Journal Pre-proof Multi-stack insulator to minimise threshold voltage drift in ZnO FET sensors operating in ionic solutions

FET biosensors operating in an electrolyte experience a monotonic, temporal and relatively slow change in threshold voltage caused by the hydration of the insulator layer between the electrolyte and the FET’s channel. Minimising this temporal change in threshold voltage is critical as, over time, the drain current of n-channel FETs decreases, making it difficult to distinguish between the signal generated in response to analyte - receptor binding events and the background noise generated by the electrolyte and the FET biosensor. While Rapid Thermal Annealing of the insulator layer is known to diminish threshold voltage drift and its negative effects, it is not compatible with a low temperature fabrication process of 200°C. Our low temperature approach to minimising threshold voltage drift involves depositing a tri-layer insulator stack, consisting of a layer of HfO 2 between two Al 2 O 3 layers. Wetting ZnO NWFETs with PBS (10 mM phosphate, 150 mM KCl, pH7.4) for an hour, showed that ZnO NWFETs with a stack insulator layer experienced a much smaller threshold voltage and drain current drift (100 mV, 0.064 nA) than ZnO NWFETs with a single material insulator layer (  4300mV, 2.72 nA), Aluminium oxide in this case. Having established the resilience enhancing properties of the stack insulator layer on FETs operating in electrolytes of physiological relevant ionic concentrations; ZnO NWFETs with a stack insulator layer were shown to be capable of detecting the presence of the miDNA-21 strands. This, in effect, paves the way for miRNA sensing experiments in the near future and for exploring the potential of ZnO NWFETs as a diagnostic


Introduction
The duration of binding events between analyte and receptors at the surface of a FET biosensor (bioFET) can last anywhere from a few seconds to thousands of seconds. The length of time depends on a number factors including the ionic strength of the surrounding solution; the concentration of analyte in solution; the binding affinity of the analyte to the receptors at the surface; and the mechanism by which the analyte is delivered to the surface of the bioFET [1]. It is, therefore, imperative that bioFETs can operate stably in solution for long periods of time (i.e.  1000 seconds) in order to register the response associated with analyte -receptor binding events.
When bioFETs come into contact with an aqueous solution during the course of a biosensing experiment, they experience a monotonic, temporal and relatively slow change in threshold voltage commonly referred to as drift. This change in threshold voltage is not caused by the analyte or by variations in the electrolyte composition but, in large part, by the hydration of the gate insulator layer between the electrolyte and the bioFET's channel [2]. During hydration, bonds are formed between OH groups in water and the metal/metalloid atoms in the gate insulator [3] such as SiO 2 , Si 3 N 4 , Al 2 O 3 and Ta 2 O 5 , with the reaction being facilitated by the presence of traps and buried surface sites in the insulator. Jamasb et al [4] modelled the growth of a thin, hydrated layer at the surface of the insulator limited by the dispersive transport of water molecules; and showed, theoretically and by experimentation, that the temporal growth of this hydrated layer reduces the effective capacitance of the gate insulator. This reduction occurs because the hydrated section of the insulator layer has a smaller dielectric constant tha n the underlying, unmodified, insulator layer which also decreases in thickness as, over time, the hydrated layer grows.
Consequently, the effective capacitance of the insulator, calculated as the capacitance of the hydrated layer in series with the thinner underlying, unmodified, insulator layer decreases as the hydrated layer grows.
Typically, for n-channel bioFETs a decreasing insulator capacitance would lead to a positive voltage drift. That is to say, over time, the threshold voltage would increase leading to a decrease in the output drain current. Not only does a decreasing output drain current, resulting from threshold voltage drift, dampen any change in current associated with analyte -receptor binding events; it makes it difficult to distinguish between the signal generated by the binding event and the background noise generated by the electrolyte and the bioFET itself. Due to the J o u r n a l P r e -p r o o f variance in the duration of binding events (i.e. a few seconds to thousands of seconds) between analyte and receptors, it is of critical importance that the magnitude of threshold voltage drift is kept to a minimum so as to minimise the negative effects of a diminishing drain current on the function of bioFETs.
Mitigating the effects of threshold voltage drift by the Rapid Thermal Annealing (RTA) of the insulator layer, which reduces the density of hydration facilitating defects such as buried surface sites and traps, has been shown to be quite effective [5] [6]. However, RTA is not a viable option when employing a low temperature fabrication process of 200°C [7]. This paper presents a low temperature approach to mitigating threshold voltage drift which is centered around depositing a multi-material stack of high- dielectric insulators via Plasma Enhanced Atomic Layer Deposition (PEALD). As opposed to the current bioFET design and fabrication standard, which is to deposit a single material high- dielectric insulator layer, a stack of appropriately chosen high- dielectric insulators will have a larger effective capacitance making it a more potent transducer [8]. More importantly, the effects of hydration are less pronounced on the stack than the single material insulator layer. This is because the change in the effective capacitance, resulting from of the presence of the hydrated layer, is smaller for the stack than for the single material insulator layer where, the effective capacitance is calculated as the capacitance of the hydrated layer in series with the underlying, unmodified, insulator layers. As a result, the bioFET with a stack insulator will experience a smaller threshold voltage drift and drain current shift than the bioFET with a single material insulator layer.

Fabrication and Experimentation
To demonstrate this phenomenon, a novel dry etch lift off technique [9] was used to fabricate two devices comprised of 512 Zinc Oxide (ZnO) Nanowire Field effect transistor (NWFETs) arrays ( Figure 1(a)) at 150°C. ZnO in its wurtzite form is naturally a n-type semiconductor [ [13] it does suffer from non-ideal effects namely, hysteresis phenomenon [14]. Al 2 O 3 has been shown to have a stronger resistance to these non-ideal effects [15] than HfO 2 and is also known to be more compatible with 3-Aminopropyltriethoxysilane (APTES), a commonly used cross linker molecule, thus a greater number of biomolecules can be immobilised at the surface of an Al 2 O 3 insulator than a HfO 2 insulator [16]. It is for these reasons that Al 2 O 3 was used as the single material insulator layer and the sandwich configuration was adopted for the stack insulator layer, in order to minimise the amount of HfO 2 used.  , [18], [19]. I D V G sweeps were recorded before and after both devices were wetted with PBS, in dry conditions. In the subsequent hour, while each device was wetted with PBS, the drain current was J o u r n a l P r e -p r o o f Journal Pre-proof recorded. For this current-time wet measurement, both devices were biased in the subthreshold region with a gate voltage of 4 V and a drain voltage of 2 V. The devices were biased in the subthreshold region as it has been shown to be the most sensitive region of bioFET operation for concentration related sensing [20]. That is to say, bioFETs are most sensitive to changes in analyte concentration when operating in this region; and as these are the kinds of future sensing experiments we aim to be conducting, biasing in the subthreshold region was deemed to be expedient. Moreover it enables the NWFETs to be employed as low power devices.   NWFETs obtained after TE Buffer was allowed to evaporate (steps 4 and 11 in Figure 2(c)).
As seen in the I D V G sweep of the same device, taken after the wet measurement Figure 3   The results obtained from the miDNA sensing experiment show a marked difference between the signal produced by the buffer solution and the signal produced by the miDNA in the buffer solution, in all drain current -time measurement phases of the investigation. The electric field of the negative charge on the miDNA strands, which physisorbed to the surface of the stack insulator, biases the ZnO NWFETs and causes its drain current to decrease as seen in Figure 4.
There was at most a 27 nA and a 36 nA difference between the output signals shown in Figure 4(a) and Figure 4(b), respectively, over the course of the measurement. This difference was most pronounced when comparing output currents when the buffer solution is evaporated Figure 4(c).

Journal Pre-proof
There was at least a 78 nA difference between the signal produced by the evaporated buffer and signal produced by evaporated buffer with miDNA-21 over a 7200 second period, which was caused by the presence of miDNA strands. It should be noted that rehydration was a precautionary measure taken to identify if the evaporation of the buffer caused an aberration in the output signals (signals overlapping) or whether they followed the same trend as seen with the other measurements.
These results demonstrate the potential of the ZnO NWFETs with stack insulator as a bioFET. Moreover, the minimal threshold voltage drift and drain current shift seen in the previous investigation highlights the enhanced stability the stack insulator gives the ZnO NWFETs, and this will be extremely beneficial in future biosensing applications. While a number of bioFET sensing experiments report contrived output metrics such as percentage change in conductance and normalized responses [25] [27] [28], probably due to the lack of a strong, stable and distinguishable output drain current; the ZnO NWFETs with a stack insulator used in these experiments do not experience any of these issues and as such, one can easily follow how the raw output signal (drain current) varies during an experiment without needing to process the output signal. This will, in the long run, make it considerably easier for non-technical end users to use these bioFETs.