Printable organic thin film transistors for glucose detection incorporating inkjet-printing of the enzyme recognition element

The demand for improvement in glucose sensing technology has increased in recent years as the incidence of type 2 diabetes mellitus (T2DM) rises worldwide.¹ The prevalence of T2DM has more than doubled across the world in recent decades, with almost 350 × 10⁶ adult sufferers in 2010¹ and that number expected to rise by approximately another 100 × 10⁶ by 2030.² Although existing blood glucose level (BGL) sampling and measuring methods used by diabetics are reasonably accurate, they typically require regular blood samples, which can be painful and inconvenient to carry out. As such, the development of a non-invasive BGL measurement for diabetes sufferers is urgently required. Monitoring the BGL of a subject through the sampling of the salivary glucose rather than blood would be a much less invasive process. It has long been recognized that the concentration of glucose in saliva is correlated with BGL, albeit at much lower concentrations.^3–6 Indeed, a very recent meta-analysis of the peer-reviewed literature in this area by Mascarenhas et al. has demonstrated that there is a definite and significant correlation between salivary glucose concentration and associated glycemia/HbA1c values.⁷ 

Enzymatic based sensors for the detection of glucose have been widely studied since the mid-1980s.⁸ During the enzymatic oxidation of glucose by glucose oxidase (GOX), H₂O₂ is liberated (Eq. (1)) and subsequently electrochemically decomposed into protons (Eq. (2))

In these early studies, field effect transistors (based on silicon MOSFET technology) were fabricated with GOX immobilized in a gel gate electrode and were used to detect changes in surface pH that were correlated with the concentration of glucose in solution.⁸ 

More recently, solution processable sensors based on organic semiconductors are attracting much attention in the rapidly emerging field of organic electronics.⁹ Furthermore, direct inkjet-printing of enzymes is also starting to be developed as researchers look for ways to take advantage of the advantages associated with inkjet-printing for sensor applications such as rapid “drop on demand” prototyping and digital design capabilities. For example, in 2011, Yun and co-workers reported the fabrication of glucose-sensing structures containing both GOX and horseradish peroxidase (HRP) mixed with the widely used polymeric conductor poly(3,4-ethylenedioxythiophene)-poly (styrene sulfonic acid) (PEDOT:PSS).¹⁰ An ink containing these three materials was inkjet-printed onto an indium tin oxide (ITO) coated substrate, which was then used as the working electrode in a three terminal electrochemical cell. Glucose concentrations down to 590 μM were detected using cyclic voltammetry and chronoamperometry. More recently, Weng et al. fabricated inkjet-printed enzymatic sensors using GOX (to sense glucose) and HRP (to sense H₂O₂) printed from an ink containing a dispersion of polypyrrole nanoparticles.¹¹ Again, the peak current observed during cyclic voltammetry experiments was used as a calibration parameter with the enzymes present on the working electrode. Also in 2014, Zhang et al. presented enzymatic sensors using GOX in combination with HRP in which the enzymes were deposited onto paper using piezoelectric inkjet-printing.¹² These sensors exhibited a change in color dependent on the concentration of glucose in a given sample due to the HRP-induced oxidation of 2,2′-azino-bis (3-ethylbenzothiazoline-6-sulfonate) (ABTS) in to ABTS^▪+.

Recently, we have reported enzymatic glucose sensors based on a modified organic thin film transistor (OTFT) architecture incorporating GOX as the recognition element. These sensors respond to glucose concentrations in the range which occurs in saliva.⁷ Glucose sensing in these devices involves the diffusion of protons (generated via the enzymatic oxidation of glucose) to, and subsequent doping of, the poly(3-hexythiophene) (P3HT) transistor channel. The diffusion step is the key determinant of response time in these devices.¹³ 

In this study, we probe the mechanisms present in OTFT glucose sensors by studying device behavior with changing device architecture. By optimizing the device architecture, we are able to demonstrate a five-fold improvement in response time. By modelling the diffusion behavior of the devices, we show that there is an initial rapid de-doping process that is followed by the doping of the P3HT layer. Moreover, the modelling data indicate that the doping region is not located at the upper P3HT interface but instead is closer to the source and drain electrodes. Finally, we demonstrate that inkjet-printing is a promising method to deposit GOX in these types of devices which both improves fabrication consistency and provides an avenue for entirely inkjet-printed sensors in the future.

The characteristic response of devices with our “standard” sensor architecture—comprising ITO source drain electrodes, a P3HT semiconductor layer, a poly (4-vinylphenol) (PVP) dielectric layer, and a gate electrode consisting of the ion exchange membrane Nafion mixed with GOX—to glucose analyte solutions has been reported previously.¹³ Subsequent investigations have revealed that removing the PVP layer serves to improve the response time of the sensor presumably by reducing the distance over which moieties diffusing from the Nafion:GOX layer to the P3HT layer have to travel (for more details on this experiment, see supplementary material¹⁴). As a result, devices used in subsequent experiments described hereafter in this letter are prepared without the PVP layer used previously.

Figure 1 shows the effect of varying the thickness of the P3HT layer of the sensors from 108 nm to 22 nm upon the rate of sensor response, represented by drain current (I_D) as a function of time. For the sake of data clarity, I_D is presented as a ratio of its stabilized level prior to analyte addition minus the minimum value of this ratio (which occurs soon after analyte addition)—a quantity referred to hereafter as “adjusted I_D.” As the thickness of the P3HT channel is reduced, the response time of the fast de-doping and slow doping processes both reduce, consistent with diffusion processes that traverse a reduced layer thickness. Both processes can be modelled by one dimensional solutions to Fick's second law of diffusion, n(x,t) = n₀ erfc{x/[2/(D_eff t)^0.5]} = n₀ erfc(A/t), where A (comprising x (diffusion distance) and D_eff (effective diffusion constant)) and n₀ (initial concentration) are treated as fitting constants with the fit solution shown as dashed lines in Figures 1(b) and 1(c). Fitting the fast (A_fast) and slow (A_slow) responses to Ficks law provides an estimate for the value of A for each process where A = x/[2(D_eff t)^0.5]. The fitted values for A (for both the slow and fast processes) decrease with decreasing P3HT layer thickness consistent with more rapid diffusive transport. Moreover, the decrease in A_slow for the very thinnest P3HT layer is more dramatic than the corresponding decrease in A_fast, indicating that there is a difference in the two processes for P3HT thicknesses below ∼36 nm.

A direct measurement of the slow and fast diffusion constants for each of the different layer thicknesses is difficult, since both the diffusion constant and the layer thickness are part of the fitting constant, A. However, it seems reasonable to assume that the fast dedoping and slow doping processes have to diffuse across the same layer thickness for a given OTFT architecture. As such, A_fast/A_slow = [D_slow/D_fast]^0.5, where A_slow and A_fast are the fitting parameters, and D_slow and D_fast are the effective diffusion constants, for the slow and fast processes, respectively. Thus, the ratio A_fast/A_slow as a function of changing layer thickness should only depend upon the ratio of the diffusion constants of the two processes. Figure 2 shows the variation of A_fast/A_slow as a function of the different device layer thicknesses and includes the data for both the devices with PVP (layer thickness 400 nm) and the devices of varying P3HT thickness without a PVP layer. The A_fast/A_slow ratio is relatively invariant across both the OTFT device with a PVP layer and the devices without a PVP layer but with P3HT layers that are at least 36 nm thick. This observation supports the assertion that, for each OTFT architecture, the fast dedoping and slow doping process diffuse across the same effective distance and indicates that the ratio of the corresponding diffusion constants is invariant. However, when the P3HT thickness drops below 36 nm, there is an abrupt increase in the A_fast/A_slow ratio. An examination of Figure 2(b) reveals that this abrupt increase is dominated by a critical change in the effective diffusion constant for the slow doping process for P3HT thicknesses below 36 nm.

The data in Figure 2(a) are consistent with the presence of some thickness of P3HT which only serves to slow the diffusion rate of protons to the active (doping) region of the channel. As such, when the P3HT thickness drops below a critical value (∼36 nm) then there is no diffusive barrier to protons accessing the doping region of the device. Indeed, this hypothesis suggests that any further reduction of the P3HT layer thickness should not affect the diffusion rate of either the fast or slow process (since there is no diffusive barrier) and reducing the P3HT thickness now merely alters the absolute number of doping sites and therefore the current in the channel.

In order to test this hypothesis, the P3HT layer thickness was further reduced from 22 nm by decreasing the concentration of the P3HT solution (from 5 mg mL⁻¹ to 2 mg mL⁻¹). The resulting P3HT layers exhibited regions of incomplete coverage and hence the layer thickness (∼9 nm) could only be estimated from the P3HT loading. Notwithstanding, functional devices could be prepared. Figure 2(b) compares I_D for devices with a 22 nm and a 9 nm thick P3HT layer, as a function of time. Figure 2(b) shows that the device response is lower for the thinner P3HT layer and that I_D at saturation has reduced from a value of ∼1.2 to a value of ∼0.5 on this scale, corresponding to a 42% reduction in current that is quantitatively consistent with the reduced P3HT thickness. Fitting the current to the Fick's law reveals that the only fitted parameter that has to be changed is that of n₀ (which governs the absolute magnitude of the response), whereas the fitted value of A is constant for both fast and slow processes for both of these P3HT thicknesses. Consequently, the data are consistent with a doping region which, for thicker P3HT layers, does not lie at the interface between the P3HT and the Nafion layer but instead lies at some small distance from the source and drain electrodes and is overlayed by undoped P3HT through which protons must diffuse (Figure 2(a), inset). As the P3HT layer is reduced, it reaches a critical thickness at which the Nafion layer interface is coincident with the doping region and subsequent decreases in P3HT thickness serve only to decrease the size of the doping region and hence the observed current.

Figure 3 shows the effect of changing gate voltage (V_GS) upon device performance after addition of 30 mM glucose solution. Figure 3(a) shows that as the V_GS is made more positive (changed from −1.0 V to −0.3 V) so the value of adjusted I_D at saturation increases (from ∼0.6 to ∼1.75 on this scale). Figure 3(b) indicates that this rise in I_D is associated with a change in the polarity of the net current flowing from gate to source. These results are consistent with a change in the net electric field experienced by the charge carriers (protons). When V_GS = −1.0 V, there is a net electric field from source to gate (since both gate and drain electrodes are held at −1.0 V relative to the source electrode), whereas when V_GS = −0.3 V, there is a net electric field from gate to drain. These electric forces act in addition to the diffusion gradient within the device serving to either retard (V_GS = −1.0 V) or enhance (V_GS = −0.3 V) protonic doping of the channel, as shown schematically in Figure 3(c). This change in V_GS increases the sensitivity of the device response and consequently subsequent measurements presented in this paper have been conducted with a gate bias voltage of −0.3 V.

Despite their sensitivity to glucose analytes of a wide range of concentrations,¹³ the glucose sensors reported previously showed significant device-to-device variation which we partially attribute to the drop-cast Nafion:GOX layer. In particular, variations in layer thickness and aggregation of the GOX lead to differences in device performance. In order to improve device consistency, Nafion was spin coated as separate layer and GOX was subsequently inkjet-printed onto the Nafion film. Importantly, inkjet-printing of the enzyme proved to be a very reliable process without any of the aggregation problems associated with drop-casting the Nafion:GOX mixture. We can attribute this improvement to both the higher solubility of GOX in water compared to the solvent mixture used in the drop-cast case as well as the slower deposition rate. (Supplementary material includes optical microscopy of inkjet-printed GOX on spin-coated Nafion compared with drop-cast Nafion:GOX mixture, as well as the electrical response of these devices to different concentrations of glucose analyte.¹⁴) In devices with inkjet-printed GOX, the enzymatic activity of the devices remained intact, and the response time of the devices improved (conceivably due to diffusion through the thinner spin-coated Nafion layer being a faster process than through the thicker, drop-cast layer as well as the enzyme being more readily available to the analyte since it is more evenly dispersed).

As a consequence of the faster I_D rise time in devices with a spin-coated Nafion layer and inkjet-printed enzyme, the calibration parameter, χ, introduced in our previous work, can now be calculated from t = 0 to 500 s (instead of 800 s used previously¹³) and thus is now defined as

Figure 4 shows the calibration curves for inkjet-printed OTFT sensor devices with P3HT thicknesses varying from 22 nm to 390 nm. In general, as the P3HT thickness becomes thicker, the gradient of the calibration curve decreases; indicating a reduced response to the GOX analyte; in agreement with the results for devices with the drop-cast Nafion:GOX layer. Interestingly, although the thinner P3HT devices have a faster response, the variability in the device response to the glucose analyte solution also increases, resulting in a reduced linearity of the calibration curve. Overall, therefore, it would appear that currently an intermediate P3HT thickness of 75–100 nm provides the optimal balance of glucose sensitivity and reproducibility, and as illustrated in Figures 4(c) and 4(d) we can expect an approximately linear response between glucose concentration and χ for glucose concentrations between 100 μM and 100 mM. Further work is currently underway in this laboratory to improve the reproducibility of printing of thin films of P3HT.

In this letter, the effect of changing device architecture and active layer thickness upon the performance of OTFT glucose sensors has been investigated. In particular, we show that streamlining the device architecture by eliminating the PVP layer and reducing the P3HT active layer thickness both act to decrease the response time of the biosensors. The fitted results are consistent with a diffusion based model of operation, whereby a rapid de-doping process is followed by protonic doping of the P3HT channel that is proportional to the concentration of glucose analyte presented to the device.

We also demonstrate that the optimized OTFT architecture can be fabricated with an inkjet-printed GOX layer at the gate electrode, which dramatically reduces the surface roughness of the deposited film. The successful inkjet-printed of the glucose-sensing enzyme provides a pathway for the realization of entirely inkjet-printed devices in the near future and, indeed, early results in these laboratories for all-printed devices of this type are encouraging.

This work was performed in part at the Materials node of the Australian National Fabrication Facility, which is a company established under the National Collaborative Research Infrastructure Strategy to provide nano and microfabrication facilities for Australia's researchers.