Superior performance of V2O5 as hole selective contact over other transition metal oxides in silicon heterojunction solar cells

UPCommons Portal del coneixement obert de la UPC http://upcommons.upc.edu/e-prints Aquesta és una còpia de la versió draft d'un article publicat a Solar energy materials and solar cells http://hdl.handle.net/2117/104004 Almora, O, Gerling Sarabia, L., Voz, C., Alcubilla, R., Puigdollers, J. Garcia-Belmonte, G. Superior performance of V2O5 as hole selective contact over other transition metal oxides in silicon heterojunction solar cells. "Solar energy materials and solar cells", 1 Agost 2017, vol. 168, p. 221-226. DOI 10.1016/j.solmat.2017.04.042 © 2017. This manuscript version is made available under the CC-BY-NC-ND 4.0 license http://creativecommons.org/licenses/by-nc-nd/4.0/ Superior Performance of V2O5 as Hole Selective Contact over other Transition Metal Oxides in Silicon Heterojunction Solar Cells Osbel Almora1, Luis G. Gerling2,3, Cristóbal Voz2, Ramón Alcubilla2,3, Joaquim Puigdollers2,3,*, and Germà Garcia-Belmonte1,* 1 Institute of Advanced Materials (INAM), Universitat Jaume I, 12006 Castelló, Spain 2 Electronic Engineering Department, Universitat Politècnica de Catalunya, Jordi Girona 1–3, Barcelona 08034, Spain 3 Centre de Recerca en Nanoenginyeria (CrNE), Pascual Vila 15, Barcelona 08028, Spain Abstract Transition metal oxides (TMOs) have recently been proved to efficiently serve as hole-selective contacts in crystalline silicon (c-Si) heterojunction solar cells. In the present work, two TMO/c-Si heterojunctions are explored using MoO3 (reference) and V2O5 as an alternative candidate. It has been found that V2O5 devices present larger (16% improvement) power conversion efficiency mainly due to their higher open-circuit voltage. While V2O5/c-Si devices with textured front surfaces exhibit larger short-circuit currents, it is also observed that flat solar cell architectures allow for passivation of the V2O5/n-Si interface, giving significant carrier lifetimes of 200 µ s (equivalent to a surface recombination velocity of Seff ~140 cm s-1) as derived from impedance analysis. As a consequence, a significant open-circuit voltage of 662 mV is achieved. It is found that, at the TMO/c-Si contact, a TMO work function enhancement ΔΦTMO occurs during the heterojunction formation with its consequent dipole layer enlargement Δ’=Δ+ΔΦTMO. Our results provide new insights into the TMO/c-Si contact energetics, carrier transport across the interface and surface recombination allowing for further understanding of the nature of TMO/c-Si heterojunctions. Keywords: Transition metal oxides, silicon solar cells, impedance spectroscopy, passivation, minority carrier lifetime. *Corresponding authors: J. Puigdollers, ([email protected]) tel.:+34 93 4011002 G. Garcia-Belmonte, ([email protected]) tel.: +34 964 387538 3 May 2017 2 1. Introduction During the last decade, crystalline silicon (c-Si) heterojunction solar cells incorporating thin films of hydrogenated amorphous silicon (a-Si:H) have become the state-of-the-art photovoltaic technology, achieving record efficiencies above 26% [1]. The key for the success of this technology lies on the superior surface passivation provided by very thin layers (<5 nm) of intrinsic a-Si:H, also allowing for carrier conduction with minimal resistive losses. However, performance limitations caused by the relatively high optical absorption of a-Si:H [2] and the highly recombining n/p doped layers [3] (which act as electron/hole selective contacts) have led to investigate novel heterojunction concepts between c-Si and dopant-free highly-transparent transition metal oxides (TMOs) [4, 5]. Originally introduced in organic electronics as electron and hole transport layers, TMOs are large band gap (Eg >3 eV) semiconductors with a wide variety of work functions (ΦTMO~3–7 eV) and conductivities (from insulating to metallic-like), providing great flexibility when used as electron- or hole-selective contact materials [6]. Additionally, they can be deposited at low temperature (T <200ºC) or by solutionprocessing methods [7], increasing the potential for process simplification and cost reduction. Until now, hole-selective MoO3 has been the study-case TMO material alongside n-type crystalline silicon (n-Si), reaching a power conversion efficiency (PCE) of 22.5% in a MoO3/(i)a-Si:H/n-Si configuration [8], where the intrinsic (i)a-Si:H chemically passivates the silicon surface while MoO3 provides the hole-selectivity. Also rear MoOx contacts with partial contact areas have been reported with an efficiency of 20.4% [9]. As an alternative, simpler structures comprising TMOs in direct contact with n-Si have also been proposed, achieving efficiencies between 12.5 and 18.4% [10-12]. The unique behavior of MoO3 and other similar TMOs (V2O5, WO3) is explained by their electronic configuration and large work function values (ΦTMO>5.0 eV), which upon Fermi level ( F E ) alignment with n-Si (Φn-Si~4.2 eV) induces a potential barrier (band bending) [10, 13]. This is believed to result in the formation of an inversion (p+) layer upon n-Si where photogenerated holes are collected and then extracted across the TMO/n-Si interface. Recent reports have also proposed V2O5 as an interesting alternative to MoO3, indicating that larger open-circuit voltage values (higher holeselectivity) can be achieved for V2O5–based devices with and without the inclusion of passivating (i)a-Si:H interlayers [10-12]. In the present paper, Impedance Spectroscopy (IS) measurements were used to compare the performance of MoO3 and V2O5 as hole-selective contacts in n-Si solar cells, showing that V2O5–based solar cells perform better due to the presence of higher built-in voltages. Additionally, temperature-dependent measurements were used to calculate barrier heights across the interface, giving further details about the energetics of TMO/n-Si heterojunctions. 3 2. Experimental Devices Fabrication: Solar cells were fabricated from n-type 2 Ω⋅cm resistivity wafers (~2.3×1015 cm-3 dopant concentration) made from float zone monocrystalline (100 orientation) material. The use of such high quality silicon allows for a very high bulk lifetime ( bulk τ ~3 ms under high injection conditions), ensuring almost all recombination effects are confined to the surfaces. Solar cells labeled t-MoO3 and t- V2O5 were subjected to random texturization of the front surface by alkaline etching (~265 µm final wafer thickness), while cells named f-V2O5 were processed as purchased (flat polished finish, ~280 µm wafer thickness). After a standard RCA cleaning 1% HF dip, all substrates were then loaded into a plasma-enhanced chemical vapor deposition (PECVD) system to deposit on the rear side an electron-selective contact consisting of an intrinsic/n-type a-SiCx:H stack (5 nm/15 nm, x~0.2). Subsequently, two rear contact strategies were used: 1) t-MoO3 and t-V2O5 cells had an a-SiCx:H back-reflector (80 nm, x~1) deposited by PECVD which was then laser-fired to obtain an array of locallydiffused point contacts (0.5% contacted area) [14]; 2) f-V2O5 cells had an indium-tin- oxide (ITO) back-reflector/electrode deposited by RF magnetron sputtering (80 nm, 1.3×10-3 mbar Ar pressure). As for the front hole-selective contacts, 20 nm thick TMO films were thermally evaporated from powdered V2O5/MoO3 sources (>99.99% purity, Sigma Aldrich) at ~8×10-6 mbar. For the front textured sample, the deposition time was ~1.7 times longer than the polished ones, in order to compensate for the increase in surface area. The deposition rate was ~0.2 Å/s, as controlled by quartz micro-balance, while the substrate remained at room temperature during the process. After a brief air exposure, an ITO front electrode/antireflective layer (80 nm) was also deposited. At this point in the process, QSSPC measurements were performed in order to determine the carrier lifetime of the solar cell precursor (figure S5). After lithographic patterning of 1 cm2 active cell areas, a front-contact Ag grid (4.3% shadow losses) was thermally evaporated by use of a shadow mask, while the back-contact metallization was done by e-beam evaporation of Ti/Al (t-MoO3 and t-V2O5) or Ag evaporation (f-V2O5). Characterizations: The IS measurements at oc V (Figures S7-S9) were carried out using an Autolab PGSTAT-30 potentiostat in the frequency range between 100 mHz and 1 MHz, being the AC perturbation of 10 mV. The spectra were recorded in open-circuit conditions under varying illumination up to 150 mW cm-2 (XE 300W Newport 6258). For achieving this, a bias voltage which corresponds to oc V was applied, hence ensuring a more homogeneous distribution of excess carriers by suppressing DC current. For practical reasons, an extra loop element was included in series connection with the circuit (Figure 5a), and at some spectra a constant phase element was considered instead of a second capacitor. For the IS measurements at short-circuit in the dark and at different temperatures (Figure S3), a Gamry Reference 3000 potentiostat/galvanostat/ZRA (same AC perturbation and frequency range as above) was employed joining the Novocontrol Quatro Cryosystem. 4 3. Results and discussion 3.1 Device structures and performance The structure of the three representative devices studied here is presented in the sketches of Figure 1, including the specifications about composition and thickness of each layer. This set of configurations will allow us to properly compare performances among V2O5– and MoO3–based Si heterojunction solar cells. The first structure, labeled t-MoO3, uses a 20 nm-thick layer of MoO3 as front hole-selective contact deposited on a textured surface, while the rear electron-selective contact is formed by a (i/n+)a- SiCx~0.2:H stack that was locally-diffused by laser firing (Figure 1a). On the other hand, there are two structures with V2O5 as hole selective contact called t-V2O5 and f-V2O5 (see Figures 1b and 1c, respectively). While t-V2O5 presents the same structure as t- MoO3, f-V2O5 exhibits a flat superposition of layers, increasing the passivation quality provided by V2O5. Moreover, f-V2O5 uses indium tin oxide (ITO) as a rear electrode, given that laser firing provides a good ohmic contact at the expense of reduced passivation quality. More details about the devices fabrication and morphological features can be found in previous works [10, 15], the experimental section and the Supplementary Information (SI). Figure 1: Sketches of the structures for the three devices which were studied in this work, as indicated. Examples of current density-voltage (J-V) curves resulting from representative devices measured under illumination are shown in Figure 2. The corresponding performance parameters such as open-circuit voltage ( oc V ), short-circuit current ( sc J ), fill factor (FF) and PCE are summarized in Table 1. Also histograms in Figure S1 present the parameter distribution for the complete set of studied samples. In this respect, devices with V2O5 show significantly higher PCE than those comprising MoO3 layers. Given that only minor improvements in the photocurrent generation and no clear trend in the FF are observed, the superiority of V2O5–based solar cells is directly connected to the enhancement in the output oc V in comparison with MoO3. Illustratively, among the samples with surface texturing (t-V2O5 and t-MoO3), t-V2O5 has a larger sc J while the flat sample (f-V2O5) has practically the same sc J as t-MoO3. Additionally, the 5 oc V differences are apparent with significantly larger values achieved by the V2O5 devices. Our findings then indicate that the V2O5/n-Si heterojunction makes up a superior contact compared to MoO3/n-Si. Figure 2: Experimental current density-voltage curves measured under 100 mW∙cm-2 of AM1.5G light spectrum for different samples as indicated. Table 1: Photovoltaic parameters corresponding to the J-V curves in Figure 2 and calculated parameters from the analysis of the MS plots in Figure 3. The depletion layer width w0 is presented at zero bias and, in parenthesis, the value calculated for the real semiconductor doping ( 15 2.3 10× cm-3). This allows comparing deviations caused by geometric effects on the capacitance area normalization. Devices Voc (mV) Jsc (mA cm-2) FF (%) PCE (%) Vbi (mV) ND ( 15 10× cm-3) w0 (nm) t-MoO3 563 33.0 72.1 13.4 599 4.37 406 (560) t-V2O5 605 34.5 74.7 15.6 617 6.86 330 (569) f-V2O5 662 32.5 70.6 15.2 712 2.10 644 (614) 3.2 TMO/n-Si heterojunction analysis. The formation of the heterojunction in the TMO/n-Si devices occurs by band bending of the n-Si energy levels producing a built-in potential bi V . However, before entering into a discussion of the energy band diagram, it is required to explore the depletion region features in order to estimate bi V values exhibited by the devices. Therefore, capacitance measurements as a function of DC applied voltage in the dark were done. Figure 3 exhibits such results in the typical Mott-Schottky (MS) plot representation, i.e. 2()CV − . The linear decrease of MS plots, as the applied voltage is swept from reverse bias to near flat band condition, evidences a constant charge density profile. Subsequently, the neutrality of charge between the negative ionized defects, presumably 6 at the thin TMO side, and the positive Si ionized impurities, allow us to assume the one side abrupt junction approximation [16]. Thus the depletion layer width w at the n-type silicon bulk can be calculated as a function of the applied DC bias V as 0 22 B bi D kT w VV qN q εε  = −−   (1) where q is the elementary charge, B k is the Boltzmann constant, T is the absolute temperature, 0 ε is the vacuum permittivity, ε is the relative dielectric constant (11.9 for Si) [16, 17], and D N is the donor density (doping concentration with negligible intrinsic defects). The 2/ B kT q term corresponds to the majority-carrier contribution in addition to the impurity concentration [16]. With this in mind, a parallel-plate capacitor model can be considered for the determination of the depletion layer capacitance per unit area as 0/ dl Cw εε = . Subsequently, by substituting equation 1 and after a few operations, the simplest expression for MS analysis is obtained as 2 0 2 2B dl bi D kT C VV qN q εε − = −−   . (2) The bi V can be obtained from the voltage intercept while D N results from the slope of the linear portion in C-2(V). By substituting these results in equation (1), w at a given applied bias can be obtained. Resulting values for bi V , ND and w0 (at V=0) are reported here for the first time for TMO/c-Si solar cells, showing the expected behavior characteristic of p-n junctions (see Table 1). Regarding the band bending, it accords well with previous experimental reports via Surface Photovoltage (SPV) and theoretical simulations for the Φn-Si range [11, 18]. ND and w0 values in Table 1 are also in good agreement with typical orders of magnitude observed in similar structures, although they do not consider effective area rectifications and/or contributions from gap trapping states. In fact, unlike the textured samples, the presence of gap defect levels is evidenced in the f-V2O5 device from its frequency dependent capacitance response [19, 20] shown in Figure S2, where the slope of the MS plot varies with the AC perturbation frequencies, even though they converge to the same bi V value. MS analysis straightforwardly gives bi V as the intercept of the linear plots with the voltages axis, irrespective of area normalization or frequency dependent capacitances [16]. Similarly to the above mentioned oc V behavior, devices with V2O5 present higher bi V than those incorporating MoO3, being f-V2O5 the structure with the larger value. Interestingly, the obtained bi V values are similar to those of standard c-Si solar cells in which 600–750 mV are distributed between the two sides of the homojunction, suggesting that TMO/n-Si heterojunction behave similarly [21, 22]. The difference here is that bi V ~ 0.6–0.7 V is restricted to the Si side, indicating that the junction achieves weak inversion at zero bias. In addition, an almost ideal linear MS plot (Figures 3 and S2) is very much less frequent in heterojunction devices like CdTe [23, 24], CIGS [25, 26], organic [27] and hybrid [28] solar cells, where the constant doping profile (regular slope) is rarely reported. Since a constant charge density profile translates into a 7 quadratic behavior of the potential within w, a quadratic bending of the energy bands can also be expected. These two elements (large bi V and constant doping) could support previous assumptions on the occurrence of a p-type inversion layer upon n-Si in the vicinity of the TMO interface (see energy-scaled band diagram of Figure 4a). Figure 3: Experimental (dots) Mott-Schottky plots and respective linear fittings (lines) under dark and room temperature conditions for the different samples, as indicated. The AC perturbation was 10 mV at 1.0 kHz. As summarized in Table ST1, the studied TMOs possess large energy bandgaps and work functions compared to silicon, and generally exhibit prominent n-type conductivity due to defect states generated from oxygen vacancies [6]. This is the reason why it has been proposed [10, 15] that a dipole layer Δ is present at the TMO/n-Si interface, in order to equilibrate the Fermi levels in the energy band diagram once the heterojunction is formed. Considering absolute values and assuming no significant changes in the TMO electron affinity after the contact is made, this can be written as -TMO n Si bi qVΦ −Φ = +∆ . (3) The spatial distribution of the work function offset is illustrated in Figure 4a-c in an illustrative typical case assuming ΦTMO ~6.7 eV, where the interfacial dipole is possibly originated by shallow O vacancies in the TMO bulk [6, 10, 29]. Here, usually reported bandgaps (Table ST1) and our calculated bi V values (Table T1) were used to sketch energy levels before and after the formation of the heterojunction, reveling two possible scenarios. In the first case, the conduction band ( C E ) bends up forming a barrier of height C E∆ of approximately ~ 25 B kT for electrons (Figure 4b), while in the second case no barrier is formed (Figure 4c). When such relative high barriers ( 10 CB E kT∆> ) are present, typically thermionic emission and/or tunneling are expected to occur [16]. Therefore, as it is known, the current density TE J follows the expression 8 2 exp[ ] exp[ ] 1 TE BB A T q qV JkT kT ϕ α  =−−   (4) where B kT is the thermal energy, A is the Richardson constant, q ϕ the distance between F E and the top of the barrier ( C qE ϕ >∆ in Figure 4b), and α is the relation between the rest mass and the effective mass ( * 0 /mm ) when thermionic emission over the barrier is the dominating mechanism. Thus, it can be proved [30] that the resistance unit surface 1 0 ( /) TE R dJ dV − = at zero applied bias follows the relation 0exp[ ] B B kq RqAT k T ϕ α = . (5) At this point it is useful to mention that when tunneling has a significant contributing role an extra term must be added to equation (4) and α can be written as a more complex expression. Furthermore, if tunneling is the dominating mechanism, 0 R cannot be easily reduced to equation (5) and the temperature should not exponentially affect 0 R . However, that situation is only expected for high doping levels above 19 3 10 cm− [30]. Conveniently, from equation (5) it is apparent that the linear fit of 0 [ /] B Ln qTR A k versus / B q kT yields an slope equaling ϕ and an intercept approaching []Ln α , allowing for testing underlying mechanisms as thermionic emission and/or tunneling across the barrier. For this purpose, J-V curves and impedance spectroscopy (IS) analyses were carried out for the textured samples (t-V2O5 and t-MoO3) in the dark and within the temperature range between 210 K and 295 K. The noisy response from f-V2O5 due to its large 0 R values prevented a reliable data processing. From the studied J-V curves (Figures S3a,b) the slope of the linear fittings around V=0 was calculated and subsequently ( ) 1 0 /R dJ dV − = obtained. As for the impedance, a Nyquist plot (negative imaginary part of impedance ''Z− versus the real component 'Z ) [31] was obtained from the IS measurements (Figures S3c,d), and fitted to the series connection of a resistance s R and a simple RC equivalent circuit (inset Figure S3c), being 0 R the resistance of the arc. The 0 R values obtained from J-V and IS measurements can be observed in Figure S3e. With these values, Figure 4d was elaborated and the barriers estimated at 16q ϕ ≈ meV for t-V2O5 and 50q ϕ ≈ meV for t-MoO3. Consequently, since such barriers are in the order of B kT and 10 10 α ≈ , thermionic emission is discarded as the main transport mechanism over the barrier. This could be interpreted implying that the energy band diagram sketched in Figure 4b with C E ∆ may occur only if tunneling is the dominating mechanism. Alternatively, it is also possible that no barrier occurs at all, as presented in the third sketch of Figure 4c. In this case, during the heterojunction formation, and due to inner bandgap defect levels, the TMO work function is augmented TMO ∆Φ as well as the dipole layer ' TMO ∆ = ∆ + ∆Φ (assuming no electron affinity change), in such a way that no interfacial barrier occurs. 15 [14] M. Colina, A.B. Morales-Vilches, C. Voz, I. Martín, P.R. Ortega, R. Alcubilla, Low Surface Recombination in Silicon-Heterojunction Solar Cells With Rear Laser- Fired Contacts From Aluminum Foils, IEEE J. Photovoltaics, 5 (2015) 805-811. [15] L.G. Gerling, C. Voz, R. Alcubilla, J. Puigdollers, Origin of Passivation in Hole- Selective Transition Metal Oxides for Crystalline Silicon Heterojunction Solar Cells, J. Mater. Res., 32 (2016) 260-268. [16] S.M. Sze, K.K. Ng, Physics of Semiconductor Devices, 3rd ed., John Wiley & Sons, Hoboken, New Jersey, USA, 2007. [17] G.A. Samara, Temperature and Pressure Dependences of the Dielectric Constants of Semiconductors, Phys. Rev. B, 27 (1983) 3494-3505. [18] M. Bivour, B. Macco, J. Temmler, W.M.M. Kessels, M. Hermle, Atomic Layer Deposited Molybdenum Oxide for the Hole-selective Contact of Silicon Solar Cells, Energy Procedia, 92 (2016) 443-449. [19] R. Herberholz, M. Igalson, H.W. Schock, Distinction Between Bulk and Interface States in CuInSe2/CdS/ZnO by Space Charge Spectroscopy, J. Appl. Phys., 83 (1998) 318-325. [20] T. Walter, R. Herberholz, C. Müller, H.W. Schock, Determination of Defect Distributions from Admittance Measurements and Application to Cu(In,Ga)Se2 Based Heterojunctions, J. Appl. Phys., 80 (1996) 4411-4420. [21] I. Mora-Sero, G. Garcia-Belmonte, P.P. Boix, M.A. Vazquez, J. Bisquert, Impedance spectroscopy characterisation of highly efficient silicon solar cells under different light illumination intensities, Energy Environ. Sci., 2 (2009) 678-686. [22] A.F. Braña, E. Forniés, N. López, B.J. García, High Efficiency Si Solar Cells Characterization Using Impedance Spectroscopy Analysis, Journal of Physics: Conference Series, 647 (2015) 012069. [23] J.V. Li, A.F. Halverson, O.V. Sulima, S. Bansal, J.M. Burst, T.M. Barnes, T.A. Gessert, D.H. Levi, Theoretical analysis of effects of deep level, back contact, and absorber thickness on capacitance–voltage profiling of CdTe thin-film solar cells, Solar Energy Materials and Solar Cells, 100 (2012) 126-131. [24] O. Almora, L. Vaillant-Roca, A. Bosio, Electrical Characterizations of CdTe/CdS Poly-Crystalline Thin Film Solar Cells, Rev. Cuba Fis., 31 (2014) 66-70. [25] P.J. Sebastian, M.E. Calixto, R.N. Bhattacharya, R. Noufi, CIS and CIGS Based Photovoltaic Structures Developed from Electrodeposited Precursors, Sol. Energy Mater. Sol. Cells, 59 (1999) 125-135. [26] T. Eisenbarth, T. Unold, R. Caballero, C.A. Kaufmann, H.-W. Schock, Interpretation of Admittance, Capacitance-Voltage, and Current-Voltage Signatures in Cu(In,Ga)Se2 Thin Film Solar Cells, J. Appl. Phys., 107 (2010) 034509. 16 [27] P.P. Boix, J. Ajuria, I. Etxebarria, R. Pacios, G. Garcia-Belmonte, J. Bisquert, Role of ZnO Electron-Selective Layers in Regular and Inverted Bulk Heterojunction Solar Cells, J. Phys. Chem. Lett., 2 (2011) 407-411. [28] O. Almora, C. Aranda, E. Mas-Marzá, G. Garcia-Belmonte, On Mott-Schottky Analysis Interpretation of Capacitance Measurements in Organometal Perovskite Solar Cells, Appl. Phys. Lett., 109 (2016) 173903. [29] Y. Guo, J. Robertson, Origin of the High Work Function and High Conductivity of MoO3, Appl. Phys. Lett., 105 (2014) 222110. [30] B. Ghosh, Electrical Contacts for II–VI Semiconducting Devices, Microelectron. Eng., 86 (2009) 2187-2206. [31] V.F. Lvovich, Impedance Spectroscopy. Applications to Electrochemical and Dielectric Phenomena, John Wiley & Sons, Inc., Hoboken, New Jersey, 2012. [32] R.A. Sinton, A. Cuevas, Contactless Determination of Current–Voltage Characteristics and Minority‐Carrier Lifetimes in Semiconductors from Quasi‐Steady‐ State Photoconductance Data, Appl. Phys. Lett., 69 (1996) 2510-2512. [33] I. Mora-Seró, Y. Luo, G. Garcia-Belmonte, J. Bisquert, D. Muñoz, C. Voz, J. Puigdollers, R. Alcubilla, Recombination Rates in Heterojunction Silicon Solar Cells Analyzed by Impedance Spectroscopy at Forward Bias and Under Illumination, Sol. Energy Mater. Sol. Cells, 92 (2008) 505-509. [34] J. Bisquert, Nanostructured Energy Devices: Equilibrium Concepts and Kinetics, CRC Press Taylor & Francis Group, Boca Raton, 2014. [35] J. Bisquert, D. Cahen, G. Hodes, S. Rühle, A. Zaban, Physical Chemical Principles of Photovoltaic Conversion with Nanoparticulate, Mesoporous Dye- Sensitized Solar Cells, J. Phys. Chem. B, 108 (2004) 8106-8118. [36] R. Brendel, Thin-Film Crystalline Silicon Solar Cells: Physics and Technology, Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim, 2003.