ZnO NW arrays were successfully grown in an aqueous solution using the CBD method. Four sample dice were involved in this experiment, with each die having four patterned regions. The fabrication processes and parameters were the same for each sample. The results from the scanning electron microscopy (SEM) photograph in Figure 2 show well-aligned ZnO NW arrays (NWAs) on a Si substrate. The high degree of alignment of the NWs was a result of the ZnO seed layer quality, which depends on the oxidizing/annealing temperature of the sputtered Zn [21]. ImageJ was used to determine the NW dimensions. The average NW dimensions were measured to be a diameter of D ~0.2 µm and a length of L ~1.8 µm.

The structural properties of the as-grown ZnO NWs were also observed at room temperature with a Renishaw inVia Raman Spectroscope, UK. An excitation laser wavelength of 532 nm (using a grating of 1800 l/m and a visible laser) and a laser power of 3 mW were used to measure the Raman spectra of ZnO NWs. As shown in Figure 2c, for the CVD-grown ZnO NWs, a high optical mode E₂ (i.e., at a high wavenumber) was observed as the dominant peak [35]. The broad small peak around 331 cm⁻¹ may be attributed to second-order Raman scattering involving acoustic phonons found with CVD ZnO NWs. The position of the E₂ (high) peak at 436.9 cm⁻¹ corresponded very well to the value of 437 cm⁻¹ reported for both the CVD ZnO NWs and bulk ZnO, which can be taken as an indication that the NWs were free of strain. Furthermore, the linewidth of the measured E₂ (high) peak was ~9.6 cm⁻¹, which is close to 6 cm⁻¹ reported for high-quality CVD ZnO NWs [35]. These results demonstrate the high-quality crystallinity of the CBD nanowires of the present study.

Further measurements were performed to determine the conductivity of the Si substrate on which the NWs were grown and to be used as the conductive bottom electrode of the PENGs. As already described above, the Si substrate was doped using the diffusion of phosphorus to improve the conductivity close to its surface. Before the electrochemical capacitance–voltage (ECV) measurement, a diffused reference sample was dipped in 6–7% HF for 10 min. This was necessary to remove any oxide layer. A wafer profiler CVP21 manufactured by Dage Electronics, Kaarst, Germany, was used for the ECV doping concentration profile measurement. The ECV measurements operated in two modes, periodically switching between them. This involved the use of an electrolyte solution to create a Schottky contact for the concentration measurement in the first mode. In the second mode, the sample was etched by anodic dissolution to a defined depth (e.g., 10 nm) until the next concentration measurement started using mode 1 [36,37]. The surface area of the sample that was exposed for etching was 1.13 mm². The electrolyte used for the ECV measurement was 0.1 M of Ammonium bifluoride (ABF). A eutectic alloy of Ga/In was applied on the surface of the sample to create good ohmic contact with the probing pins. Figure 2d above shows the results of the ECV profile measurement of n⁺ Si. The highest concentration was measured to be in the range of (8.5 to 8.7) × 10¹⁹ atoms/ cm² at a depth range of 50 to 190 nm.

Testing of the PENG device without (and with a polymer coating, not shown here) was done with a Keithley SCS4200 I–V parameter analyzer from Keithley Instruments, Inc., Cleveland, USA (input impedance: >10¹³ Ω, input leakage current: 30 pA), as shown in Figure 3.

Prior to testing, the four patterned areas of the sample with a 24 × 24 mm² area were cut into individual dice. A bulk Si carrier die with Cr/Au deposited on it was used as a separate top electrode. The top and bottom electrodes, which formed Schottky and Ohmic contacts, respectively, were then connected to the I–V analyzer via 0.8 mm outer diameter Cu/Pb wires (Conrad Electronic SE, Berlin, Germany) connected to the electrodes’ surfaces using a conductive silver (Ag) paste. The separate top electrode was placed directly on top of the NWs (or if they were coated with a thin layer of SU-8, on top of an evaporated Au/Cr contact layer) and the output voltage was then measured via a 2-pin probe when a mass of 91.5 g was placed on it and removed from it manually and repeatedly. In this contact design, a uniform NW length was decisive to ensure electrical contact with all NWs of an array. Also, it must be noted that as the force applied to the top of the nanowires in this design was done manually, uniform compressive stress to all nanowires’ tips may not have been achieved. Again, due to the different lengths of the NWs, they may not be exposed to identical stress by the mass-loaded separate electrode on top. These effects, which will become more pronounced with large-area NWAs, are evident in Figure 4, which shows non-uniform output voltage peaks of the NG in both the compressed and released modes.

From the piezoelectric output in Figure 4 above, a maximum peak output voltage of ~7.9 mV was measured from the PENG after the release of the application of a 91.5 g mass (~0.9 N per NW, see calculation below). It can be observed that when the ZnO NWs were in contact with the top electrode and an external mass was placed on it, a negative potential was induced at the compressed side of the NWs. When the mass was released from the nanowires, a positive potential was induced. The PENG’s output voltage can be increased when the four patterns are connected in series [10], and in parallel connection, the output current can be enhanced [8,28]. When integrated on one substrate, each unit of the patterned region of the ZnO NWs can work independently.

To study the effect of a polymer matrix, arrays of ZnO NWs with the same NW dimensions as above and the same sample size of 24 × 24 mm² were spin-coated with SU-8 polymer. However, due to its high viscosity, the spin-coated SU-8 formed a thin layer of about 100 nm thick (measured with ImageJ 1.53t) on the NW array instead of embedding them, as depicted by the SEM photograph in Figure 5.

Despite the many advantages associated with SU-8 in MEMS applications, the inability of SU-8 polymer to penetrate the densely grown nanowire arrays is a major drawback. This phenomenon can be linked to the high viscosity of the SU-8 (45 cST at 25 °C) that was used in our work. Regardless of this drawback, we continued to measure the piezoelectric output of the fabricated NGs and compare the results with those from other polymers that will be utilized in our future work.

To be able to measure the piezoelectric output voltage of the fabricated NG, a separate top electrode was used as in the first experimental setup (see Figure 3a). Beforehand, a 300 nm thick Cr/Au top electrode was evaporated onto the SU-8 layer. A Si carrier die was subsequently placed on the sample and the piezoelectric output was measured. To measure the piezoelectric output, masses of 22 g, 42 g, and 91.5 g were successively put on top of the separate top electrode of the NG, generating an impulse-type compressive-mode excitation of the piezoelectric device. The height h from which the masses were dropped onto the NG determined its velocity when the surface was hit according to v = sqrt (2 × g × h) (g = 9.81 m/s²). By estimating the typical height to be h ≈ 0.5 cm, we obtained 0.31 m/s, which was slightly lower than the value of 0.8–0.9 m/s measured using a cam-based bending setup for flexible PENGs [38].

These masses were applied to each patterned region (which were of different sizes), and their output was measured using a Keithley SCS4200 I–V parameter analyzer. In the measurement process, each patterned region (P1: 4 × 10 mm², P2: 5 × 7 mm², P3: 3 × 8 mm², P4: 4 × 8 mm²) of the device was measured individually using the separate top electrode. The output voltage signals from the I–V parameter analyzer for each patterned region under repeated mass loadings of 22 g, 42 g, and 91.5 g successively are shown in Figure 6. The measured peak output voltage for each pattern was observed to be almost the same, which is an indication that the nanowires’ geometry was uniformly realized by the CBD growth process that was employed. These measurements were performed to study the output behavior of the ZnO-based PENG device when the applied force (mass) was varied. The peak amplitudes of the generated voltage varied due to the not-well-defined and low-reproducibility load application and release using manually dropped mass pieces. Nevertheless, although there was some amount of noise influence on the measured signals, the piezoelectric output voltage signals were easily distinguishable.

The maximum peak output voltages from the nanogenerator device that were recorded with varied applied forces (masses) are presented in Table 1.

To verify that the measured voltage signal was not from environmental noise but indeed from the ZnO NWs, further measurements were performed with reference samples without ZnO NWAs. The first set of samples consisted of a ~200 nm thick ZnO seed layer (SL) deposited on a Si substrate, and the second sample was a ~2 µm thick layer of SU-8 polymer on Si. Afterward, a 30/300 nm thick Cr/Au electrode was evaporated on the ZnO SL, SU-8 polymer layer, and the back side of the Si substrates for each sample. The potential difference between the Au/Cr/ZnO SL/Si/Cr/Au and Au/Cr/SU-8/Si/Cr/Au layers was measured using the I–V parameter analyzer when a mass of 91.5 g was manually applied in the repeated-compression mode. The measurement procedure was the same as with the PENGs presented earlier. The results from both measurements are presented in Figure 7a, which shows the maximum peak output voltage of ~110.3 µV for the ZnO SL sample and ~13.7 µV for the SU-8 sample. It can also be seen that there was no clear piezoelectric behavior of the measured signals, as the signals exhibited almost no negative output and mainly positive signals upon both the compression and release of the applied mass. We can therefore conclude that the output signals from our nanogenerator device were actually piezoelectric output voltages generated in the ZnO NWAs.

Again, to verify the formation of a Schottky contact on the ZnO nanowires, an I–V curve of the NG was obtained using the I–V parameter analyzer. To do this, the current through the NG was measured as the voltage was increased from −8 V to +8 V. The resultant curve in Figure 7a depicts a Schottky-like, non-Ohmic contact behavior of the device, which was assigned to the Au/Cr/SU-8/ZnO junction/interface. In a further measurement, we found an Ohmic characteristic for the ZnO/n⁺Si/Cr/Au bottom contact [39].

To estimate the performance and better understand the electric potential development in the ZnO-NW-based NG device, a single ZnO NW under a compressive force was simulated using COMSOL Multiphysics. The axial compressive force was applied along the z-axis of the NW with a <100> crystal orientation, i.e., Zn surface polarity. During the simulation, the following boundary conditions were defined: (i) the boundary load, i.e., the face where the compressive force was applied, was defined as the entire top face of the NW; (ii) the fixed constraint, which modeled the structurally blocked face of the system with respect to the x, y, and z displacements, was defined as the bottom face of the NW; and (iii) the ground, or zero electric potential plane, was also defined as the bottom face. Points (i) and (ii) were mechanical conditions and point (iii) represented an electrical condition [12,40]. In this simulation, it was assumed that the whole nanowire top surface was in contact with the applied force. We would like to emphasize here that due to the complexity of implementing a doping profile in COMSOL coupled with its simulation challenges, the built-in silicon material in the COMSOL material library was utilized instead of the phosphorus-doped contact layer on n-type silicon. The material properties of silicon, ZnO, and SU-8 that were used in our simulation are presented in Table 2, Table 3 and Table 4, respectively.

When the ZnO NW was subjected to an axial compressive force of 0.9 nN, for a single NW of length L = 1800 nm and diameter D = 200 nm, a generated potential of −6.5 mV was calculated at the top, with the zero electrical potential plane defined at the bottom of the NW as a boundary condition for the FEM. A force of 0.9 nN was applied to the single NW in accordance with the experimentally applied force (91.5 g corresponding to a gravitational force of ~0.9 N) to an array of ~1 × 10⁹ NWs (estimated from the base area of a single NW of ~3 × 10⁻⁸ mm², which was assumed to be arranged in a closed-packed array in the device area of 4 × 8 mm² = 32 mm²). The resulting areal density of ~3 × 10⁹ NWs/cm² compared well with the experimental result [21]. Nevertheless, the small deviation of the simulated potential of −6.5 mV from the measured maximum peak voltage of −7.9 mV could be attributed to a too-rough estimation of the number of NWs in the array considering the non-ideal NW arrangement visible in Figure 2. A non-uniform height distribution may lead to the application of a larger force than 0.9 nN to the NWs of a larger height. Furthermore, NW bending will have to be considered in addition to compression in a more realistic simulation due to the non-perfect vertical alignment of the ZnO NWs. In the case of coexisting piezoelectric phases with Zn and O surface polarities (i.e., if not all NWs have a Zn surface polarity as assumed in FEM), the induced negative and positive potentials, respectively, could have canceled out. Finally, the screening of the piezoelectric field by free electrons in the ZnO NWs was not considered in our FEM, i.e., doping, surface traps, etc., were not considered in our COMSOL simulations. Therefore, the presented COMSOL simulation results only give an estimate of the output voltage values, which, nevertheless, can be used to evaluate the magnitude of experimental results.

For an evaluation of the measured performance of the NG devices based on ZnO NWAs with SU-8 top layer, an array area of 3 µm × 3 µm containing 10 × 10 NWs was simulated using COMSOL Multiphysics under an axial and static compressive force. Distributed forces of 22 nN, 41 nN, and 90 nN were applied by inducing compressive stresses of 2.444 kPa, 6.625 kPa, and 10 kPa to the NWA area along the z-axis of the NWAs. A 100 nm thick SU-8 polymer layer was added on top of the NWA followed by a 30/300 nm Cr/Au layer as the top electrode. The calculations from the simulations gave generated piezoelectric potentials of −2.4 mV, −4.5 mV, and −9.9 mV, respectively, which were much smaller than the maximum measured values of the four NWA patterns but nearly agreed with their averages. The results from the simulation with applied forces of 0.9 nN for a single NW, as well as 22 nN, 42 nN, and 90 nN to an NWA of 100 NWs in a 3 × 3 µm² area, are depicted in Figure 8 below.

To mitigate the adverse effect of manually applying a force, measurements were repeatedly done for improved statistics. Mean values of measured peak output voltages and their corresponding standard deviations, which were realized from the experimental and simulation results in this work, are summarized in Table 5 below. For this, the peak values of both the polarity of the NG output measured upon both load application and release were read from Figure 6a–c and averaged. Standard deviations were calculated and taken as the measurement uncertainty [38].

From Table 5, when the applied mass was increased in steps of 22 g, 42 g, and 91.5 g, there was a corresponding output voltage increment for each patterned region of the PENG device. The trend was the same for the simulation results when the applied force was increased in steps of 22 nN, 41 nN, and 90 nN, which was to be expected.

The large uncertainties of the measured data could be attributed to the reproducibility issues of the used manual force application to the nanowire arrays, i.e., the number of nanowires in contact with the applied force may have varied. Nevertheless, within these error bounds, as is visible in Figure 9, linear dependences of the average output voltage on the applied mass can be observed for all array areas, as expected by the results of the FEM.

The peak values of open-circuit voltage V_oc-peak and short-circuit current I_sc-peak can be used to calculate the peak power P_peak = V_oc-peak × I_sc-peak of a PENG [38]. To characterize our nanogenerator device in terms of the power output, it was connected to a resistive load of 100 kΩ, which was much lower than the input impedance of the Keithley analyzer (10 TΩ). From a measurement with the NWA in pattern P1 with repeated applications/releases of a load of 91.5 g, we obtained the current output shown in Figure 9b. Here, the peak current amplitudes were much higher than the leakage current of the Keithley analyzer of <30 pA. Taking the peak value of 0.8 nA as an estimate for I_sc-peak and the corresponding V_oc-peak = 21.6 mV (Table 1), we calculated a peak output power of ~0.017 nW at a load of 0.9 N. The active volume, where the piezoelectric power was generated, was given by the considered NW array area times the sum of NW and polymer heights [38]. With the corresponding values of 0.4 cm² × 2 µm, the power density per active volume amounted to 0.22 µW/cm³.

The presented results are consistent with the literature and may be a basis for optimizing the NG output in the future. In a similar work, C. Oshman et al. [12] reported a PENG based on a ZnO NW array that was spin-coated by PDMS and contacted by sputtered Ti/Au, yielding an (open-circuit) output voltage of 3 mV under an applied compressive load of 0.451 N. This compares very well with the range of 2.6 ± 0.9 mV to 3.8 ± 0.8 mV we found with our devices at a load of 22 g (corresponding to a gravitational force of 0.22 N; see Table 5). Furthermore, the calculated peak power density per active volume of 0.22 µW/cm³ from our device is very similar to the value ~288 nW/cm³ reported by C. Opoku et al. [41] with a ZnO NW array coated by a PDMS polymer. Nevertheless, much larger values of open-circuit voltage (272 mV) and peak power (17 nW) under 3 N compressive force will be possible, which is related to the seed layer quality and load resistor adjustment [42].

The piezoelectric coefficient d₃₃ of ZnO nanowires, which is the parameter that associates polarization with stress at a constant field and strain with the electric field at constant stress was calculated in our experiment. The bulk sample piezoelectric d₃₃ measurements were carried out using the new measurement tool ESPY33 developed under project Nanowires (EMPIR 19ENG05), which enables accurate and traceable measurement of a charge that develops across a piezoelectric sample undergoing a compressive cyclic stress [43]. The ESPY33 applies a known cyclic mechanical force to the sample (device under test) between two aligned mechanical probes. A DC static preload can be applied to hold the sample in place whilst the AC force is executed. The charge is measured using electrical contact probes that also act as mechanical probes. Details of the device are found on the company website [43] and in patent GB2572334A [44]. Charge and force are recorded using digital electronics and storage scope and the effective piezoelectric charge coefficient is calculated using d₃₃ = (Q/A₁)/(F/A₂), where Q is the charge in C; F is the force in N; and A₁ and A₂ are the electrode areas and force areas, which, in the cases discussed in this paper, were equivalent. Details of this measurement method are found in a metrology handbook [45].

The sample was prepared using a silicon bottom wafer (1–10 Ωcm, thickness 275 µm) and four lithographically patterned NWA areas (NWs of 200 nm diameter, 1.8 µm in length) with vias providing top access to the bottom planar electrode. This sample was modified such that the earth (bottom electrode) via pads were connected to the ESPY33 tool’s bottom probe. The sample had four top patterns processed at various locations and a thin SU-8 polymer top infill that allowed Cr/Au top electrodes to be deposited. The device was loaded into the ESPY33 tool such that the direction of the load was along the long axes of the nanowires. The ESPY33 tool’s top probe contacted the top surface of the sample (the pattern or metalized surface) and the bottom probe contacted the backside of the silicon wafer that was electrically connected to the device’s bottom electrode.

The ESPY33 tool was previously calibrated against a standard mass force and known reference capacitor and voltage source, which enabled the forces and charges to be traceably measured. The force was measured using an instrumented load cell and charge via a transimpedance virtual earth operational amplifier. The tool was then used to measure the piezoelectric charge response of two standard reference artifacts: quartz and lithium niobate. The values recorded were within 1% of the true values for these single-crystal piezoelectric materials (quartz: d₁₁ = 2.27 ± 0.30 pC/N, lithium niobate = 20.0 ± 0.2 pC/N).

Identical pre-loads of 0.1 N were applied to the samples and a small signal force of 100 mN was applied in increments of 10 mN to examine any load dependence of the piezo response. The results for the ZnO samples are shown in Figure 10. All four patterns of the sample were examined. The theoretical value of effective ‘d₁₁ (or equivalent d₃₃)’ response along the nanowire length from the compressive force is of the order of 9.9 pC/N. Our data indicate that an effective non-zero and statistically significant ‘bulk’ piezoelectric response was measured in the sample, with nominal values of 1.9, 2.1, and 3.6 ± 0.5 pC/N for three out of the four working patterns (P1, P3, and P4).

Values of up to 3.6 ± 0.5 pC/N were found with the patterned NWAs, which had a thin (~100 nm) SU-8 top layer. Here, the larger diameter of ~200 nm and the semiconducting properties of the NWs may have determined their piezoelectric performance. This interpretation is supported by piezoresponse force microscopy (PFM) with single ZnO NWs, where the effective piezoelectric coefficient d₃₃^eff was observed to decrease with the NW diameter and SL thickness [46]. For a diameter and SL thickness of ~210 nm and 150 nm, respectively, which are close to those of the present study (200 nm and 100 nm, respectively), d₃₃^eff was between 2.9 pm/V and 3.8 pm/V, i.e., in reasonable agreement with our findings.