Work function control of ITO
The device structure and ideal band structure of the Schottky-type solar cell are shown in Fig. 1(a). The discrepancy in the work function (WF) between electrode A and the semiconductor produces a built-in potential, which separates the photogenerated electron-hole pairs. Once the generated carriers travel to the opposite electrode, power generation can be realized. In our previous study, it was proven that the greater the WF difference between the asymmetric electrodes, the higher the resultant PCE; this is consistent with this power generation model. Thus, in this study, the WF control of ITO was the first objective towards obtaining an optimal band structure for the Schottky-type solar cell.
To modulate the WF, different types of thin metal films were coated on ITO. After an Mx coating (Fig. 1(b, c); M is a metal and x is thickness in nm), some Mx/ITOs can maintain a high AVT of greater than 80%, such as Ni1, Ni5, Fe1, Al1, Al5, Cu1, and Ag1; this indicates their potential to be used as electrodes in TSCs. The WF of the Mx/ITO was then measured using photoelectron yield spectroscopy (PYS). It was found that the WF of the Mx/ITO varied in the range of 4.2– 5.4 eV (Fig. 1(d)). Moreover, most of them had a transparency of more than 80%, indicating that the WF of ITO was successfully modulated by maintaining the original high transparency of ITO (Fig. 1(e)). As a monolayer WS2 grown by CVD is naturally n-doped by impurities and the WF is measured to be around 4.9 eV, Mx/ITOs with WF higher and lower than 4.9 eV are promising candidates as transparent electrode with Schottky contacts (Ni1, Ni5, Cu1 and Fe1), and Ohmic-like contacts (Al5, Al1, Ag1, ITO), respectively against a monolayer WS2.
Schottky barrier height measurement and control: The purpose of controlling the WF was to control ΦB in both the contacts. Thus, it is important to understand the band structure of a real device. Here, the spatially resolved photoexcited charge-carrier mapping (SPCM) method was used to measure the potential profile of the real device24. By using this method, the detailed band structure between the Mx/ITO and TMD could be obtained under the ambient condition, which is the operating condition of the solar cell. The monolayer WS2 was used as the photoactive channel material in contact with Cu/ITO (left) and Ni/ITO (right) (Fig. 2(a) and 2(b)). The potential profile was obtained by the following process: (1) the source-drain current (IDS) was measured under laser irradiation at a specific spot, and the position dependency of IDS was obtained by scanning the laser from A to B (Fig. 2 (c)); (2) IDS was integrated from A to B (Fig. 2 (d), black line); (3) The y-axis was switched to present the electron negative potential (potential A is always zero) (Fig. 2 (d), red line), which could be considered to be the same as that of the conduction band (Fig. 2(g)) because of the simple approximation between the potential and carrier generation24.
First, we focused on the Cu/ITO side under different source-drain bias voltages (VDS) (Fig. 2(e) (i)–(v)). The negative slope of the potential near the Cu/ITO interface decreased with increasing VDS and became almost flat at 120 mV, indicating that ΦB at the Cu/ITO side was approximately 120 meV (Fig. 2(h)). Then, the slope changed from positive to negative slope at approximately 20 mV at the Ni/ITO side (Fig. 2 (f) (i)–(v)), indicating that ΦB on the Ni/ITO side was approximately 20 meV (Fig. 2(i)). Using the SPCM method, different types of Mx/ITO devices were systematically measured, and each ΦB was carefully estimated from the VDS dependency of the SPCM (Table S3). The correlation between ΦB and the WF for different Mx/ITO interfaces is displayed in Fig. 3(a). It was found that all the data points except those for Ni followed a linear tendency in that ΦB was proportional to the WF following the equation \({\varPhi }_{B}\) = SF × \({(\varPhi }_{\text{m}}-{\chi })\), where SF, \({\varPhi }_{\text{m},}\) and \({\chi }\)are Fermi level pinning (FLP) factor, the WF of the conductive material and electron affinity of the semiconductor, respectively. Through curve-fitting our data using this formula, SF was calculated to be approximately 0.25 (blue dashed line in Fig. 3(a)), which was consistent with the results for non-transparent bulk metal contact against TMDs obtained by other groups (SF = 0.1–0.3)25,26. According to this equation, ΦB of Ni should be approximately 200 meV, which is completely different from the experimental results (ΦB = 10–110 meV). It is known that metals with stronger/weaker binding energies (Eb) owing to their shorter/longer bonding distances (d) are considered to cause a stronger/weaker FLP effect27. Based on the previous theoretical calculation of the binding energy against TMDs, the Eb of Ni was obtained to be 510 meV, which was much higher than that of Cu (400 meV), Ag (350 meV), and Au (300 meV)27, thus denoting a stronger FLP effect of Ni as compared to Cu, Ag, and Au. This relatively stronger FLP effect may account for the much lower ΦB than the theoretical value for Ni.
Based on the above, it can be hypothesized that a slight increase/decrease in d (Eb) could weaken the FLP effect and increase ΦB, resulting in an improved performance of the Schottky-type solar cell. To confirm this hypothesis, a thin WO3 (a few nm) layer was inserted as an insulating oxide layer. As a result, ΦB of WO3/Cu/ITO increased up to 220 meV (Fig. S1), which was much higher than that of Cu/ITO (ΦB = 120 meV) (Fig. 3 (a)); this confirms the significant effect of WO3 insertion on ΦB. The SF of WO3/Cu/ITO was calculated to be approximately 0.56, which indicates a weakening of the FLP effect.
Because it is now possible to control ΦB by tuning the surface of the ITO electrode, the PCE of our solar cell devices could be measured. The relationship between ΦB and PCE is summarized in Fig. 3(b, c). The PCE and VOC increased with ΦB at the drain side (ΦB−drain) (Fig. 3(b)), and decreased with ΦB at the source side (ΦB−source) (Fig. 3 (c)); this suggested that the higher PCE and VOC originated from the higher ΦB−drain and lower ΦB−source. Because the drain and source electrodes were designed for carrier generation and collection electrode of the Schottky-type solar cell, these results agreed well with those of our power generation model. Among these solar cells, WO3/Cu/ITO with the highest ΦB−drain exhibited the best PCE (1.45\(\times\)10−3 ), which was 10 times higher than that of Cu/ITO (1.44\(\times\)10−4 %, and more than 1000 times that of pure ITO (Fig. 3 (d)) (Fig. S2).
Scale-up of highly transparent Schottky solar cells
Even though a very high PCE could be obtained from a small device at a µm-scale, the PT of the entire device would be considerably limited by the device size. Thus, PT is an important parameter for determining the potential of a solar cell for practical applications. As the solar cell study using TMDs is in a relatively nascent phase, PT has not been discussed till date. In this study, we attempted to increase PT to a practical level (more than 100 pW)23 by scaling up the device. However, it can be seen that scaling up by increasing the channel width and number of parallel connections cannot effectively increase PT, and may sometimes cause the PT to drop instead (Fig. S3), indicating that it is necessary to choose a suitable architectural design to scale up the TMD-based solar cells. Here, some concrete strategies adopted for the scale-up include (1) designing the structure of the unit device (UD); (2) exploring parallel connections (named as unit module A (UDM-A)); (3) investigating series connections (named as unit device module B (UDM-B)); and (4) combining parallel and series connections (named as unit device module C (UDM-C)). The optimization of the architecture of each device is vital.
As a first step in designing a unit device (UD) structure, solar cells with various widths (W) and channel lengths (Lch) were fabricated (Fig. 4 (a, b)). The performance of the Schottky-type solar cells has been mainly discussed in terms of three aspects: PT, VOC, and short circuit current (ISC). When Lch = 1 µm, PT increased with W up to 33.5 µm (Fig. 4 (c)). However, when W was larger than 33.5 µm, PT significantly decreased with W, that is, there was a threshold value of the critical width (Wth) essential to maintain a high PT. The drop in PT was mainly owing to VOC dropping (Fig. 4 (d, e)). Coincidentally, a similar tendency was observed in the devices with Lch = 2, 3, and 4 µm, with different Wth values of 81.3, 116.3 and 142.0 µm, respectively. Interestingly, there was an approximately linear relationship between Wth and Lch (Fig. 4 (f)), indicating that the aspect ratio of the device (W/Lch) was critical to designing large-scale solar cells with TMDs, and it should be lower than approximately 36. This could be explained by the decrease in the parallel resistance. The UD can be treated as a combination of several small channels connected in parallel, and each channel has a shunt resistance of Rsh(i) (i = 1,2.., n) (Fig. S4). The total shunt resistance (Rsh−total) would depend on the resistance of each channel (1/Rsh−total =\(\varSigma\)(1/Rsh (i)). If channel (i) contains a metal-like pass, Rsh(i) would be very low, resulting in a low Rsh−total and a low VOC. Candidates that would cause a low Rsh(i) are possibly impurities, such as the 1T phase of TMD or other impurities existing in TMD, which are inevitably induced by chemical vapor deposition or mechanical exfoliation28. The presence of impurities leads to the formation of band tail localized states, thereafter, percolation transport would arise in 2D disordered materials with energy variations along the current-carrying path29,30. When W increased, the possibility that the unexpected metal-like pass involved within the photoactive channel would increase, resulting in a low Rsh−total; this is known as the percolation model31,32, and can explain the existence of Wth and the linear correlation between Wth and Lch. Hereafter, we use Lch = 2 µm and W = 10 µm to avoid unexpected VOC drops to scale up the TMD-based Schottky-type solar cell.
To increase PT, we attempted to connect multiple UDs in parallel. With the increasing number of parallel connections (Npa) of UD, PT increased up to Npa = 11 (Fig. S5). However, a further increase in Npa caused a sharp drop in VOC, resulting in a lower PT; this could also be explained by the aforementioned percolation model.
Once we found a suitable structure for UD and Npa, the effect of the number of series connections (Nse) was investigated. When Nse was less than 4, PT and VOC increased (Fig. S6). When Nse was larger than 4, the PT drop mainly came from the decrease in not only VOC but also ISC, implying an excessive carrier loss owing to the longer travelling distance of the carriers.
The combination of parallel and series connections, as well as further paralleling of UDM-B (unit device module C (UDM-C)) was also investigated. It was found that PT increased by more than 106 times than that of UD upon repeatedly connecting 18,750 units of UDM-B at the cm-scale (Fig. 5(a) (i–ii) and 5(b)). As a control experiment, simply scaled up devices (Lch = 2 µm and W = 3000 µm; all parallel connections) were fabricated, in which PT did not increase even with a device area 106 times larger than that of UD (Fig. 5(a) (iii–iv) and 5(b)) (Fig. S7). These results indicate that the appropriate series–parallel composite design is of significant importance to PT optimization of TMD-based NISCs.
Finally, a similarly optimized device designed on an SiO2/Si substrate ((WO3/Cu/ITO-ITO) and architecture (UDM-C: UD × 3 parallel connections (UDM-A) × 4 series connections (UDM-B) × 18,750 parallel connections)) were fabricated on a quartz substrate (Fig. 5 (c)) (Fig. S8). As a result, the AVT and PT reached up to 79% and 420 pW, respectively (Fig. 5(d) and 5(e)), which can drive several real devices23. To the best of our knowledge, this is the first investigation on realizing an NISC with TMD, and with the PT obtained being the highest value for a solar cell using a monolayer or a few layers of TMD regardless of the AVT (Table S2).