Fabrication
The fabrication of nanoelectrode arrays started with the fabrication of metal pads connecting the nanoelectrodes. The metal pads specifications for reading the output signal from nanoelectrodes were matched explicitly to a 60-electrode microelectrode array chip of multichannelsystems (MCS GmbH). It consists of 60 individually addressable square metal pads of 2.2 mm x 2.2 mm separated by 150 µm(see supplementary information figure S2). In the future, it should be possible to obtain a signal from nanoelectrodes through these metal pads using the MEA2100 overhead stage (see supplementary information figure S2).
In our 3-D NEA, the total area written by EBL exceeds 5 mm2, and the dimension of structure varies from 20 µm down to tens of nanometers. Therefore, we needed to achieve sufficiently fast electron-beam lithography while maintaining high resolution. To achieve it, we divided the EBL area into three regions based on the minimum feature size that is to be written, and during EBL patterning, we used a 30 µm aperture for smaller feature sizes, whereas, for medium and large size patterns, we used 60 µm and 120 µm apertures, respectively. In conventional EBL systems, aperture sizes are directly related to the beam current, which in turn controls the writing speed of EBL, as explained in Eq. (1)[22].
$$\begin{array}{c}{T}_{singlepixel}\approx \frac{Dose*{Area}^{2}}{Beam Current} \#\left(1\right)\end{array}$$
In the above equation, \(Dose\) is defined as the clearance dose of the resist (given in terms of µC/cm2), \(Area\) is the total area to be exposed (given in terms of cm2), and the \(Beam Current\) is the current of the electron beam (given in terms of nA). Please note that in Eq. 1, the stage movement and settling time have not been taken into account, as we individually define these parameters for every aperture (i.e., beam current). Figure 1 (a) shows an after-metal lift-off optical micrograph of the total area written by EBL. Regions with smaller structures written by 30 µm and 60 µm apertures are shown in Fig. 1(b) and (c). During EBL, four alignment markers were also written (shown in a red circle in Figure S2 of supplementary information) using 120 µm aperture to align the photolithography pattern with EBL exposed area. Figure 1(d) shows the alignment of EBL patterns with photolithographically defined patterns after metal lift-off. A fully patterned wafer with gold metal electrodes on a silicon substrate is shown in Fig. 1(e).
Furthermore, to fabricate 3D nanoelectrodes, we use focused ion beam (FIB), and focused electron beam (FEB) based platinum deposition techniques. Although the platinum deposited by FEB was smoother and more conductive than the FIB deposited platinum, the backscattering from the electron beam was higher than the ion beam leading to shorting of electrodes due to deposition of a thin platinum film all-around 3D electrodes (See supplementary Figure S3). Figure 2(a) shows a scanning electron microscope (SEM) image of circular gold nano pads of 2D nanoelectrode. The platinum nanoelectrodes were deposited individually at this predefined array of circular gold pads. Figure 2(b) and (c) respectively show the top and tilted SEM images of 3D platinum nanoelectrodes deposited by FIB. The top diameters of nanoelectrodes were approximately 350 nm, whereas the bottom diameter of the nanowire was around 550 nm giving it a conical shape. The calculated filling ratio using these parameters comes out to be 0.16, which is very similar to what is required for achieving a highly interconnected neural network. Finally, a SU-8 thin film planarization technique was used to isolate these electrodes from each other, as discussed in reference[23]. A further increase in filling factor can be achieved by optimizing the FIB and FEB depositions to achieve near cylindrical nanoelectrodes.
Electrochemical Impedance Spectroscopy (EIS) Measurement: A nanoelectrode can be understood as a circuit element mediating the charge transfer from an electronic conductor to an ionic conductor, and the electrochemical impedance spectroscopy (EIS) provides information about the effectiveness and mechanism of this charge transfer, making it one of the primary measurements required to assess the quality of nanoelectrodes. Also, a neural signal is distributed over a range of frequencies corresponding to different neural processes. For example, local field potentials signifying synaptic activity are contained in the low-frequency band (i.e., approximately 1-250 Hz), whereas individual neuronal spikes fall in a relatively high-frequency range (i.e., around 500–3000 Hz).[24] This makes it essential to study the EIS over a range of frequencies, and in our case, we perform EIS measurements for 0.1Hz to 105 Hz.
To ensure that we only obtain electrochemical measurement data from 3D nanoelectrodes, we employ several levels of epoxy and polymer coatings to eliminate any contribution from silicon substrate or gold wire connecting the nanoelectrodes, as shown in Fig. 3(a). To prepare the substrate for EIS measurements, we uniformly cover the back of the silicon substrate using epoxy tape just after platinum metal deposition. It covers the back of the silicon substrate and the edges of the silicon. Thereafter, a 250 nm thick film of PMMA is spin-coated to isolate the gold wire connecting the nanoelectrodes. The choice of PMMA was based on the fact that it can be dissolved in acetone to expose the gold contact, but it does not dissolve in IPA or methanol, and therefore they can be used for cleaning substrates before measurement. Subsequently, a SU-8 layer was deposited over a small area (not covering the gold metal pads) to isolate the 3D nanoelectrodes. To remove polymers coated on the surface of platinum nanoelectrodes, we use an alternate step of oxygen plasma cleaning followed by oxide removal in a highly diluted HF (1:50 HF:Water), as detailed in reference.[23] We confirm the complete removal of SU-8 using SEM, as already shown in Fig. 2(d). Finally, the SU-8 is cured to form a hard contact between the nanoelectrodes that cannot readily dissolve in organic solvents. Each gold contact was individually exposed during the measurement of nanoelectrodes, ensuring no contribution from any other gold contact connecting the 3D nanoelectrodes. Moreover, to study nanoelectrodes' length-dependent EIS behavior, we control the thickness of SU8 by etching it for different durations in the presence of dense oxygen plasma before its curing.
Moreover, before fitting the EIS data to explain the charge transfer mechanisms, we check the measured data against the Kramers-Kronig (K-K) relation to ascertain its quality.[25] The K-K relation connects the real and imaginary parts of a complex function and is a measure of linearity, stability, and casualty of the system. The system's linearity means that the response is linear and perturbation is small; stability of the system means that it does not change with time (for the duration of measurement), whereas casualty implies that the measured response is only due to the excitation signal. The check against the K-K model is performed using a built-in command in the NOVA software (obtained from Autolab Instruments) that is based on the work presented in reference [26]. In short, during K-K model fitting, the data is fitted against a circuit that always satisfies the K-K assumptions. In NOVA software, this circuit consists of a series of RC circuits equal to the number of measured data points, and the quality of data is measured in terms of \({\chi }_{ps}^{2}\) (pseudo chi-squared fit value) given by Eq. (2).
$$\begin{array}{c}{\chi }_{ps}^{2}=\sum _{i=1}^{n}\frac{\left[{Z}_{re,i}-{Z}_{re}\left({\omega }_{i}\right)\right]-\left[{Z}_{im,i}-{Z}_{im}\left({\omega }_{i}\right)\right]}{\left|Z\left({\omega }_{i}\right)\right|} \#\left(2\right)\end{array}$$
Where, \({Z}_{re,i}\) and \({Z}_{im,i}\) are the measured real and imaginary parts of impedance, \({Z}_{re}\left({\omega }_{i}\right)\) and \({Z}_{im}\left({\omega }_{i}\right)\) are the real and imaginary parts of impedance simulated as a function of the radial frequency \({\omega }_{i}\), and \(\left|Z\left({\omega }_{i}\right)\right|\) is the vector length (absolute value) of the modeling function. The lower is the \({\chi }_{ps}^{2}\), the higher is the data quality, and vice-versa. In our case, for all presented data, the \({\chi }_{ps}^{2}\) was between 10− 3 to 10− 4, and the \({\chi }_{ps}^{2}\) reduced below 10− 5 if the data points are restricted to frequencies more than 10 Hz. The data quality in the future can further be improved by employing a 2- or 4-electrode measurement, which reduces the contribution of uncertainties induced by the reference electrode. Also, high-quality wire bonding or soldering can be investigated instead of crocodile/alligator clips for applying and reading the current through electrodes.
After confirming the quality and reliability of EIS data, we moved on to fitting the Nyquist and Bode plots. Here onwards, for the sake of discussion, we categorize our samples into samples 1–3 depending on the thickness of platinum nanowire exposed (for details, see Figure caption of Figs. 3 (b) – (d)). Figure 3(b)-(d) shows equivalent circuit models along with a figurative representation for nanoelectrodes with varying thicknesses of SU-8, whereas Fig. 4 (a)-(c) shows the corresponding fitting of Nyquist and Bode plots. Additionally, from here onwards, we use R, C, Q, RC, and RQ to respectively represent the resistor, the capacitor, the constant phase element, the capacitor in parallel to resistors, and the constant phase element in parallel to the resistor. Also, Relec denote the resistance due to electrolyte, Rct denote charge transfer resistances, Cpt denotes platinum nanoelectrode capacitance at the electrode/electrolyte interface, and Qpt indicates the constant phase element representing the non-ideal capacitive behavior of platinum nanoelectrode.
It's apparent from the fitting of Nyquist and Bode plots that there is a significant shift in the impedance spectra as a function of SU-8 etching time (i.e., the thickness of platinum nanowire). For example, for sample 1, where only the tip of platinum nanoelectrode is exposed. An SEM image of platinum nanoelectrode with only tip exposed is shown in Figures S3(e)-(f). The Nyquist and Bode plots can be fitted by a relatively simple circuit consisting of a resistor (Relec) in series connection with Rct and Cct in parallel, giving an almost perfect semi-circle in the Nyquist plot. As expected, because a very small portion of the electrode is exposed, the impedance is very high and exceeds more than 106Ω for almost all measured frequency region, and more than 108Ω for smaller frequency ranges. A high charge transfer resistance and low capacitance of ~ 260 MΩ and 3 pF signify that the impedance originates from the double charge layer formation at the nanoelectrode/electrolyte for most frequency ranges. A more clear picture is obtained by analyzing the phase vs. frequency Bode plot, which shows that Cpt is low only at sufficiently higher frequencies; however, as the frequency drops, the impedance corresponding to Cpt increases, and most current flow through the resistor.
In comparison to sample 1, sample 2 (see Fig. 3(c)) has very different EIS characteristics and requires a more complex circuit to explain the Nyquist and Bode's plots shown in Fig. 4(b). The circuit presented in Fig. 3(c) consists of Relec connected in series with one RC component and 2 RQ components. The magnitude of the capacitive element for sample 2 is several hundred nF, suggesting an increased surface area of the tip of the nanoelectrodes. At the same time, the requirement of constant phase element for fitting points to the polycrystalline and rough radial surface of platinum nanoelectrode, which is also evident in the SEM images shown in Fig. 2.[27, 28] The constant phase element is frequently used to model an imperfect capacitor, and its impedance (\({Z}_{cpe})\) is given by:
$$\begin{array}{c}{Z}_{cpe}=\frac{1}{{{Q}_{0}\bullet \left(i\bullet \omega \right)}^{n}} \#\left(3\right)\end{array}$$
In the above equation, \({Q}_{0}\) and \(n\) are frequency-dependent parameters, defined such that \({Q}_{0}=1/\left|{Z}_{cpe}\right|\) at \(\omega =1 rad/s\), and phase is always –(\(90 .n)\)° with \(0\le n\le 1\). Therefore, an ideal capacitor is when \(n\) = 1. In the current case, the parameter n for both the RQ elements is near to 1 for shorter nanoelectrodes (~ 200 nm) and decreases to 0.6–0.7 for longer nanoelectrodes (~ 500 nm). The roughness of nanoelectrode also increases the effective surface area of the nanoelectrode, which can also explain almost an exponential decrease in impedance of nanoelectrodes with a linear increase in the thickness of platinum nanoelectrodes. We also postulate that although one of the RQ elements represents the polarization of the radial surface of nanoelectrode due to applied sinusoidal potential, the other RQ element might be due to the secondary polarization of nanoelectrodes adjacent to the original electrode where the potential is applied. We measured more than five different samples, and we see similar behavior that an additional RQ element (to account for secondary polarization) is required to achieve fitting of EIS data when there are more than a few hundred nanometers of nanoelectrode is exposed. Another important clue that the second RQ element must represent secondary polarization is that the polarization resistance corresponding to secondary polarization is significantly high and is of the order of MΩ to TΩ, whereas the polarization resistance of the primary electrode is only a few kΩ. Also, increasing the number of this element significantly improves the fitting parameter of EIS data. This may be because a single lump sum RQ element is not sufficient to achieve a high-quality fit. There are several nanoelectrodes near the primary electrode (i.e., electrode to which potential is applied), and each needs to be fitted by an individual RQ element because of their nonuniformity. Nonetheless, even with 1 RC and 2RQ in series with Relec, \({\chi }^{2}\) (fitting parameter) of ~ 0.1 can be achieved. Here, we would also like to mention that except for sample 1, the fitting of EIS for all other samples significantly improved by increasing the number of RQ elements. Still, we have tried to minimize the number of fitting elements to explain the results better and give readers a clearer view of the charge transfer mechanism in these nanoelectrodes. In the future, it would be interesting to see how does the spacing between the electrodes influence the secondary polarization because so far, such a study has not been attempted.
During our fabrication, we also found cases where nanoelectrodes were shorted with each other, most probably during platinum deposition. In this case, there is a steep drop in the impedance with increasing frequency (see supplementary Figure S4(a) and (b)), which might be due to the simultaneous polarization of two electrodes. Although, even in this case, EIS can be reliably fitted with 1 RC and 2RQ in series with Relec, there are subtle differences in the Bode plot of sample 2, as should be evident by comparing Fig. 4(c) and (d), with supplementary information Figure S4 (a) and (b). For platinum deposited by electron beam, almost all nanoelectrodes showed shorting behavior as shown in supplementary information Figure S4 (c) and (d).
Sample 3 EIS data can be fitted with a very similar circuit to sample 2, except that now there is another constant phase element in series with the Relec, and 1RC and 2RQ elements, as shown in Fig. 3(d). The origin of 1RC and 2RQ elements is very similar to samples 2, whereas the origin of an additional constant phase element in series can be explained by sufficient thinning of SU-8. When SU-8 thickness is sufficiently reduced, a double layer will be formed near the electrolyte/SU8/gold contact interface, which is explained by an additional constant phase element in series.
Nevertheless, the EIS of sample 2 is very similar to that of commercial MEAs (see Fig. 5 (a)-(c)), showing the usefulness of a 3D NEA. The commercial MEAs were obtained from multichannelsystems and had the product code: 60EcoMEA-w/o. Our results indicate that even though NEAs have a small size compared to MEAs, they can still reach a very similar impedance to MEAs and offer a very similar charge transfer mechanism, as apparent from both Nyquist and Bode plots. One of the significant differences between NEAs and MEAs is the lack of secondary polarization in the case of MEAs. This is understandable considering the electrodes in the case of MEAs are significantly further apart compared to NEAs. Also, the rough top surface explains the lack of RC components and the presence of RQ elements.
Finally, one of the primary reasons for performing EIS measurements is to estimate the electrode-electrolyte interface noise, which is mainly controlled by thermal noise at frequencies higher than 10 Hz.[9, 29] The corresponding equivalent thermal noise for nanoelectrodes can be estimated by Eq. (4):
$$\begin{array}{c}{V}_{th}=\sqrt{4\bullet k\bullet T\bullet Re\left(Z\right)\bullet \nabla f} \#\left(4\right)\end{array}$$
In the above equation, \(k\) is the Boltzman’s constant, T is the absolute temperature, \(Re\left(Z\right)\) is the real component of the electrode impedance, and \(\nabla f\) is the frequency bandwidth. Figure 5(d) shows the calculated thermal noise for samples 1–3 in comparison to the commercial electrodes. As expected, thermal noise decreases with decreasing impedance, and in the NEAs, the thermal noise is directly related to the thickness of platinum nanoelectrode exposed to the electrolyte. Also, for sample 3, where almost more than half of platinum nanoelectrode is exposed, thermal noise reduces to as low as ~ 1µV. We believe such a low impedance and resulting low thermal noise, in the case of NEAs, is a result of the high surface roughness of platinum nanoelectrode deposited using a focused ion beam.