Detection of ultra-low protein concentrations with the simplest possible field effect transistor

JNT sensors with various nanowire widths (10, 20, 30, and 50 nm), lengths (0.5, 1, and 10 μm) and numbers (1, 3, and 20 NWs) were fabricated on highly p-doped (10¹⁸ cm⁻³) SOI wafers with top Si device layer of 45 nm and buried oxide (BOX) layer of 145 nm SiO₂ (see figure 1(b)). The patterning was done with electron beam lithography (EBL) using a high-contrast and low roughness development process for the negative tone resist hydrogen silsesquioxane (HSQ) [18, 19], followed by a chlorine (Cl₂) chemistry based reactive ion etching (RIE). Then an additional 200 nm SiO₂ layer was deposited on the whole surface except the device regions, to minimise the leakage current through the BOX. Next, the source/drain metal contacts, interconnects and contact pads were created by deposition of a 100 nm thick nickel (Ni) layer and thermal annealing for 30 min at 400 °C in forming gas (10% H₂/90% N₂). These two subsequent depositions were done by electron beam evaporation and were respectively combined with two steps of photolitho-graphy and lift-off. A final third step of photolithography was performed with 2 μm thick negative photoresist SU8, which was used to passivate the whole surface of the chips, except the device regions and the metal contact pads. The fabricated devices were inspected by both optical and scanning electron microscopes (SEM). Detailed description of the fabrication process is given elsewhere [17].

To make the JNT sensors sensitive only to the protein streptavidin, the Si nanowires were functionalised with biotin. Streptavidin has a high affinity for biotin and they form a very strong non-covalent binding. The functionalisation was done in several steps. First, the SiO_x nanowire surface was treated in a piranha solution (3:1 H₂SO₄:H₂O₂) to produce a clean hydroxyl-terminated surface. Next (3-Aminopropyl)triethoxysilane (APTES) was covalently attached to the nanowire surface by immersing the samples in a solution of 5 v/v % APTES in anhydrous toluene for 3 h at 50 °C. Then the samples were rinsed with anhydrous toluene, deionised (DI) water and dried under nitrogen (N₂). The aminosilanised nanowires (Si-APTES) were subsequently immersed for three hours at room temperature in 2 ml phosphate buffered saline (PBS, pH 7.4) with added 100 μl of E, Z link-NHS-LC-Biotin in dimethylformamide (DMF) (1 mg ml⁻¹). During this time, the NHS-LC-Biotin reacts with the primary amine of the surface-tethered silane to leave a biotinylated surface. Finally, surfaces were rinsed with PBS and DI water and dried under nitrogen. Contact angle measurements changed from 19° (highly hydrophilic) to ∼70° (less hydrophilic), indicating successful surface modification.

The electrical characterisation was done at room temperature using a cascade manual probe station and an Agilent B1500 semiconductor parameter analyser.

Efficient fluid delivery to the sensors was enabled by bonding polydimethylsiloxane (PDMS) stamps with a 150 μm width channel to the front surface of the chips (figures 2(a), (b)). Inlet/outlets access holes with diameters ∼400 μm were drilled on the top of the stamp to link the microfluidic channel to external PTFE tubing (figures 2(a), (b)). The tubing was relatively long (∼1.5 m) to enable administration and collection of the fluid from the outside of the probe station. The solutions containing the target analyte were delivered to each sensor at a rate of 100 μl min⁻¹ with a syringe (BD Plastipak, 10 ml) propelled by a pump (Harvard, Pump 11+). From the syringe, the flow passed through PTFE tubing with inner diameter (ID) of 0.4 mm and outer diameter (OD) of 1.0 mm. Approximately 10 cm from exiting the syringe, the flow passed through a T-junction where a small amount of a different solution could be added into the main fluid flow as required. From the T-junction the flow continued along the PTFE tubing (∼1 m) into the probe station. Standard fittings from Upchurch Scientific were used for all connections. Inside the probe station the tubing was downsized to OD 0.4 mm and ID 0.1 mm PTFE tubing using interconnect junctions fabricated in-house from PDMS. The smaller tubing was required to interface to the PDMS stamp but could not be used for the entire length of the fluid supply lines, as excessive backpressure would be generated. The Si NWs were accessible through the small windows on the SU-8 passivation layer.

In some of the measurements under 'liquid conditions', an Ag/AgCl reference electrode (ALS RE-1S, 4 mm barrel diameter) was used to maintain a constant liquid potential. The reference electrode was incorporated into a flow cell and attached to the microfluidic channels by the PTFE tubing. The flow cell consists of a small volume chamber (∼1 μl) at the base of the electrode, which is supplied via a 1 mm diameter channel custom made in PDMS (figure 2(c)).

To check the quality of the fabricated and biotinylated sensors, they were first electrically tested under dry conditions. Figure 3(a) shows a sample transfer characteristics of a device constructed from a single nanowire with a diameter of 50 nm and length of 1 μm. The drain current I_d was measured as a function of the back-gate potential V_bg by sweeping V_bg forward (circles) and backward (triangles) between −20 and 10 V while keeping the drain potential V_d constant at V_d = 1 V. The two transfer characteristics overlap well, indicating the high quality of the fabricated sensors. The threshold voltage V_th shifts slightly towards the right indicating a small negative charge accumulation. Injected interface charges, which do not dissipate as the gate bias polarity changes, lead to a shift in the threshold voltage [20]. Hysteresis can affect the stability of the sensors causing short- and long-term drift of the sensor response [21]. The little hysteresis observed here (<0.33 V of the back-gate potential) indicates small defect-induced charge trapping at the interface Si/SiO₂ [22].

Zoom In Zoom Out Reset image size

As a next step, a similar measurement was done while continuously flowing (100 μl min⁻¹) PBS through the microfluidic channels aligned over the sensors. The PBS potential was controlled by the reference electrode. Figure 3(b) presents sample transfer characteristics of a device having an array of 20 nanowires of 50 nm width and 10 μm length. The drain current I_d was measured as a function of the reference electrode potential V_ref by sweeping V_ref forward (circles) and backward (triangles) between −4 V and 2 V while maintaining a constant drain potential of V_d = 1 V. Here the two transfer characteristics overlap almost perfectly, which is yet another demonstration of the excellent quality of the fabricated and functionalised sensors.

In this work, streptavidin-biotin has been chosen as a model target-probe pair, with streptavidin in solution being sensed using biotin covalently tethered to the NW surface. This target-probe pair is well known, has one of the strongest non-covalent binding interactions in nature, and very fast association kinetics. It is widely used for evaluation of the sensing performance of various sensors and is, therefore, appropriate for comparison of different sensor concepts. The Si nanowires functionalisation with biotin is described in section 2.2.

In the first series of streptavidin sensing experiments, the drain current I_d was measured as a function of the back-gate potential V_bg by sweeping V_bg forward and backward between −20 and 5 V ten times at constant drain potential V_d = 1 V while continuously flowing (100 μl min⁻¹) plain PBS over biotinylated sensors. The measurements of a device with an array of 20 NWs of 0.5 μm length and 50 nm width resulted in on-currents I_on of ∼1.5 μA and off-currents I_off of ∼0.15 nA (I_on/I_off ratio of ∼10⁴) as shown in figure 4(a). Then PBS solution with relatively high concentration of streptavidin (0.4 μM) was flowed at the same rate over the sensors instead of the pure PBS and again a series of ten consequent I_d–V_bg measurements was done at identical conditions. The obtained I_d–V_bg curves for the same device have an obviously much smoother subthreshold slope (SS) and much higher currents, I_on ∼ 10 μA and I_off ∼ 0.1 μA (I_on/I_off ratio of ∼10²), see figure 4(a). These results indicate successful binding of streptavidin molecules to the biotinylated NW surfaces. Since streptavidin has a weakly acidic isoelectric point (Pi ∼ 5.6) [22], its molecules are negatively charged when dissolved in a PBS solution of pH = 7.4, as in our case. Their binding to the nanowire surfaces acts as a negative gating of the sensor and caused accumulation of majority carriers (holes) in the p-type device channel. Therefore, a significant increase of drain current was observed: up to a factor of 10 for the I_on and a factor of 10³ for the I_off (figure 4(a)).

Zoom In Zoom Out Reset image size

As a final step, solutions of different streptavidin concentrations in PBS (pH = 7.4) were prepared from a solution with an initial molarity of ∼380 nM by consecutively diluting it in PBS by a factor of 30 eight times until an extremely low molarity of ∼580 zM (580 × 10⁻²¹ M) was achieved. During the sensing experiment, a steady stream of pure PBS was flowed through the microfluidic channel at a rate of 100 μl min⁻¹ and periodically small amounts (100 μl) of nine different solutions with increasing streptavidin concentrations (580 zM, 17 aM, 520 aM, 16 fM, 370 fM, 14 pM, 420 pM, 13 nM and 380 nM) were administered one by one into this steady stream, while continuously measuring the drain current I_d of the sensing device. The reference electrode was used to maintain constant potential of the solutions. In figure 4(b), I_d of a biotinylated sensors having an array of 20 nanowires of 50 nm width and 10 μm length and biased at V_ref = −1.5 V and V_d = 1 V is shown as a function of time. First, as a control experiment, 100 μl of plain PBS were injected separately into the main channel solution at the time mark of ∼900 s (figure 4(b)), by the use of a T-junction. From the T-junction, the flow continued along the PE tubing (∼1 m) into the probe station. As expected, this injection caused no significant change in the device current, see figure 4(b). This is a clear proof that all the subsequent signals obtained by the sensor are caused exclusively by streptavidin attachment to the nanowire surface.

After the initial plain PBS injection, 100 μl PBS solution with the lowest streptavidin concentration of 580 zM was injected into the main channel solution at the time mark of ∼1100 s (figure 4(b)). A remarkable response of the sensor to this extremely low streptavidin concentration can be clearly seen in figure 4(b). Following this first streptavidin injection, the microfluidic channel and the device were rinsed by plain PBS for ∼3–4 min. Then, the PBS solutions with increasing streptavidin concentrations were introduced step by step within the continuous stream of pure PBS buffer. After each streptavidin injection, the microfluidic channel and the sensor were consequently rinsed from potentially unbound streptavidin molecules by plain PBS for the same time of ∼3–4 min. With each increasing streptavidin concentration, an increasing response of the sensor is clearly observed in figure 4(b) up to the highest concentration of 380 nM.

As mentioned above, the remarkable responses of the sensor to the streptavidin injections are reflective of the protein attached to the nanowire surface. The possible sensor response to the charged streptavidin molecules passing close to the surface of the nanowires might be ruled out for the following reason. In our experiments we used a solution of PBS with pH = 7.4 and medium range concentration of 10 mM, where the Debye screening length is λ_D ≈ 3.1 nm. This means that the unbound streptavidin molecules passing by the surface of the nanowires at a distance >3.1 nm will not have any influence on the sensor current. The ones that come closer then 3.1 nm will either bind to the biotinylated surface of the nanowires because of the extraordinarily high affinity of streptavidin for biotin or non-covalently attach to the nanowire surface. Such a non-covalent attachment might be indicative of incomplete biotinylation of the APTES layer and may be facilitated by the expected positive charge of this layer at pH = 7.4 due to the partial protonation of its amine groups [23]. The low positive potential (in the range of few millivolts) of the APTES layer can electrostatically attract and capture the negatively charged (at pH = 7.4) streptavidin molecules. These molecules, together with the covalently bound ones, most probably cause the high peaks in the measured current observed in figure 4(b). However, the non-covalent attachment of streptavidin is weaker than the covalent binding and streptavidin molecules that are not bound to biotin can be washed off by the constantly flowing PBS. This may be the reason for the current drops after the streptavidin injections have passed over the sensor (figure 4(b)). In such a case, the 'peak' current values are indicative of streptavidin molecules both covalently and non- covalently attached to the nanowire surface, whereas the 'valley' current values should be indicative of streptavidin molecules covalently bound to biotin.

The dashed line (red online) in figure 4(b) extrapolates the possible current drift before the first streptavidin injection, while the solid line (light blue online) connects the 'valley' current values after each streptavidin injection. When one compares the two lines, it becomes obvious that even at the level of those valleys there is a significant jump in the measured current already after the first injection with the extremely low streptavidin concentration of 580 zM, which cannot be attributed to a continuous current drift. Therefore, one could convincingly conclude that the sensor response in our case is indeed reflective of streptavidin molecules binding to the biotinylated nanowire surface.

In one of our previous works [6], we performed similar experiment using another type of sensors, namely 3D vertically stacked Si nanowire FETs. In spite of the fact that the sensing devices were quite different from the current ones and had the opposite doping (n-type instead of p-type), the results were comparable with the ones we present here and we were able to detect streptavidin at a concentration as low as 17 aM. This might be considered as an indirect proof that streptavidin can indeed be reproducibly and reliably detected at these low concentrations.

Plots of the peak current values and valley current values for each streptavidin injection as functions of the respective streptavidin concentrations are presented in figure 5. Logistic fits of the two dependences were suggested with the function

$$\begin{eqnarray}&&y={a}_{2}+\left({a}_{1}\mbox{--}{a}_{2}\right)/\left[1+{\left(x/{x}_{0}\right)}^{p}\right].\end{eqnarray} \tag{ 1 }$$

The logistic function is widely used, among others, in pharmacy and chemistry to model the concentration of reactants and products in autocatalytic reaction.

Zoom In Zoom Out Reset image size

The plots in figure 5 show the distinct increase of drain current with the increase in the streptavidin concentration. There is, however, some decrease in the increment of the sensing signal current for each subsequent injection from about 25% between the first two injections to about 15% between the injections 7 and 8. This might be caused by a partial saturation of the binding sites of the self-assembled biotin monolayer. There is also a slight decrease in the peak current for the last injection of streptavidin with the highest concentration of 380 nM. It is difficult to speculate what could be the reason for this but it is definitely not a result of breaking streptavidin bonds to biotin. A possible explanation might be either oversaturation of the biotin monolayer or a partial desorption of this layer.

In figure 5 it is clearly seen that the slopes of the two curves are different: the dependence of the peak currents on streptavidin concentration is steeper than the one of valley currents. This indicates that different phenomena determine the values of the peak currents and the valley currents. As discussed above, the peak current values are most probably reflective of streptavidin molecules attached both covalently and non-covalently to the nanowire surface, whereas the values of the valley currents are reflective of streptavidin molecules covalently bound to biotin. If they were a result just of a current drift with time, they would have the same slope. This is yet another proof that the sensor response in our case is indeed reflective of streptavidin molecules binding to the biotinylated nanowire surface.

To the best of our knowledge, the concentration of 580 zM is the lowest streptavidin concentration ever detected by any kind of sensor. At this concentration, the number of streptavidin molecules in the injected volume of the PBS solution (100 μl) was ∼35. The upper limit estimation of the microfluidic channel volume is (W × H × L): 200 μm × 200 μm × 1 cm = 0.4 μl. This means that we had in average only one molecule (statistically less than one molecule) in the microfluidic channel at any time. In addition, the functionalisation approach used in this work implies that surface modification with APTES/biotin happens not only at the nanowires but also at the area around them, which is much larger than the NW surface itself. This fact leads to the existence of a vast number of competing sites outside the nanowires, which are capable of binding the target molecules. Statistically, the number of the binding sites outside the nanowires is greater than the number of the binding sites at the nanowires with a ratio much higher than 35:1. Moreover, due to the constant flow of the PBS solution, streptavidin has only a limited time to bind to the nanowires before the unbound molecules are carried away by the flowing PBS. Taking into account all these factors, it is possible to state with a great certainty that at the concentration of 580 zM we most probably detected only few streptavidin molecules closly approaching detection at the single molecule level.

The observed superior performance of JNT sensors might be due to at least two advantages of JNTs over the classical MOSFETs [10–13], which are especially important for their application as sensors:

• (i)  
  The current flow in JNTs is not controlled by a reverse biased p–n junction as in standard MOSFETs but entirely by the gate potential, which modulates the carrier density in the channel. Therefore, they are more sensitive to any change in the electrostatic potential on the channel surface acting as a gate potential.
• (ii)  
  JNTs demonstrate bulk conductance near the centre of the channel (NW), in contrast to the conductance in a thin surface inversion or accumulation layer near the gate in the inversion mode (IM) or accumulation mode (AM) transistors, which leads to higher drive currents. Moreover, this fact makes the modulation of depletion and the conduction in JNTs less affected by the noise-inducing parasitic surface states than in the case of conventional MOSFETs, which is very important for achieving high signal-to-noise ratio and hence the ultimate detection limit of a single molecule [7–9].