pH Response of Silicon Nanowire Sensors: Impact of Nanowire Width and Gate Oxide

We present a systematic study of the performance of silicon nanowires (SiNWs) with different widths when they are used as ion-sensitive ﬁeld-effect transistors (ISFETs) in pH-sensing experiments. The SiNW widths ranged from 100 nm to 1 µm. The SiNW-ISFETs were successfully fabricated from silicon-on-insulator (SOI) wafers with Al 2 O 3 or HfO 2 as gate dielectric. All the SiNWs showed a pH Response close to the Nernstian limit of 59.5 mV/pH at 300 K, independent of their width, or the investigated gate dielectric or operating mode. Even nanowires (NWs) in the 100 nm range operated reliably without degradation of their functionality. This result is of importance for a broad research ﬁeld using SiNW sensors as a candidate for future applications.


Introduction
In recent years, the ion-sensitive field-effect transistor (ISFET) (1,2) has been scaled down to a silicon nanowire ISFET (SiNW-ISFET) with sub-100 dimensions. (3)(4)(5)(6) The reduced size offers unique possibilities, such as the dense integration of multiplexed NWs, (7,8) which sense a large number of different target molecules simultaneously using differently functionalized NWs. (9) The key feature for biological applications is that nanostructures are on the scale of biological specimens. Recently, Tian et al. (10) have demonstrated a manipulation of the living cell with nano-FETs without damaging the cell.
The sensing mechanism of SiNW-ISFETs is based on the accumulation of charged molecules on the NW surface, which leads to a surface potential shift. The transistor then responds to changes in the surface potential with a threshold voltage shift ΔV th . In a pH-sensing experiment, the accumulated surface charge originates from OH groups located on the surface of the gate dielectric, which are positively (protonated) or negatively (deprotonated) charged, according to the pH of the buffer solution. This surface charge provokes the pH response of the SiNW-ISFET in the form of a threshold voltage shift. Consequently, the pH response is given in mV/pH. Several publications addressed the question on whether the NW width (4,(11)(12)(13) or thickness (14,15) has an effect on the pH response; however, the results were inconsistent. An effect of the NW width on the response in pH-sensing experiments has been shown in refs. 12 and 13. Elfström et al. (12) observed an increasing threshold voltage shift with decreasing NW width. No pH response was found for NW widths > 150 nm. Vu et al. (4) reported preliminary results for single 80-and 400-nm-wide wires that exhibit the same threshold voltage shift of 41 mV/pH. Moreover, the ideal working point showing the highest pH response is also a controversial discussion in the literature. Gao et al. (16) demonstrated an enhancement of the pH response in the subthreshold regime compared with that in the linear regime, whereas preliminary results described in ref. 4 suggest the same pH response in both regimes. To clarify these disagreements, we explore in a systematic manner the effect of the NW width of SiNW-ISFETs on the pH response with widths ranging from 1 µm down to 100 nm in both operating modes, the subthreshold and linear modes. In addition, we study the sensing properties of the SiNW-ISFET coated with Al 2 O 3 and HfO 2 grown by atomic layer deposition (ALD). Compared with SiO 2 , (4,10,11,16) these oxides provide a leakage-free (17) operation with an ideal Nernstian pH response that is independent of the salt concentration. (5,6,18) For the application of wide pH ranges, HfO 2 is a predestined candidate as a sensing surface due to its chemical inertness in most acidic or basic solutions.

Materials and Methods
The SiNW-ISFETs were fabricated from silicon-on-insulator (SOI) wafers following the top-down approach. (11) The SOI wafer (Soitec, France) has a buried oxide layer of 145 nm and an 85-nm-thick low-doped p-Si(100) device layer with a resistivity of 8.5 to 11.5 Ω·cm. First, the Si device layer was thermally oxidized with a 15-nm-thick SiO 2 layer. Then, the pattern of the SiNW-ISFET was defined by electron beam lithography (EBL) in a negative resist (AZ nLOF 2000) and transferred to the SiO 2 layer by reactive ion etching. The patterned SiO 2 was used as an etching mask for the wet chemical etching of the Si device layer in tetra-methyl ammonium hydroxide (TMAH) with 10 vol.% isopropanol (IPA) at 45 °C. The wet chemical etching of the Si(100) layer is highly anisotropic with etch rates depending on the crystallographic orientation of the silicon. (19) The Si(111) planes are etched approximately 100 times slower (11) than the other planes, which results in a trapezoidal shape of the NW cross section. A key feature of our sensors is that a single SiNW-ISFET chip contains 48 NWs with eight top widths W top [ Fig. 1(a)] ranging from 1 µm down to 100 nm. The 48 NWs are arranged in four arrays with twelve NWs sharing the common drain contact. The length of the NWs in the PMMA resist by EBL. The resist acted as an implantation mask for the following implantation with a BF 2 + dopant dose of 2.3×10 15 cm −2 at 33 keV. The implanted area approached the NW to a distance of ∼200 nm. The NW retained its original low p-type doping concentration. After implantation, the dopants were activated in a thermal annealing step at 950 °C for 6 min in forming gas. The NW array with three NWs connected to the implanted source and common drain contact is shown in Fig. 1(c).
The passivation of the SiNW-ISFET, which is necessary for measurements in liquids, was carried out by atomic layer deposition (ALD) of Al 2 O 3 and HfO 2 (Savannah S100 Cambridge NanoTech). The ALD oxides were deposited directly on the NW surface. Immediately prior to the ALD oxide growth, a Radio Corporation of America (RCA) cleaning was done. During the RCA cleaning, the 15-nm-thick thermal SiO 2 was etched in buffered hydrofluoric acid (BHF). After the growth of the ALD oxides, the pattern of the source/drain contact windows was defined by optical lithography in the resist AZ nLOF 2000. The contact windows were then opened in BHF. The etch rate of Al 2 O 3 is comparable to that of thermal SiO 2 , whereas HfO 2 etches 14 times slower. Immediately after the etching, the opened contacts were metalized with Al-Si(1%) in a lift-off process. The formation of the ohmic contacts was completed in a thermal annealing process in forming gas at 450 °C. In the same step, the ALD oxides were also postannealed. It is known that the postannealing of the ALD oxides in forming gas has the effect of reducing the density of interface states D it as they are passivated by hydrogen. (20) Also, the dielectric constants of the oxides are affected as the oxide structure changes from amorphous to (poly)-crystalline, depending on the annealing temperature. However, the ALD oxides have a large excess of O 2 as they are grown in an oxygen-rich environment. Therefore, there is the risk that during the postannealing, the excess O 2 diffuses to the Si where it leads to the growth of the interfacial SiO 2 layer (20,21) that already formed between the Si substrate and the ALD oxide during the ALD deposition. (20) The smaller dielectric constant of 3.9 for SiO 2 will reduce the total dielectric constant of the ALD layer. To integrate the fluidic system, microchannels were fabricated in the SU-8 resist by optical lithography. Figure 1(d) shows a micrograph of the SiNW-ISFET after this process step. One microchannel aligned perpendicular to the NWs can be seen. After the wire bonding, the contact pads including the wire bonds were sealed with epoxy. In the experiment, the microchannel was filled with an electrolyte, and the electrostatic potential of this electrolyte was mediated through the liquid to the NWs. This is known as liquid gating. The pH-sensing measurements were carried out in standard pH buffer solutions (Titrisol, Merck) with different pH values. In Fig. 1(e), a schematic of the measurement setup is shown. (17,22) The device was operated at a source-drain voltage V sd of 0.1 V and the conductance G through the device was measured. The conductance was modulated by a top liquid gate voltage V lg that was applied to a platinum wire immersed in the buffer solution. A calomel reference electrode was used to measure the effective potential V ref in the buffer solution against ground. Additionally, a back gate voltage V bg can be applied to the back side of the sample. The measurements were performed with buffer solutions with pH values ranging from 3 to 9 and from 3 to 10, for Al 2 O 3 and HfO 2 , respectively. For Al 2 O 3 the highest pH was 9 as the Al 2 O 3 is etched at higher pH values. The basic buffers did not attack the HfO 2 layer due to its chemical inertness to acidic or basic solutions, apart from hydrofluoric (HF) acid-based solutions. We started with the buffer at pH 3 and exchanged sequentially the buffers with steps of ΔpH = 1 until the buffer with the highest pH was reached. We then returned stepwise to pH 3.
All the 48 NWs of one SiNW-ISFET chip were measured automatically at different pH values with a LabView-controlled measurement setup including a switching box (Keithley 3706).  (8) which is in agreement with the site-binding model. (2) Here, to determine the pH response, we do not extract the threshold voltage V th according to   Fig. 2(a) dashed line] is plotted as a function of pH. A linear dependence is found over the whole pH range. The pH response is defined by the slope of the best fit line. The NW with W top = 1 µm shows the pH response for pH values measured from 3 to 9 and back to 3 with a negligible hysteresis. one of the classical methods, (23) as we are not interested in the absolute V th values. It is sufficient to know the relative shift of the transfer curves. Therefore, we defined V th as the reference potential V ref at any source-drain conductance value in the subthreshold regime. In our experiments, this value was 20 nS as indicated by the dashed line in the semilogarithmic plot of Fig. 2(a) Fig. 2(b). A linear dependence is found over the whole pH range. Moreover, a small negligible hysteresis in the forward and backward sweeps of the pH values is observed. (The drift rate of V ref is a few mV/h for samples covered with 20-nm-thick Al 2 O 3 .) The slope of a linear fit to the data defines the pH response of the NW in mV/pH. These slopes are shown in Fig. 2(b) for NWs with W top ranging from 100 nm to 1 µm. All the NWs show an ideal Nernstian behavior of 60 mV/pH at 300 K, independent of the NW width or gate oxide type and without degradation of the pH response over time. To the best of our knowledge, this is the first systematic study showing a clear absence of any size dependence on pH-sensing properties of NWs. The observed ideal pH response confirms recent results showing that ALD-grown Al 2 O 3 (18) and HfO 2 (6) are excellent pH-sensing interfaces, owing to the high density of active surface groups that buffer the pH changes in the bulk solution. Within the site-binding model, (2) this oxide property is called the surface buffer capacity. As long as the oxide/ electrolyte interface of the surface of the NW provides a large surface buffer capacity for protons, no size dependence is expected. Our observation is therefore in strong contrast to the work described in refs. 12 and 13.

Results and Discussion
In an application, it is convenient to measure a large conductance. For fast readout, it is therefore beneficial to operate the ISFETs in the linear regime and not in the subthreshold regime. In the linear regime, the measured signal is the conductance change ΔG (pH) at constant V ref . It can be converted to a voltage shift ΔV ref using the transconductance g m = dG/dV gate : (1) To investigate the influence of the NW width on the response of ΔG to pH changes in the linear regime, one has to carefully analyze the transconductance g m . A comparison between different NW widths requires a normalization of the conductance because it changes with the wire size and shape. (24,25) Different methods could be assumed, such as normalization with the cross-sectional area of the wire, the top width W top , the top perimeter W eff = W top + 2W side , the bottom width W bottom , or even the full perimeter of the wire W eff + W bottom . It is not clear a priori which method is the best, because we do not know the exact current distribution along the NW cross section. If the current flows homogeneously over the full cross section, one would have to scale with the area. There are many reasons, however, why this might not be the case. For example, we can gate the NW asymmetrically and thereby accumulate more carriers at the top surface as compared with the bottom surface, or vice versa. It is also known that surfaces with different crystallographic orientations have different band alignments; (21) hence, the accumulation of carriers may start differently at different crystal faces. Since a detailed systematic study of these variabilities is beyond the scope of this work, we assume that the current scales with the top perimeter, i.e., W eff = W top + 2W side . This assumption is motivated by the gating scheme via the liquid gate (top gate), which is used in the experiments in this work. It is further justified, because the electrolyte is in contact simultaneously with both the top and side faces. We thereby neglect the possible dependence of the current on the crystal orientation.
In Fig. 3(a), the transfer curves normalized by W eff are shown for two NWs with W top of 100 nm and 1 µm at different pH values. The linear regimes of both normalized transfer curves are very similar, which indicates an identical scaled transconductance for different wire widths. Moreover, the scaled transconductance values for a set of NWs with different W top and HfO 2 or Al 2 O 3 as gate oxide are shown in Fig. 2(b). The error bars result from an average over two NWs with an equivalent W top . All the NWs with different widths show a similar scaled transconductance. On the basis of this result, we can conclude that in the linear regime, the normalized conductance change ΔG, and therefore, the response to pH changes, is also independent of the NW width and the operating regime. Our result is in contrast with the results shown by Gao et al. (16) Therefore, our results prove that one can fabricate NW sensors with ideal Nernstian behaviour with NWs down to 100 nm width, and an operation in the linear and subthreshold regimes is favourable. Although the scaled transconductance values are constant for wires of different widths, we have to understand the different g m /W eff values for different top gate oxides. In the linear regime, g m is determined by the mobility µ of the charge carriers and the liquid gate capacitance per unit area, C lg : Here, L denotes the length of the NWs, which was constant for all the wires. We assume that for one sample, the mobility is constant for all the wires of different widths as it is indicated by the constant value of the scaled transconductance. Sample-to-sample variations in the mobility may still occur in the fabrication process. Therefore, the scaling of g m with W eff will then equally be reflected in the capacitance. As a consequence, the value of the scaled transconductance g m /W eff should be determined using C lg . The liquid gate voltage is applied to the NW through the liquid via the top oxide (TOX). It, therefore, involves the double-layer C dl in the electrolyte and the top-oxide capacitance C TOX connected in series. Owing to the high ion concentration in the electrolyte of 100 mM, we have C dl >> C TOX . In the series connection, this leads to C lg ≈ C TOX and g m /W eff = µC TOX /L.
Looking at Fig. 3(b), one would naively expect a factor of two difference in the measured g m /W eff values between the HfO 2 samples with 10 and 20 nm thicknesses, instead of 1.3. To resolve this discrepancy, we determined the dielectric constants ε r by capacitance-voltage (C-V) spectroscopy. We obtained ε r = 12 and 17 for HfO 2 layers with 10 and 20 nm thicknesses, respectively. The 20-nm-thick Al 2 O 3 layer had a dielectric constant ε r = 7.2. Our results are comparable to the reported values of ε r = 10 -18 (26) and ε r = 4-10 (21) for HfO 2 and Al 2 O 3 , respectively. From our C-V measurements, we see that the dielectric constants are thickness dependent as suggested by the g m values of the HfO 2 data in Fig. 3(b). The reason for this might be the presence of an interfacial SiO 2 layer of 2 nm maximum thickness, which grows during the ALD process (20) between the Si substrate and the ALD oxides. It reduces the dielectric constant of the total layer, according to the series capacitance of the SiO 2 and the ALD oxide layer. Knowing the dielectric constants of HfO 2 layers, we estimated a C TOX capacitance ratio of 1.4 between the 10-and 20-nm-thick HfO 2 samples. Indeed, this result shows that we cannot expect a factor of two difference in the measured g m /W eff values in Fig. 3(b). The slight discrepancy between the C TOX capacitance ratio, which is 1.4, and the ratio of 1.3 for g m /W eff can be attributed to slight differences in the mobility. On the basis of our data, we estimated mobilities of 58 and 62 cm 2 /Vs for the 10-and 20-nm-thick HfO 2 samples, respectively. For the Al 2 O 3 sample, we obtained an estimated hole mobility of 75 cm 2 /Vs. The mobility values are in good agreement with the reported hole mobilities for SOI wafers in the accumulation mode (50-139 cm 2 /Vs), (11) with similar doping concentrations. The slightly reduced mobility observed in both samples covered with HfO 2 is in agreement with previous results in the literature, which report a mobility degradation in MOSFETs fabricated with high-κ materials. (27)

Conclusions
In conclusion, we have systematically studied the effect of the NW width W top ranging from 100 nm to 1 µm on the pH response of SiNW-ISFET. Here, the pH response is determined from the voltage shift ΔV ref of the transfer curve in mV/pH. SiNW-ISFETs with Al 2 O 3 or HfO 2 exhibit an ideal linear Nernstian response of 60 mV/pH at 300 K in the subthreshold and linear regimes, without degradation of the Nernstian pH response over time. The drift rate of V ref is a few mV/h. For future experiments, we suggest to correct and thereby minimize the influence of drift on the sensor response by differential measurements, where a passivated NW is used as the reference NW. (28) No effect of the NW width on the pH response was observed. The results of our systematic study are in strong contrast to those described in previous publications as no effect of the wire width or the operating regime on the pH response is observed. Not only the sensing properties but also the electrical properties of the NWs are exceptionally good and reproducible, e.g., the 48 NWs integrated on one sample have a variation in V th of ∼50 mV from NW to NW of the same width. Also, the negligible hysteresis in the transfer curves and leakage currents below 2 nA are the result of our reliable fabrication technique. The hole mobilities and dielectric constants of the ALD oxides are in good agreement with expected values from the literature. We also have proposed a scaling method of the wire conductance using the full top perimeter width in liquid gating experiments, which leads to a constant scaled transconductance for all the wires. The excellent sensing and electrical properties of wires in the 100 nm range prove the feasibility of a dense integration of SiNW arrays, which is demanded for future systems with multiplexing functionality. For multiplexed detection, we suggest differential measurements with a passivated NW as reference wire, in order to minimize disturbances of the signal, such as drift, temperature or nonspecific adsorption.