to SiO 2 gate dielectric FETs with heavily doped poly silicon gate. The optimized SiO 2 /HfO 2 FET has a low gate current density of about 10 −3 /cm 2 at a gate voltage of 1.0 V (Supplementary, Fig. S1). This current density is about 500 times lower than that in a standard SiO 2 FET of similar gate dielectric thickness 25 . Since electrochemically induced sensing surface degradation depends on the gate current density, reduced gate current results in improved sensing surface life time. The threshold voltage, sub-threshold swing s(SS) and drain current hysteresis (Δ H) are other parameters impacting sensor performance. For example, low threshold voltage implies that the use voltage would be small, a low SS value results in enhanced sensitivity as discussed later, and small Δ H indicates that the sensing signal is reversible with gate voltage sweep. The optimized SiO 2 /HfO 2 FET device has threshold voltage of ~0.24 V, SS = 71 mV/decade and Δ H < 1 mV (Supplementary, Fig. S1); these values are comparable to those for a SiO 2 FET device. Hence, these SiO 2 /HfO 2 FETs have significantly lower gate currents, an important advantage over SiO 2 FETs.
The FET electrochemical sensor measurements are made as follows. The drain current (I D ) is the sensing signal. The voltages applied at the drain, source, substrate and reference electrode are V D , V S , V SUB and V G , respectively. Sensing measurements are made at room temperature with V S = V SUB = 0 V, V D = 50 mV while V G applied at the reference electrode is either varied or set at a fixed value. For an electrochemical sensor with FET transducer, the equation for the sensing current I D is the same as the transfer curve equation for a standard FET device except V G is replaced with (V G + ψ s ) 1 .

Comparison between BJT and FET Transducers.
Using the electrochemical sensors shown in Fig. 2, sensing measurements are performed for FET and BJT sensors. Measurements are performed in an aqueous KCl solution of varying concentrations (0.01-100 mM). Since KCl is fully ionized in water at room temperature, Cl − and KCl concentrations are assumed to be the same. Figure 3A shows the dependence of the sensing signal I C on the applied voltage V BE measured at various chloride concentrations ([Cl − ]) for the BJT sensor. Symbols denote measurements and solid lines are exponential fits to the data in accordance with equation 1. At each Cl − concentration, I C increases exponentially as V BE increases with a sub-threshold swing (SS) of 59 mV/decade, consistent with the room temperature Nernst value. This SS value is same as that for the stand-alone BJT device (Supplementary, Fig. S2), and therefore SS is an intrinsic transducer property. Figure 3B shows the dependence of I D on gate voltage (V G ) measured at various [Cl − ] for the FET sensor. At each Cl − concentration, the sensing signal I D increases exponentially with SS = 71 mV/decade at lower V G and has a linear dependence at higher V G , consistent with the FET transfer curve equation. This SS = 71 mV/decade for the FET transducer is same as that for the stand-alone FET device (Supplementary, Fig. S1), thereby indicating once again SS is an intrinsic transducer property. From Fig. 3A,B three main observations can be made. (i) Sensing current versus applied voltage curves are significantly different for BJT and FET sensors. (ii) Sub-threshold swing are independent of chloride concentration for both sensors and SS for BJT sensor is smaller than that for the FET sensor. (iii) For both sensors, I C and I D curves shift as chloride concentration varies, thus indicating that threshold voltage V T depends on [Cl − ]; V T is defined as the applied voltage value corresponding to a constant sensing current of 1 nA. As chloride concentration increases in the solution, more Cl − bind to the sensing surface, thereby causing surface potential ψ s to shift with a concomitant shift in V T : i.e. Δ V T = Δ ψ s . Figure 3C shows dependence of V T on [Cl − ] for BJT and FET sensors: V T varies reversibly with Δ V T = 57 ± 1 mV/pCl for both sensors. This similarity is understandable since both electrochemical sensors have the same sensing surface with similar Δ ψ s response. In summary, Fig. 3C shows that Δ V T is independent of the transducer device choice.
Transconductance (g m ) is another important property impacting sensor performance. g m is defined as (δ I C /δ V BE ) and (δ I D /δ V G ) for BJT and FET sensors, respectively. Using data from Fig. 3A,B, g m is estimated. Figure 3D shows the dependence of g m on the sensing current for BJT and FET sensors. For sensing currents < 10 nA, g m increases with a power law dependence on the sensing signal with an exponent of 0.99 at for both sensors, irrespective of the chloride concentration. At sensing currents > 10 nA, g m corresponding to the FET sensor starts to decrease and is less than that for the BJT sensor. These gm curves for BJT and FET transducers are similar to those observed for the stand-alone devices (Supplementary section, Fig. S3), thereby indicating that g m is an intrinsic transducer property. Using these results from Fig. 3, the performance metrics for BJT and FET sensors are compared as discussed.
An electrochemical sensor sensitivity can be defined as the relative change in the sensing current (I) for a fixed change in the target analyte concentration. From equation 1, sensor sensitivity can be written as: s m In the above equation, Δ ψ s is the change in the surface potential corresponding to a fixed analyte concentration change. Figure 3 shows that Δ ψ s is same for both BJT and FET sensors. Hence, for the purpose of transducer comparison, Δ ψ s can be set to 1 V and sensor sensitivities are estimated from equation 3. For the BJT sensor, sensitivity Δ I C /I C = g m /I C, where the transconductance is g m = (δ I C /δ V BE ). Similarly, the FET sensor sensitivity can be written as: Δ I D /I D = (δ I D /δ V G )/I D . Using the transfer and transconductance curves shown in Fig. 3, sensor sensitivities at various chloride ion concentrations are estimated and compared. Figure 4A compares the BJT sensor sensitivity with the FET sensor sensitivity. The BJT sensor sensitivity (Δ I C /I C ) is observed to be independent of I C , implying that sensing measurements can be made with the same high sensitivity over the entire I C range of several decades, irrespective of applied V BE . In contrast, the FET sensor sensitivity Δ I D /I D varies with the sensing signal, and therefore applied V G has to be chosen such that I D is in the range of 0.05 to 0. 5 nA to achieve maximum sensor sensitivity. Also, BJT sensor sensitivity (Δ I C /I C ) is increasingly higher than the FET sensor sensitivity (Δ I D /I D ) for sensing currents > 1 nA. Hence, an electrochemical sensor with a BJT transducer is demonstrated to have superior sensitivity characteristics.
We now discuss calibration curves for BJT and FET electrochemical sensors: the calibration curve is defined as the sensing signal dependence on analyte concentration measured at a fixed applied voltage. To compare these two calibration curves, the sensing signal is normalized by the signal measured at 100 mM KCl concentration. Figure 4B shows the dependence of normalized sensing current on the chloride concentration for BJT and FET sensors at various applied voltages. For the BJT sensor, the normalized I C increases by a factor of 10 per pCl, irrespective of the applied voltage V BE value. In contrast, normalized calibration curve depends on the applied gate voltage V G for the FET sensor. Normalized I D increases by a factor of 8 per pCl over a narrow range of V G ~ 0.2 to 0.25 V and this dependence becomes weaker at higher V G . In summary, normalized calibration curves are independent of applied voltage for a BJT sensor, and therefore it is better suited for mobile sensing applications.
Signal to noise ratio (SNR) is another important sensor performance metric. As discussed elsewhere 26 , SNR measures the sensor resolution (i.e. smallest measurable change in ion concentration) and can be written as:  Using equation 5, the SNR is estimated as follows. Sensing current noise power density at 1 Hz (S I ) and transconductance (g m ) are measured for BJT and FET sensors in KCl solution of varying concentrations, and also for the stand alone device (i.e. no sensing surface or solution). To measure S I for the BJT sensor, the collector current is measured as a function of time at a sampling rate of 50 ms at different V BE values with V C = V B = 0 V. The collector current time domain data is analyzed using the Fast Fourier Transform Welch method 27 and the I C noise power density (S I ) as a function of frequency is estimated. Similar noise measurements associated with I D are performed for the FET sensor. Figure 5A shows the dependence of S I on the sensing current for sensors with BJT and FET transducers. For both sensors, S I shows no dependence on the chloride concentration and is the same as for the stand-alone device without the sensing surface and solution. This implies that the measured sensor noise is mainly due the transducer. S 1 has a power law dependence on the sensing current with an exponent of 2 and 1.3 for BJT and FET sensors, respectively. Also, S 1 is lower for the BJT sensor, particularly at sensing currents < 10 −7 A. Transconductance (g m ) values for BJT and FET sensors are estimated from Fig. 3D. Using the noise data (Fig. 5A) and g m (Fig. 3D), signal to noise ratio (SNR) values are calculated from equation 5 for the two sensors. Figure 5B compares the signal to noise ratio (SNR) for the two chloride ion sensors. The SNR is observed to be independent of the chloride concentration and is the same as for the stand-alone device (i.e. no solution or sensing surface) for both sensors, implying that the transducer device is the dominant cause of the measured noise. For the BJT sensor, the SNR is higher and remains constant over the entire sensing range. This is a significant advantage because the signal can be measured with the same high SNR, irrespective of applied voltage V BE . In contrast, the maximum SNR is observed at I D ~100 nA for the FET sensor, and therefore the corresponding V G has to be carefully chosen to achieve maximum SNR. From Figs 4A and 5B, we observe that the maximum sensitivity and SNR occur at different sensing current values, and as a result a tradeoff has to be made between sensitivity and SNR for a FET sensor. Since SNR is a measure of the sensor resolution, the above SNR observations are also true for the sensor resolution.
To further highlight the sensor resolution dependence on the transducer choice, a dilution experiment is performed using BJT and FET sensors. Starting with a 10 mM KCl solution of 1.6 mL volume, the solution is diluted in steps by adding 20 μ L of deionized water. Each dilution step results in chloride concentration change of Δ pCl ~0.0051. To ensure that the comparison is accurate, the applied voltage (V BE or V G ) is fixed at a value such that the corresponding sensing current at the initial 10 mM KCl concentration is the same. At each dilution step, the sensing current is measured for about 100 sec after a wait of about 1 minute. Figure 5C shows the BJT sensing signal I C measured at a fixed V BE = 0.26 V at various dilution steps. Figure 5D shows dilution experiment results for a FET sensor, where the sensing signal I D is measured at a fixed V G = 0.284 V. Figures 4C and 5D measurements are analyzed and results are shown in Fig. 5E,F. Figure 5E shows mean and standard deviation values of BJT sensing current I C at various dilution steps. Mean I C value increases by Δ I C ~3 pA per dilution step and the standard deviation δ I C ~ ± 0.15 pA. Since the ratio Δ I C /δ I C ~ 10 and Δ pCl ~ 0.0051 at each dilution step, estimated BJT sensor resolution is about 0.0005 which is the highest reported value to the best of our knowledge 20,21 . Figure 5F shows mean and standard deviation values of sensing current I D at various dilution steps for the FET sensor. Mean I D value increases by Δ I C ~ 2.2 pA per dilution step and the standard deviation δ I C ~ ± 1 pA. Since signal change and signal noise are of comparable magnitude for the FET sensor, FET sensor is unable to resolve Δ pCl ~ 0.0051 at each dilution step. A key observation from Fig. 5C,D is that sensing signal I D has higher (about 6 times) noise in comparison to that for I C , thus indicating that the FET transducer is noisier than the BJT transducer. This higher noise in I D can be attributed to the presence of gate dielectric related Si/SiO 2 interfacial traps in the FET device, whereas the BJT device has no such gate dielectric related interfacial traps. In summary, the BJT sensor has lower noise and therefore has higher resolution in comparison to the FET sensor.
Another difference between these two electrochemical sensors is the voltage applied to the aqueous solution by the reference electrode. In the case of a BJT sensor, the reference electrode applies 0 V to the solution during sensing measurements. In contrast for the FET sensor, the reference electrode applies a voltage |V G | > 0 V to the solution during sensing measurements. This |V G | > 0 V can cause interference, particularly if the sensing is being performed on voltage gated ion channel proteins such as those embedded in neuron membranes. Hence, BJT sensors are better suited for a wider range of biological systems. Lastly, like a FET transducer, a BJT is also compatible with silicon processing technology and can be integrated with automation circuitry on a single chip. Both FET and BJT can easily be miniaturized to 0.1 um 2 but the FET is easier to further scale down. In summary, this comparative study demonstrates that the electrochemical sensor with a BJT as the transducer has significantly enhanced sensing characteristics in comparison to that with a FET transducer.
Cl − Sensing in Artificial Human Sweat for Diagnostic Applications. Since electrochemical sensors with BJT as transducers have significantly enhanced sensing characteristics, they are further investigated for as a potential candidate for hand-held diagnostic application for Cystic Fibrosis disease that requires the measurement of [Cl − ] in human eccrine sweat 19 . Chloride sensing measurements are performed using artificial human sweat that are similar in chemical composition to the human eccrine sweat. Measurements are performed on a set of seven artificial sweat samples that have been diluted by varying amounts with ultrapure water: the sweat percentage ranged from 100 (no dilution) to 0.8 (125 times dilution). Figure 6A shows measured BJT sensor transfer curves for undiluted and diluted sweat samples. For each sample, I C increases with increasing V BE in accordance with equation 2 with an exponent that corresponds to 25 °C. This dependence is similar to that observed in aqueous KCl solutions (Fig. 3A). Also, I C curves are observed to shift towards the left with decreasing sweat percentages, thereby indicating that [Cl − ] is decreasing. In Fig. 6B, the sensing signal (I C ) reversibility and reproducibility in sweat is investigated. I C is measured for about 100 sec at a fixed V BE = 0.2 V for each artificial sweat sample. Measurements are made on samples in the order of first decreasing and then increasing sweat concentration. I C is observed to be repeatable and reversible within 4% of error. Though, the focus is on BJT sensor, similar sensing measurements are also performed with the FET sensor to verify that chloride sensitivity comparisons are same in sweat as those of Fig. 4B. In Fig. 6C Since artificial sweat has amino acids, minerals and metabolites, it is possible that these amino acids and metabolites would nonspecifically bind to the sensing surface, and thus cause signal drifts or degrade the sensitivity to chloride ions. To evaluate these effects, the silver chloride sensing surface is incubated in 100% artificial sweat for total time of 168 hours (1 week); intermittently the incubation is interrupted and chloride sensing measurements are made in the incubating sweat. Figure 7A shows the measured dependence of I C on V BE after various incubation times. From the figure, the transfer curves show a small shift of ~3 mV after a week of incubation. Since the curve shifted back when re-measured in fresh sweat, the observed drift is attributed to a slight increase in the chloride ion concentration due to the sweat evaporation during the long incubation period. Hence, the sensing signal shows negligible (<1 mV) drift after one week of incubation in artificial human sweat. This result is understandable because sweat is mostly (99%) water with low concentrations of biomolecules and as a result non-specific binding is minimal.
To measure the impact of week long incubation in sweat on the sensing surface sensitivity to chloride ions, the calibration curve at a fixed V BE = 0.2 V is measured as a function of incubation time as shown in Fig. 7B. The calibration curve (I C versus [Cl − ]) remains unchanged with incubation time, thus indicating that the silver chloride sensing surface is not degraded with prolonged incubation in sweat. At the end of the week long incubation, the sensing signal I C at a fixed V BE = 0.3 V is repeatedly measured 6000 times in sweat as shown in Fig. 7C. The signal shows no drifts with repeated measurements and has 0.13% noise which is similar to the noise measured before incubation. In summary, incubation in artificial sweat does not induce signal drifts, calibration curve degradation or noise increase for the BJT sensor. Conclusion BJT and FET transducers are compared by performing sensing measurements on two electrochemical sensors that are identical in all details, except for the transducer device type. This comparative study demonstrates that an electrochemical sensor with the BJT transducer has significantly enhanced sensing characteristics in comparison to that with the widely used FET transducer. The BJT sensor has sensitivity and resolution and calibration curves that are independent of the use voltages. Hence, BJT sensors are particularly well suited for making high sensitivity and resolution sensing measurements with minimal calibration requirements and are well suited for mobile sensing applications. As a demonstration for mobile diagnostic applications, the BJT electrochemical sensor is investigated by measuring chloride levels in artificial human sweat and is shown to be a viable option for portable cystic fibrosis diagnosis.

Methods
Chloride ion sensing surface preparation. The silver chloride is used as the chloride ion (Cl − ) sensing surface for both BJT and FET sensors. It is prepared by the electrochemical anodization of a silver (Ag) wire (Sigma Aldrich, part# 265586, ≥ 99.99%; 1 mm diameter) in a chloride solution. To prepare a silver chloride coated silver wire, a silver wire is first cleaned by sonication in ethanol (~5 mins) followed by rinse in ultrapure Millipore water. The cleaned silver wire is used as the anode and a platinum wire (Sigma Aldrich, part# 267201, 99.99%) is used as the cathode in an electrolytic cell. Since KCl 28 , HCl 29 and their mixtures 30 have been widely used as an electrolyte for anodization of the Ag wire, we have evaluated different electrolytic solution with the aim of identifying the best recipe. The results are summarized in Table S1 in Supporting Information Section. As shown in Table S1, the recipe that uses 1 M HCl as the electrolytic solution with a constant current of 2 mA/cm 2 for 15 minutes produced a silver chloride surface with highest chloride sensitivity and least drifts during electrical measurements. The silver chloride surface prepared by this recipe is used for all sensing measurements.