A similar version of this paper will appear in the journal Solar Energy and Solar Cells (1999).
National Renewable Energy Laboratory
Thin Film PV Partnership
1617 Cole Boulevard
Golden, CO 80401
303-384-6441 (-6430 fax)
ken_zweibel at nrel.gov
Key Words: photovoltaics, thin films, copper indium diselenide, cadmium telluride, amorphous silicon, thin film silicon, manufacturing costs
To improve module prices, progress is needed on three fronts: the performance of the modules (efficiency, or W/m2), their direct manufacturing cost ($/m2), and increased volume production. It is direct manufacturing cost ($/m2) and some effects of volume production that are the focus of this paper.
The simple combination of cost per square meter divided by output per square meter yields the key parameter, $/Wp used as a measure of PV cost-effectiveness. Table 1 shows for thin films some likely future combinations of direct manufacturing cost and efficiency in order to indicate the multiple paths by which low module cost in $/Wp can be approached. The highest cost ($3.3/Wp) represents about where we are today in commercial thin films. The lowest cost, which is a combination of 15% module efficiency with $50/m2 direct manufacturing cost, represents the US Department of Energy's long-term goal for thin film modules. It is about a tenfold reduction, which will require substantial progress. Actually, any combination of cost and performance below about $0.5/Wp would represent a good approximation of this low-cost goal. Despite its importance, the predominant emphasis of this paper will not be on PV module efficiency. That has been covered elsewhere . This paper will be on direct module manufacturing cost, and efficiency will only be of concern in that it must be maintained or enhanced despite strategies that reduce manufacturing cost.
With progress toward low cost, new energy-significant markets should open up for PV. Examples of such markets are: residential and commercial rooftops, building facades, parking lot structures, distributed utility grid support, centralized PV for daytime loads, hybrid systems for decentralized rural electrification, hybrid systems (e.g., with gas-powered fuel cells or microturbines) for distributed power networks, and PV with storage (compressed air) or PV for hydrogen production. By contributing energy on this scale, PV can fulfill its promise to contribute to the betterment of the world's environment (e.g., better air quality and reduced carbon dioxide) while reducing high risks associated with international energy supply tensions resulting from more traditional sources.
|Amorphous Silicon||Solarex, United Solar, Canon, others||Commercial Production under 10 MW at Several Plants|
|Cadmium Telluride||First Solar, BP Solar, Matsushita||First Solar Production under 1 MW, Others Lower|
|Copper Indium Diselenide||Siemens Solar Industries (SSI), Global Solar||Initial Small Quantity Manufacture under 100 kW at SSI|
In addition to previous studies [1-13], the author has used a detailed cost model specifically created for NREL by First Solar (then Solar Cells Inc. of Toledo, OH) to enable a better prediction of manufacturing costs. In cooperation with NREL, First Solar has provided specific manufacturing information. First Solar currently manufactures small volumes (under 1 MWp) of CdTe modules. However, they have published plans to increase production to over 20 MWp in the next several years. The manufacturing model details their projected manufacturing cost for a thin film manufacturing facility based on an existing CdTe sublimation process. The First Solar data is the basis for a baseline thin film manufacturing process that is used throughout this paper. Other technologies are compared using this baseline and data from the public domain, except for Si Film. AstroPower has supplied the data for their experimental, high-rate, high-temperature process on ceramic substrates, which is being developed in their Thin Film Partnership subcontract with DOE/NREL (as are almost all of the other technologies examined in this paper).
Naturally, the First Solar model is far more representative of their own approach than it is of any other. In addition, the First Solar plant has not yet been scaled-up to the levels implicit in the model, which means that the numbers may underestimate real costs. In fact, since prior projections by some thin film PV companies have seriously underestimated costs, this concern is a valid one, even for this study. In the end, only each company scaling up its own technology can be certain of the real costs of thin film module manufacturing.
Projections concerning the other solar technologies (some of which are already commercial) are made merely as guides and in order to establish the potential of thin films in general. The only reason this is a valid activity at this time is that, today, no detailed cost numbers are available concerning any of the basic thin film manufacturing approaches. The model examined here is a substitute for such information. The purpose is to point out obvious, first-order conclusions.
Even within each of these categories, there are similarities. For example, almost all thin films require transparent, conductive oxides for top contacts. These can be made or purchased pre-deposited by a vender on glass superstrates (substrates that end up on top of the module and thus must be transparent to sunlight). In addition, there is also almost always a metal deposition step for the back contact, which is frequently DC magnetron sputtering of a metal like aluminum. Finally, the heart of the module, the formation of the alternating n-type and p-type layers to form a junction, also has overriding similarities such as similar thicknesses, low materials costs when done well, and acceptable energy requirements for manufacture. The great similarities of structure and processing among the various thin films allows us to gain substantial insights into their potential costs and the barriers that they must overcome to reach those costs.
Given these similarities, where do the cost differences among the thin films come from? First, thin film modules differ in performance. Though this is important and sometimes crucial, it is not the subject of this paper. (But see Ref. .) In terms of manufacturing cost (cost per unit area), the most substantial factors are:
Given the similarities among thin films, which essentially determine that all of them will have order-of-magnitude similarities in direct manufacturing costs, these differences provide the most substantial causes for the costs to differ from some typical value among different materials, device, and processing options. For example, the baseline First Solar thin film technology uses only a few dollars of semiconductor material, requires very little capital cost, has a low-cost glass substrate, and sustains some added cost for encapsulation. In their model, these costs sum to a cost per square meter of less than $100 at the 20 MW/yr production level, and they will trend down towards some lower cost (perhaps about $50) as their technology is optimized for larger volumes (see below). Other technologies vary from this baseline because of the differences stated above. Table 3 gives a summary of the costs of the First Solar CdTe manufacturing process used as a baseline in this paper. The costs were taken from the First Solar-supplied model. The data can provide rough comparisons with other technologies.
|Component||Direct Manufacturing Cost ($/m2)||Comments|
|Materials (all)||$48||Semiconductors only about $5; mostly encapsulation, substrate, and modularization|
|Capital (all)||$10||Semiconductors only about $5|
|Heat, Electricity, Water||$3||Energy payback < 3 months for energy added during manufacturing|
|Labor||$12||Plant labor and operations management|
|Maintenance of Equipment||$3||4% of Initial Capital Cost|
|R&D||$4||Must maintain technical lead|
|Warranty||$5||3% of sales (very high for early high prices)|
|Rent and factory overhead||$5||Factory overhead at 1.5% sales|
|Total direct manufacturing||$90/m2||Projected from existing technology, not yet optimized|
The simplest comparisons of the capital costs of different PV technologies compare dollars per installed megawatt of annual capacity. However, this can be misleading, because two parameters are varying: cost per square meter and module wattage per square meter (efficiency). To get a true picture of manufacturing costs, analysis must be based on cost per square meter. Performance can be folded in later to get $/Wp comparisons. It is possible to come to a better understanding of the direct manufacturing cost per square meter of different thin film approaches if one breaks process costs into components. However, such insights depend on making simplifying assumptions that have substantial uncertainties. With this in mind, the estimates presented below must be viewed as merely heuristic assistance in comparing different thin film approaches.
Two major categories of capital cost are: (1) the equipment used to make the active semiconductor layers; and (2) the equipment used to make the rest of the module (e.g., metal contacts, lamination, scribing, annealing). The most critical aspect of capital cost is the first: equipment to make the semiconductors. The rest of the capital costs will be very similar for all thin films and can be minimized separately (e.g., by doing away with a second piece of glass in encapsulation). For the purpose of this study, I will assume that this non-semiconductor capital cost is the same for all approaches and equals $5/m2. This is the amount in the First Solar baseline. This assumption helps in the examination of differences in the semiconductor aspect of manufacturing.
Based on the baseline case and a seven-year depreciation, the total capital cost (C) can be parameterized by the ratios of three properties to the baseline. The three properties are: the semiconductor deposition rates (R), the semiconductor thicknesses (T), and some measure of the relative complexity of the equipment (E). This latter is needed because equipment costs vary per deposition zone area. Vacuum equipment is more expensive than a simple annealing oven. For an in-line process, this parameter can be calculated by dividing the equipment cost by the area of the deposition zone. For a batch process, it is the equipment cost divided by the deposition zone area and then divided by the number of plates being simultaneously processed. This points out that batch processing has an inherent advantage over in-line processing, if it can be done while maintaining module performance and production throughput.
Based on these assumptions, and using the baseline data from First Solar, the formula for semiconductor deposition capital cost ($/m2) is:
The figure of 80 Wm-2 comes from the assumed baseline efficiency; the 7 comes from the depreciation period, and the 18 MW comes from the plant's designed throughput. For the semiconductor deposition alone, this reduces to semiconductor equipment (SE) costs of:
Recall that these parameters (E, T, R) are ratios to the baseline. If all the ratios are 1 (i.e., we are using the baseline), C reduces to about $9.4/m2. SE reduces to $4.4/m2. These are the First Solar values of total capital equipment cost and semiconductor capital equipment cost; and the difference is the 'front and back end' capital equipment cost. An interesting comparative number is the ratio of capital cost for just the semiconductor layers (compared against the First Solar baseline), which can be defined as SR = E x T/R. This figure shows the relative contribution of that equipment as compared to the baseline value of 1. Most ratios are higher than one, with a few exceptions.
Table 4 gives the set of assumptions used in comparing various approaches to thin film manufacturing for amorphous silicon, cadmium telluride, copper indium diselenide, and Si-Film. I have chosen these because I was able to make reasonable estimates of the parameters. One process not examined is CIGS fabrication using a two-step sputtering and selenization process. Although this is a mainstream process in CIGS, the batch process parameters are difficult to estimate without company input. Also, I have assumed that the semiconductor layers already include secondary, very thin layers such as buffers, windows, heterojunction partners; and these are all done by the selected process and are included in its thickness. This assumption is clearly an oversimplification but is probably not a major source of error.
|Process (status)||Deposition Rate (micron/min)||Equipment Cost per Deposition Zone Area ($M/m2)||Semiconductor Thickness (micron)||Companies||Comments|
|Sublimation of CdTe (commercial)||6||8||4||First Solar||Fast in-line|
|Electrodeposition CdTe (pilot line)||0.006||0.005||2||BP Solar||Slow batch|
|In-Line a-Si Glow Discharge (commercial)||0.02||0.7||1||Solarex, USSC||Slow in-line|
|Box Carrier (Batch) a-Si (commercial)||0.02||0.1||1||EPV||Slow batch|
|High-Rate a-Si (experimental)||0.1||1.8||1||Moderate in-line|
|High-Rate CIGS Evaporation (experimental)||0.3||5||2||Global Solar||Moderate in-line|
|Silicon-Film (experimental)||500||30||50||AstroPower||Fast in-line|
Table 5 is a compilation of various SE and SR values for different semiconductor options. Recall, however, that these are very sensitive to assumptions. This is especially true for the unknown value of the equipment ratio (E), which could vary over a large range. Unlike rate and thickness, E can only be estimated. For example, a rough estimate of the baseline value would be: $7M/0.9 m2 = $8 M/m2. For in-line glow discharge, a rough estimate would be: $7M/10 m2 = $0.7M/m2. Then the ratio, E, for in-line glow discharge would be 0.09 (as in Table 5). Since these are only rough estimates, the uncertainties in Table 5 are quite large. Table 5 should only be used to give a sense of magnitude differences between semiconductor deposition processes.
The last column in Table 5 gives a rough approximation of the capital cost of a 100 MW plant for each semiconductor deposition process (it does not include back end costs). Note that the uncertainty of this number is very large and should only be used to indicate trends. This is because the equipment ratio (E) is quite uncertain (as explained above). The value of the semiconductor capital equipment cost for a 100 MW plant shown in Table 5 is calculated from the other assumptions by multiplying the baseline cost for the First Solar approach ($31 million) by the ratio SR, as defined above. Thus these are derived values, and are not independently verified. Again, they are estimates useful for understanding trends.
In addition to the previously mentioned uncertainties, other caveats are necessary. For example, the above comparisons cannot be made assuming the same module efficiency. In fact, it is because efficiency does vary substantially (from about 5% on the low end, to almost 12% on the high end) that companies may make decisions based on this factor rather than manufacturing cost, especially for initial manufacturing plants. At costs of $1/Wp, a change of about 1% in module efficiency is worth about $9/m2 in cost (although this factor changes at different cost/performance levels), and about double that at the system level.
Similarly, manufacturability, uptime, and yield vary for different processes, and they can be overwhelmingly critical in making choices among competing technologies.
In the future, most processes will be optimized for higher rates than those given in Table 4. In addition, most absorber layer thicknesses will be 1 micron or less, since that is all that is needed to absorb over 90% of the incident sunlight. Today's greater thicknesses reflect technical compromises needed to achieve high-yield in early manufacturing. Clearly, there are options for improving most of the approaches given in Table 5, even in-line GD (through higher rates) and CIGS evaporation (through higher rates, thinner layers).
|Process (status)||Rate Ratio (R)||Equipment Ratio (E)||Thickness Ratio (T)||SE ($/m2)||SR||Companies||Initial Cost ($M) for 100 MW @ 10% Eff.|
|Sublimation of CdTe (commercial)||1||1||1||$4.4||1||First Solar||$31M|
|Electrodeposition CdTe (pilot line)||0.001||0.006||0.5||$1.8||0.3||BP Solar||$10M*|
|In-Line a-Si GD (commercial)||0.0033||0.09||0.25||$30||6.8||Solarex, USSC||$210M|
|Box Carrier (Batch) a-Si (commercial)||0.0033||0.012||0.25||$4||0.9||EPV||$28M|
|High-Rate a-Si (experimental)||0.017||0.22||0.25||$14||3.2||$99M|
|High-Rate CIGS Evaporation (experimental)||0.05||0.625||0.5||$26||6||Global Solar||$186M|
An important aspect of capital cost is that it can be a very high initial hurdle for new manufacturing. Not only are the up-front costs high in terms of financing a new factory, they have serious impact on the first few years of production. Like any form of capacity costs, if the factory does not operate at full capacity, capital costs are multiplied. In other words, if module production yields are lower than expected and downtime stretches out production volumes, capital costs can balloon. Low initial capital costs have substantial value.
|Process (status)||Capital Cost ($/m2)||Maintenance $/m2 (28% of capital cost)||Total (capital plus maintenance) $/m2||Companies|
|Sublimation of CdTe (commercial)||$4.4||$1.2||$5.6||First Solar|
|Electrodeposition CdTe (pilot line)||$2||$0.6||$2.5||BP Solar|
|In-Line a-Si GD (commercial)||$30||$8||$38||Solarex, USSC|
|Box Carrier (Batch) a-Si (commercial)||$4||$1.1||$5.1||EPV|
|High-Rate a-Si (experimental)||$14||$3.9||$18|
|High-Rate CIGS Evaporation (experimental)||$26||$7||$33||Global Solar|
Since the estimates given in Tables 5 and 6 have a great deal of uncertainty, especially in the E value, we must take these estimates with substantial caution. However, based on very large observable differences, some hypotheses are:
|Material||Amount (g) @ 75% Utilization and 4 micron thickness||Cost per kg||Cost/m2|
|S, Al, etc.||2 g||$100||$0.2|
Active material cost is dominated by one component, the tellurium, which is the most costly feedstock. Notice also that these costs are for 4 micron CdTe films. In the future, these numbers will be reduced as CdTe thickness approaches 1 micron.
The amounts for other technologies are fairly similar. Table 8 gives some estimates for these other technologies, based on reasonable assumptions. Only the most costly element is given, and this is usually the one that dominates such costs. If a technology has more than one costly element that is used poorly, it will have problems. None of the ones examined have this problem.
|Process (status)||Material||Utilization Rate||Cost ($/kg)||Thickness (micron)||Amount (g) per micron-m2||Approximate Cost ($/m2)|
|Sublimation of CdTe (commercial)||Te||75%||275||4||3.2||$3.5|
|Electrodeposition CdTe (pilot line)||Te||95%||275||1.5||3.2||$1.3|
|In-Line a-Si GD (commercial)||Ge||10%||3000||1||0.4||$12|
|Box Carrier (Batch) a-Si (commercial)||Ge||25%||3000||1||0.4||$5|
|High-Rate a-Si (experimental)||Ge||10%||3000||1||0.4||$12|
|High-Rate CIGS Evaporation (experimental)||In||50%||400||2||2||$3|
Note that these costs are fairly reasonable for all approaches with the possible exception of germane for amorphous silicon multijunctions made in an in-line system. Substantial R&D is being done to improve germane utilization in order to cut these costs for in-line systems; they are already low for batch processing.
Of the non-semiconductor costs (about $43/m2 in the baseline), substrates are one of the most important and best understood. Table 9 gives a list of common substrates and their approximate costs. One key question is whether to make or buy tin oxide-coated glass for those technologies (CdTe and a-Si) that have this option. It is quite likely that the price of tin oxide-coated glass will drop substantially as volumes increase, since the cost to the glass maker for large pieces is only about $4/m2 (with transportation). Rough estimates of future costs are also given in the table.
Projection Based on Large Volume
|Float Glass||$4||$3||Requires making own TCO or metal contact|
|Float Glass/tin oxide||$10||$4||Baseline for First Solar & Solarex|
|Stainless Steel Foil||$4||$3||USSC|
|Polyimide Roll||$10||$5||Global Solar and others (temperature limited)|
|Ceramic||$9||$5||Device-design driven, AstroPower|
These are all similar with the possible exceptions of ceramic and polyimide. However, choices are also made based on the thermal stability needed for high-temperature processes, or on product design, such as the need for flexibility or light weight. Some substrates require extra processing (e.g., conductive stainless steel has extra handling/processing steps as compared to nonconductive glass) and other minor differences (breakage, rolls versus stacks, etc.). These can alter costs. In the future, new glass superstrates may be chosen if they maximize transmission while maintaining low cost (e.g., borosilicate glass made on a float line); and new ceramic or high-temperature plastic substrates may become available.
The value of R&D includes:
The value of increased volume manufacturing include:
Table 10 gives a summary of reasonable goals for cost reductions by category based on R&D and volume production improvements.
Of these various cost reductions, perhaps the most difficult will be materials cost reductions. However, today's PV R&D has focused on module efficiency and semiconductor process development. In the future, increased resources will be devoted to improving encapsulation methods and other similar processes and their costs. Simpler designs (e.g., eliminating either front or back glass; substituting other materials for current EVA pottants; and avoiding the costs of lamination) should provide avenues for improvement.
|Component||Direct Manufacturing Cost ($/m2)||Comments|
|Materials (all)||$28||Volume purchases reduce all costs, especially substrates; lower-cost encapsulation|
|Capital (all)||$5||All process can be optimized, rates increased, layers thinner|
|Heat, Electricity, Water||$2||Larger volumes, thinner layers|
|Maintenance of Equipment||$2||4% of lower capital cost|
|R&D||$1||Rising sales, lower prices|
|Warranty||$1||Lower prices, greater product reliability|
|Rent and factory overhead||$5||Larger volumes|
|Total Direct Manufacturing||$50||Optimization, R&D, and higher volumes|
In addition to this fundamental observation, a number of other key findings can be made:
It is interesting to speculate about the ultimate cost of thin films. It could be lower than the $50/m2 goal, since achieving that goal can be visualized with current knowledge. Since thin film PV is a scientifically based technology that will eventually be produced in ultra-large volumes, it should ultimately cost the sum total of the materials costs of its elemental constituents processed and transported with the greatest efficiencies. This cost could be lower than $50/m2. Efficiencies will also improve beyond the 15% performance goal (but that is not the subject of this paper). Thus the module cost per watt goal of $0.33/Wp could also be surpassed as the stimulus of positive cash flow allows the thin film PV technologies to mature and blossom during the 21st century.