A New Mass Production Technology for High-Efficiency ... - IEEE Xplore

IEEE JOURNAL OF PHOTOVOLTAICS, VOL. 4, NO. 2, MARCH 2014

687

A New Mass Production Technology for High-Efficiency Thin-Film CIS-Absorber Formation Volker Probst, Immo Koetschau, Emmerich Novak, Axel Jasenek, Heinz Eschrich, Frank Hergert, Thomas Hahn, Jochen Feichtinger, Markus Maier, Bernd Walther, and Volker Nadenau

Abstract—A new mass production technology for CIS-absorber formation yielding high-average module efficiencies is introduced. A novel custom-designed oven very successfully exploits the principle of forced convection during heating, CIS formation reaction, and cooling. Cu(In,Ga)(Se,S)2 absorbers are formed by metal precursor deposition on soda lime glass followed by reaction in selenium/sulfur atmosphere. Processing is performed in a multiplechamber equipment which handles corrosive, flammable, and toxic process gases from atmospheric pressure to vacuum at high durability. The substrates (size: 50 cm × 120 cm) are processed in batches up to 102 substrates, applying forced convection for very homogenous heat transfer and high heating and cooling rates. Multiple-chamber design and batch size yield high throughput at cycle times above 1 h. This approach combines the specific advantages of batch type and inline processing. An excellent average efficiency of 14.3% with a narrow distribution (+/−0.31%) and a peak efficiency of 15.1% is shown with this technology. Module characteristic distributions during pilot production are presented. Detailed layer analytics is discussed. This straightforward reliable mass production technology is a key for highest module performance and for upscaling. Module efficiencies of 17% can be reached, enabling production costs below 0.38 US$/Wp in a projected GWp plant.

Fig. 1.

Process sequence for CIS module front end processing.

of solar modules remain steadily challenging. Only those thin film technologies that demonstrate 1) the highest efficiency and 2) the largest cost reduction potential will persist under these surrounding conditions. Within a typical CIS process sequence, the strongest impact on those two aspects is found in the CISabsorber formation. Thus, we have introduced a new CIS formation approach involving forced convection (FC) which quickly yielded the high-average aperture efficiencies in production ambient. This new technology combines batch and inline processing and its benefits in a very effective way and provides an excellent average module efficiency at high throughput and small production floor footprint.

Index Terms—Batch, CIGSSe, CIS, costs, high efficiency, inline, mass production, photovoltaic module, thin film.

II. EXPERIMENTAL SETUP I. INTRODUCTION OPPER–indium–diselenide-based solar cells have exhibited the maximum cell efficiencies of all thin-film approaches for the past 20 years [1] with the latest world record of 20.4% recently [2]. Nonetheless, the market share of CIS solar cells is still insignificant, since those top efficiencies could not successfully enough be transferred into commercial production so far. Additionally, the efficiency gap to multicrystalline Si solar modules and the ongoing price decline for all types

C

Manuscript received June 11, 2013; revised September 16, 2013; accepted October 6, 2013. Date of current version February 17, 2014. V. Probst, I. Koetschau, E. Novak, A. Jasenek, H. Eschrich, F. Hergert, T. Hahn, M. Maier, and B. Walther are with the Research and Development, Bosch Solar CISTech GmbH, Brandenburg 14772, Germany (e-mail: volker. [email protected]; [email protected]; Emmerich.Novak@ bosch.com; [email protected]; [email protected]; frank. [email protected]; [email protected]; [email protected]; [email protected]). J. Feichtinger and V. Nadenau are with the Corporate Research, Robert Bosch GmbH, Gerlingen-Schillerhöhe 70893, Germany (e-mail: [email protected]; [email protected]). Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/JPHOTOV.2014.2302235

A. Module Preparation The process flow for front-end processing is shown in Fig. 1. On standard float glass substrates of 120 cm × 50 cm in size, a molybdenum back contact is formed by dc magnetron sputtering. Subsequently, laser patterning of the back contact (P1) is performed to define the cell width for the monolithic cell integration. Then, the two-stage absorber formation process is started by dc magnetron sputter deposition of the CuGaIn precursor (CIG) using alternating CuGa and indium targets. For the subsequent Cu(In,Ga)(Se,S)2 absorber formation (CIGSSe or simply CIS), the new FC technology is applied which is described more in detail next. After CIS formation, a CdS buffer layer is deposited in a volume-minimized chemical bath that is followed by a sputtered intermediate layer of intrinsic ZnO. Mechanical patterning is performed to prepare for the integrated series connection (P2). The transparent front electrode is dc magnetron sputtered from ZnO:Al targets, and a mechanical insulation cut (P3) finalizes the front-end processing. Details on the typical device structure and module completion done by a chemical-bath-deposited buffer layer, a sputtered window layer, and scribing processes are given in [3].

2156-3381 © 2014 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications standards/publications/rights/index.html for more information.

688

Fig. 2. Schematic cross section of the processing chamber of the new FC oven normal to the production flow.

Fig. 3. Schematic section showing the production flow through a threechamber FC system. The system is extendable to multiple-processing chambers and entrance/exit load locks.

B. Forced Convection Absorber Formation Technology Fig. 2 schematically displays the processing chamber that is operated at typically from atmospheric pressure and above down to vacuum. It consists of a vacuum chamber with walls kept to an adjustable constant temperature in a range from room temperature to, e.g., 200 ◦ C. Condensation of processing gases can be avoided this way. At the inner surface of the chamber, a thermal insulation is applied in order to avoid heat loss and protect the chamber walls against processing gas interaction like, e.g., corrosion or condensation. At least one ventilator per chamber provides circulation of the processing gases through the substrate batch and the heater matrix. One important aspect is the chosen flow regime: for a given number of substrates and distance between the glass plates, the fan speed can be adjusted in such a way that from laminar up to turbulent all flow conditions can occur. Switching from heating to cooling can be accomplished by changing the position of the flaps accordingly. Intermediate positions can be controlled and are helpful for the adjustment of cooling rates in a wide range. Fig. 3 displays a schematic cross section in the plane of the production flow of a three-chamber FC system. The absorber formation process step is divided into different sections which are processed in the different chambers specified for the respective task. In the production mode, each chamber contains a full


batch of 72 or optionally 102 substrates. However, the equipment is not limited to this carrier capacity. Tests for even larger substrate load are currently under way. The full process recipe is applied to each batch by transferring the carriers in an indexing mode from one processing chamber to the next. Heating, metal precursor reaction, and cooling are performed via FC, which allows for very homogenous temperature distribution and CIS formation across the substrates and batch, even at ramp rates significantly above that of conventional batch-type furnaces. Flexible thermal profiling up to the transition point temperature of soda lime glass is possible and precisely controlled by fast response heating units. During the reaction process, the carrier gas is well controlled in order to exactly adjust the required partial pressure of the respective chalcogens. The furnace setup enables fast ramp-up and cool-down cycles as well as short-defined temperature plateaus. Process atmosphere and temperature can be changed very easily, and process cycles can be performed significantly faster than those in conventional batch-type diffusion furnaces. A typical total pressure range is from 100 to 1000 mbar. During the reaction process, the carrier gas for FC contains selenium and subsequently sulfur in a well-controlled manner in order to exactly adjust the required partial pressure of the respective chalcogens. The processing gas can also contain H2 Se or H2 S, either on its own or in combination with Se-vapor or S-vapor, respectively. Hydrogen may be present as well. These benefits result in very low costs of ownership for this new type of CIS formation process and equipment. III. RESULTS A. Module Electrical Characteristics and Distribution In a continuous 24-h/7-d production over a week, an aperture efficiency of η = 14.3% ± 0.31% was achieved in average over all modules processed. The furnace showed homogeneous performance nearly independent of carrier position over the whole load of 72 substrates. Correspondent statistics of the module electrical characteristics are shown in Fig. 4. The excellent reproducibility of absorber formation is not only reflected in a low variation of photovoltaic parameters of the produced modules but in the absence of any significant variation of CIGS material properties as well, as depicted in Fig. 5. Here, a region from X-ray diffraction (XRD) patterns taken in the Bragg–Brentano (BB) geometry covering the two chalcopyrite reflections 112 and 211 is shown together with the CIGSSe composition, as measured by wavelength dispersive X-ray fluorescence (XRF) for series of test modules from consecutive batches. The shape of the 112 reflection is especially determined by the elemental gradients in particular by the Ga-gradient, as will be shown later. Thus, the low variation between the different diffraction patterns concerning shape, intensity, and position of the reflections, as seen in Fig. 5(a), is a direct sign for a high reproducibility of the core CIGS material parameters as is the low-compositional variation from different runs shown in Fig. 5(b). Here, the anion-to-cation ratio shows up slightly above unity because of an additional amount of Se

PROBST et al.: NEW MASS PRODUCTION TECHNOLOGY FOR HIGH-EFFICIENCY THIN-FILM CIS-ABSORBER FORMATION

689

Fig. 4. Result of a 24-h/7-d test run with fixed process parameters. The electrical module data are shown as histograms and box plots (median, quartile, and whiskers of 1.5∗interquartile range, possible outliers). The diamonds symbolize the mean values plus their 95% confidence levels, and the red brackets depict the shortest halves.

that is incorporated during the annealing process in the Mo back contact. Thus, the stable amount of MoSe2 that is formed during annealing is an additional sign for high process stability.

Fig. 5. (a) XRD patterns of CIGS thin films on Mo-coated float glass substrates from eight consecutive runs. The intensities of the diffraction patterns have been shifted for the sake of visibility. (b) Composition of CIGS samples from 40 consecutive runs, derived from WD-XRF measurements. Here, the results from three different module positions are given (, top left “◦,” and bottom right “∗”), together with the average composition (solid line).

B. Absorber Characterization In order to investigate the structural properties of the absorber, XRD patterns were recorded in the BB and grazing incidence geometry [grazing-incidence X-ray diffraction (GIXRD)]. Fig. 6 shows the BB-XRD pattern in the range from 16◦ to 47◦ (diffraction angle 2θ). All visible peaks can unambiguously be identified with either the chalcopyrite phase, or the features of the back contact, namely, the 110 reflection of metallic molybdenum and a broad peak at 31.7◦ , which can be attributed to the 100 reflection of the MoSe2 phase because of its increased peak width clearly exceeding those of the neighbored chalcopyrite reflections 112 and 211. The 112 chalcopyrite reflection shows additional diffraction intensity at the peak tail on the right side (see arrow) at 27.5◦ . This very likely indicates the presence of chalcopyrite absorber material with a smaller lattice constant and may not be attributed to a separate phase. It is well known that the CIGSSe chalcopyrite lattice adopts a smaller lattice constant according to Vegard’s law if either Ga is substituted for In and/or sulfur is substituted for selenium [4], [5]. Under the assumption of a lateral homogeneous absorber formation process, as evidenced later on by electroluminescence (EL) and Raman mappings, the diffraction signal at 27.5◦ can be interpreted in terms of a compositional depth profile of the thin film. A nonde-

Fig. 6. BB X-ray diffraction pattern of a typical Cu(In,Ga)(Se,S)2 thin film. All peaks can be either assigned to the chalcopyrite phase or to the molybdenumrelated phases of MoSe2 (100 reflection) and Mo (110 reflection).

structive method to access this depth profile is GIXRD, which is used to sample the thin film at varying penetration depths [6]. The pattern acquired at the lowest incidence angle generates the most surface-sensitive diffraction pattern, and the measurement at highest incidence angle yields the most bulk-sensitive pattern.

690


Fig. 8. (a) EL image of a typical module from a continuous production run. The image is assembled from three single images, each covering one third of the module area, in order to enhance its resolution. (b) Mapping of the S/(Se + S)ratio at the surface of the same module as recorded by 5 × 6 single Raman spectra covering the whole module area. The mapping is scaled to relative deviations from the target value; the value 100 represents roughly S/(Se + S) = 0.2. Fig. 7. Determination of the gallium profile by GIXRD. Backside GIXRD patterns of the same Cu(In,Ga)(Se,S)2 thin film for incident angles between 0.2◦ and 10◦ . The maximum intensity decreases with smaller incidence angles. The simulations (solid lines) fit the measured data well and result in the Ga/(Ga+In) concentration depth profile of the lower chart.

It is also possible to quantitatively refine a set of diffraction patterns taken at various incidence angles simultaneously by a suitable simulation program in order to extract the depth profile of either the Ga/(Ga+In) ratio or the S/(S+Se) ratio [7]. However, some rough information about the depth profiles is essential in order to define a starting point of such quantitative refinements. Cross-sectional energy dispersive X-ray spectroscopy (EDX) measurement gives a first insight here and—as seen in other reactive sequential absorber processes—we also found a strong concentration of Ga near the interface to the back contact [8], [9]. Furthermore, our two-step process is a sulfurization after selenization, and we found that sulfur is enriched at the absorber surface. This finding will further be corroborated by Raman mappings of the sulfur A1-mode. Hence, the absorber is sulfur-rich at the surface and gallium-rich at the back contact. In such a configuration, it is difficult to unambiguously refine GIXRD patterns for a Ga-depth profile at the back contact. The incident X-ray beam penetrates from the absorber surface and the most surface-sensitive measurements are dominated by the sulfur gradient at the surface. In order to circumvent this problem, the GIXRD patterns were recorded from the backside of the film. For this purpose, an amorphous piece of polyethylene (8 mm × 12 mm) was glued to the absorber surface, and the absorber thin film was forcedly separated from the back contact. This sample was then subject to the GIXRD measurements, as shown in Fig. 7. Since the X-rays of the incident beam enter now from the backside of the thin film, the Ga-depth profile, which concentrates at the back contact, is most sensitively ac-

cessed (i.e., low absorption), and at the same time, the influence of sulfur is now strongly attenuated by the bulk material. In order to obtain a quantitative Ga-depth profile, the parameters of an empirical function describing solid state diffusion were subject to iterative refinements of simulation data [10]. The Ga/(Ga+In)-depth profile in the lower chart of Fig. 7 shows the final result, and the fit between simulated and measured GIXRD data is shown in the upper one. We ensured that the integral Ga content which is given by the simulated Ga/(Ga+In) profile meets the boundary condition of the overall integral Ga content as measured by XRF of the thin film (Ga/(Ga+In) ≈ 0.2). Given that most of the sulfur is in fact found at the surface [see Raman mapping in Fig. 8(b)], the influence of small amounts of sulfur in the bulk of the absorber on the refined Ga profile is rather low and yields only an overall offset of all measured peak positions. In summary, the structural investigation confirms a device structure where most of the Ga is located at the back contact and where only a small amount of Ga (< 2 at.%) is found in the space charge region (0–0.4 μm) of the device. The homogeneity of our modules is checked regularly with various imaging and mapping methods. A typical example is given in Fig. 8(a), which depicts an EL image and a Raman mapping of a module from a continuous production run. The EL image shown in Fig. 8(a) reveals uniform EL intensity throughout the whole module area, without the appearance of significant shunts or lateral intensity variations. Thus, current density as well as the bandgap at the surface of our modules can be assumed to have a high homogeneity [11]. The latter is confirmed by the S-content in the surface, as measured by a Raman mapping given in Fig. 8(b). Since the Ga content of the CIGS layers is accumulating toward the back contact, any increase of the bandgap at the surface above that of CuInSe2 can be attributed

PROBST et al.: NEW MASS PRODUCTION TECHNOLOGY FOR HIGH-EFFICIENCY THIN-FILM CIS-ABSORBER FORMATION

691

Fig. 10. EQE spectra reveal decreasing IR response when the absorber thickness is reduced (black: 100%, red: 92%, blue: 85%, green: 77%). The electric data refer to the standard (100%).

Fig. 9. SEM pictures from two samples descending from the same annealing process. In (a), the plain view of the CIGSSe surface is shown, and in (b), the cross section of a cleaved CIGSSe layer is shown.

to the amount of sulfur incorporated there. Raman has been shown to be a sensitive quantitative measurement technique for the Se/S balance in CIGSSe surfaces [12]. In the example given, most of the surface area lies within ±20% deviation from the target value for the sulfur content. Thus, a low lateral variation of the bandgap at the surface resulting from a low lateral variation of the CIGSSe composition can be regarded as a prerequisite for the high EL uniformity shown in Fig. 8(a). The morphology of CIGSSe layers shows up most prominently in scanning electron microscope (SEM) images (see Fig. 9): The CIGSSe layer is dense and shows a compact surface without significant holes or pronounced grain boundaries; the cross-sectional view confirms the compactness of the absorber material with grain sizes typically in the range of the film thickness. C. Spectral Response Measurement In order to understand the electrical properties, we first investigated the spectral response as a function of absorber thickness. Fig. 10 shows four characteristic measurements of the external quantum efficiency (EQE), and as can be readily seen, all devices have an effective bandgap Eg ≈ 1.0 eV. This finding fits well to the device structure of a Ga-rich back contact. Within the space charge region, only a little Ga is found, and the device works basically like a CuInSe2 cell with an excellent current collection due to the back surface field, which is induced by the

Fig. 11. I(V) and P(V) curves of the champion module yielding an aperture area efficiency of 15.1%. The module voltage corresponds to an average cell voltage of 611.5 mV.

Ga gradient. However, the open-circuit voltage of typical cells is around 590–610 mV and exceeds the expected “rule of thumb value” of Eg /e − 0.5 V = 0.5 V), which can be only explained by the sulfur incorporation near the surface. Motivated by material saving, an experiment to reduce the absorber layer thickness showed us clearly that the infrared (IR) response diminishes in thinner absorbers (see Fig. 10). Still, the standard thickness (100%, on which all other investigations of this study are based) yields the best results. This indicates that the crystal quality of the standard thickness is excellent—carrier lifetime and mobility are sufficient to separate and collect the generated charge carriers throughout the absorber thickness. Consequently, a further increase of the IR response, just as efficiency, is expected for thicker absorbers. D. Champion Module From a 24/7-Run Fig. 11 displays the current–voltage characteristics of our champion module so far. It yields an aperture area efficiency of 15.1%, a fill factor of 73.3%, and an open-circuit voltage 73.4 V ¨ Rheinland. measured and confirmed by TUV The processed modules show excellent stability during light soak and climate testing. IV. PRODUCTION COSTS PERSPECTIVE We assume a 17% module efficiency (total area) to be achieved, even on a doubled substrate size by using FC technology. For this case, production costs were calculated using a detailed model for the overall process, taking into account

692


REFERENCES

Fig. 12.

Items of expenditures for a 500 MWp CIS factory.

total cost of ownership of each process step. We performed a study for fab sites from 250 MWp to 1 GWp. Scaling up fab and substrate size leads to production costs below 0.38 US$/Wp in a continuous 1-GWp production. Because of excellent performance of the FC oven (short cycle time and high capacity), low-capital expenditures (CAPEX) and depreciation rates can be reached, and consequently, the total costs will be dominated by material prices (see Fig. 12). V. SUMMARY AND OUTLOOK A new type of mass production technology for high-efficiency CIS formation has been introduced. It is based on FC during the process steps: heating, CIS reaction, and cooling. The new technology is very beneficial because of 1) high average efficiencies at a narrow distribution reproducibly demonstrated in 24-h/7-d production runs; 2) high throughput at very low production cost, made possible by the combination of batch type and inline processing. The new FC CIS technology is scalable fourfold: in terms of substrate size, batch size, substrate packing density, and number of equipment chambers. In consequence, highest throughput per production unit is exploitable to lowest photovoltaic module production costs. For a projected 1 GWp, plant production costs below 0.38 US$/Wp were calculated.

[1] National Center for Photovoltaics, NAt. Renewable Energy Lab., Golden, CO, USA. Chart of best research-cell efficiencies. (2013). [Online]. Available: http://www.nrel.gov/ncpv/ [2] Swiss Federal Laboratories for Materials Science and Technology, EMPA, Dübendorf, Switzerland. “New world record 20.4% for flexible CIGS on polyimide foil.” Jan. 17, 2013 (Press Release). Available: http://www.empa.ch/plugin/template/empa/1/131438/—/l=2 [3] V. Probst, F. Hergert, B. Walther, R. Thyen, G. Batereau-Neumann, B. Neumann, A. Windeck, T. Letzig, and A. Gerlach, “High performance CIS solar modules: Status of production and development at Johanna solar technology,” in Proc. 24th Eur. Photovoltaic Sol. Energy Conf., Hamburg, Germany, 2009, pp. 2455–2459. [4] D. K. Suri, K. C. Nagpal, and G. K. Chadha, “X-ray study of CuGax In1 −x Se2 solid solutions,” J. Appl. Cryst., vol. 22, pp. 578–583, 1989. [5] I. M. Kötschau, Strukturelle Eigenschaften von Cu(In,Ga)(Se,S)2 Dünnschichten, Ph.D. dissertation, Univ. Stuttgart, Stuttgart, Germany, 2002. [6] B. L. Ballard, X. Zhu, P. K. Predecki, D. Albin, A. Gabor, J. Tuttle, and R. Noufi, “Determination of composition and phase depth-profiles in multilayer and gradient solid solution photovoltaic films using grazing incidence,” Adv. X-Ray Anal., vol. 38, pp. 269–276, 1995. [7] I. M. Kötschau and H.-W. Schock, “Compositional depth profiling of polycrystalline thin films by grazing-incidence X-ray diffraction,” J. Appl. Cryst., vol. 39, pp. 683–696, 2006. [8] T. Dalibor, S. Jost, H. Vogt, R. Brenning, A. Heiß, S. Visbeck, T. Happ, J. Palm, A. Avellán, T. Niesen, and F. Karg, “Advanced CIGSSe device for module efficiencies above 15%,” in Proc. 25th Eur. Photovoltaic Sol. Energy Conf., Valencia, Spain, 2010, pp. 2854–2857. [9] G. M. Hanket, W. N. Shafarman, B. E. McCandless, and R. W. Birkmire, “Incongruent reaction of Cu–(InGa) intermetallic precursors in H2 Se and H2 S,” J. Appl. Phys., vol. 102, pp. 074922-1–074922-10, 2007. [10] I. M. Kötschau and H.-W. Schock, “Depth profile of the lattice constant of the Cu-poor surface layer in (Cu2Se)1 −x (In2Se3)x evidenced by grazing incidence x-ray diffraction,” J. Phys. Chem. Solids, vol. 64, pp. 1559– 1563, 2003. [11] T. Ott, P. Mack, F. Trudel, Y. Schulz, T. Walter, D. Hariskos, O. Kiowski, and R. Schäffler, “Accelerated ageing and luminescence of CIGS solar cells,” in Proc. 26th Eur. Photovoltaic Sol. Energy Conf., Hamburg, Germany, 2011, pp. 2421–2424. [12] V. Izquierdo-Roca, A. Pérez-Rodr´ıguez, A. Romano-Rodr´ıguez, ´ J. R. Morante, J. Alvarez-Garc´ ıa, L. Calvo-Barrio, V. Bermudez, P. P. Grand, O. Ramdani, L. Parissi, and O. Kerrec, “Raman microprobe characterization of electrodeposited S-rich CuIn(S,Se)2 for photovoltaic applications: Microstructural analysis,” J. Appl. Phys., vol. 101, pp. 103517-1–103517-8, 2007. Authors’ photographs and biographies not available at the time of publication.