In previous article, we discussed the project development of First Solar, Inc. (NASDAQ: FSLR), which is its cash cow in the foreseeable future. In this article, we will analyze First Solar’s module business – the cost and efficiency of its CdTe thin-film modules.
Currently the mainstream solar PV modules can be divided into two camps: c-Si and thin-film, with c-Si dominating the market with close to 90% of market share. There are some varieties in each camp: multicrystalline, quasi-monocrystalline and monocrystalline belong to the c-Si camp and amorphous-silicon (a-Si), CdTe and CIGS are in the thin-film camp.
The main complain about FSLR’s CdTe module is its low conversion efficiency as compared to the mainstream crystalline silicon (c-Si) modules. While FSLR used to be the king of module cost, the rapid fall of c-Si module price eroded FSLR’s lead. Now analysts usually say both types of modules have roughly the same level of costs. Yet the momentum of c-Si module cost gives an impression that c-Si module cost will soon fall below that of CdTe and FSLR will be a laggard in the race of reducing module cost. If the cost of c-Si module is on par with CdTe, then the lower conversion efficiency of CdTe module will certainly tilt the balance in favor of c-Si module. As a result, the c-Si module manufacturers like Suntech Power Holdings Co., Ltd. (ADR)(NYSE:STP) should win the module war. Yet the headlines are telling a different story, all the c-Si module manufacturers are racking up big losses quarter after quarter. So what is going on?
First Solar’s Module Efficiency and Costs
In the latest earnings conference call (2012-Q2), CEO Hughes stated following:
“Module manufacturing cost per watt for the second quarter, excluding our German plant, was $0.72. On a comparable basis, module manufacturing cost per watt was up $0.02 quarter-over-quarter. The sequential increase was due to higher plant underutilization costs and associated inefficiencies. Had our plants run at full utilization, then our module manufacturing cost per watt would have been $0.64 per watt or $0.04 below the Q1 2012 on a comparable basis.
Our best plant is manufacturing modules at a cost of $0.63 per watt, assuming full utilization. The average line conversion efficiency for our modules was 12.6% in the second quarter, which is up 0.9 percentage points year-over-year and up 0.2 percentage points quarter-over-quarter. The year - the current efficiency rate of modules produced on our best line was 13.1% last quarter compared to 13% this quarter.”
As a comparison, the average conversion efficiency is 12.2% at 2011-Q4 and 12.4% at 2012-Q1. Here FSLR showed steady improvements for its modules. FSLR has mentioned in the past that its target is to achieve efficiency of 13.5% to 14.5% at the end of 2014. Given progress in recent quarters, the target efficiency sounds credible. The increased module efficiency will bode well for the future BOS (Balance of System) cost reduction.
As for module cost, FSLR excluded German plant (with higher cost) because it already decided to close the plant at the end of the year. On surface, the $0.72 cost is nothing to be proud of. Here is a short history of its module cost: 2006 $1.40; 2007 $1.23; 2008 $1.08; 2009 $0.87; 2010 $0.77; 2011-Q2 $0.75; 2011-Q3 $0.74; 2011-Q4 $0.73. Therefore, its cost reduction appeared to slow down dramatically in recent years. However, as Hughes pointed out, the cost at full utilization in second quarter would be $0.64, a decent fall from $0.75 of a year ago (2011-Q2). Sure the reduction is likely related to the reduced output at German plant (or the German plant is not considered), yet it exactly showed that the management had made judicious decision in shutting down the plant.
Previously, FSLR has called for a module cost target of $0.5-0.55 by the end of 2014, which is a bit aggressive. Still, anywhere between $0.5 and $0.6 will keep FSLR in the leader’s group in terms of module cost.
One final note about FSLR’s module cost is that it is ‘all-inclusive’, covering freight, warranty and recycling. On the other hand, the stated “cost” for c-Si manufacturers normally does not include freight (instead it is put into selling expenses). No recycling program is in place for most c-Si manufacturers.
c-Si Module Efficiency and Costs
Current Efficiency
c-Si module has a rather long manufacturing chain starting with polysilicon, which is small piece, high purity silicon. In the next step ingots (square or round cylinders) are made from polysilicon, which are cut into thin silicon wafers. Solar cells are made by adding electric circuits to the wafers. In the final step solar cells are packaged to make modules so that they can last 20 or more years in the field.
The c-Si module efficiency is a mixed bag due to the existence of many manufacturers. Each manufacturer typically has modules with different specifications. The difference in efficiency can be attributed to the type of wafer used, the equipments (production line) making cells and modules, as well as the components and consumables used in the production.
The type of wafer (multi, quasi-mono and mono) used in cells has a large impact on its efficiency. Mono wafers are more expensive and have highest efficiency. Multi wafers are the most economical and have reasonable efficiency. Quasi-mono wafers are still evolving; its cost should be slightly lower than mono while its efficiency is in between multi and mono. For now, quasi-mono remains a niche product with ReneSola Ltd. (ADR)(NYSE: SOL) as the primary manufacturer.
The typical module efficiencies using different type of wafers are: 14.7% (multi), 15.5% (quasi-mono) and 16.1% (mono). Due to the lower cost, multi modules have always had higher market share than the other two. Although at one time mono modules were hot at Europe due to the strong rooftop market, now mono modules are a tough sell due to a shift in customers’ preference in favor of lower cost. More specifically, higher efficiency achieved through optimized/upgraded cell and module production lines costs less than using mono wafers. As a result, the c-Si modules in current market are dominated by multi modules and the average efficiency should be slightly above 15%.
The High Efficiency Game
As a trend in the PV industry, top module companies have constantly tried to excel in the high efficiency game. The level of success is varied and generally subdued. SPVI has published several articles related to the c-Si module efficiency this year: a general discussion and FSLR’s roadmap, on Panda which is Yingli’s flagship module; EPLS which is Canadian Solar’s top line; the relative success of Trina’s Honey; the predicament of quasi-mono wafers. The take-home message from these articles is that there is a delicate balance between efficiency and manufacturing cost. It is very hard to achieve commercial success (aka making money proportional to efforts) by playing the high efficiency game. The success of Honey appears to be a rarity mostly because of its modest target and the adoption of the most economic efficiency-enhancement technologies.
Besides the above-mentioned high efficiency products, Suntech Power’s “Pluto” line and SunPower’s E Series are also well known. Overall “Pluto” has failed as a product due to cost and quality issues, despite years’ effort by Suntech. SunPower is the world leader in terms of module efficiency. All of its product lines qualified to be high-efficiency. But the high-efficiency does come with high cost. Although cost is the focus of next section, here we just point out that the average module cost for SunPower is $1.46 at 2011-Q4, and the target for 2012-Q4 is $1.10. Both are well above costs of Chinese tier-1 manufacturers.
As discussed in the SPVI articles above, there are a variety of technologies which improve efficiency developed over the last 50 years, such as MWT/EWT, SE, PESC/PERC/PERL, IBC etc. Readers can research these terms at their own interests. The theories for these technologies are well established but so far only SE has been widely adopted. The reason, of course, is that it is very difficult to apply these technologies economically. Most of the time, manufactures find the efficiency gain does not justify the cost. One example is Canadian Solar’s ELPS development. At one time, CSIQ’s CEO sounded very confident in CSIQ’s capability to commercialize the MWT technology. However, after repetitive tries, CSIQ achieved little success. In its latest conference call, its CEO stated its target capacity for ELPS is 120MW by Q1-2013. Considering that the original announcement of ELPS is June 2011 and its total module capacity is 2+ GW, one can hardly deem it successful even if ELPS reaches the 120 MW target next year.
The Efficiency Trend
In this environment when all the companies are racing to achieve lower costs, ambitions for efficiency one or two percentage higher (module > 16.5%) are tamed. They are more inclined to adopt “incremental” strategy such as using anti-glare coating, making bus bars thinner - without much disruption to their processing flows. This will give them limited efficiency gain at minimal extra costs. Another route to higher efficiency is to purchase latest turn-key production lines from equipment makers. These lines usually are not much different from a few years back in terms of processing flow; rather the individual pieces of these lines are more optimized to squeeze out additional efficiency. For example, Centrotherm’s new Centaurus line can produce cells with efficiency of almost 20%. That is why Yingli Green Energy Hold. Co. Ltd. (ADR) (NYSE: YGE) is planning expansion and Canadian Solar Inc. (NASDAQ: CSIQ) is also considering expansion despite the obvious industry-wide glut. They simply want to update their lines to have a better product mix (more high-efficiency modules). Still expansion at this time is tough as almost all the module companies have heavy debt load, including Yingli and Canadian Solar.
Given continued losses and weak financials for most c-Si module manufactures, it is expected that the efficiency gain of c-Si modules will be very modest in the next few years – companies simply do not have much capital to invest. It should be at least 3-4 years before they can upgrade most of their production lines to the latest ones. The average efficiency gain from now is likely to be 1-2% by 2016. We suspect it is going to be close to the lower end instead of higher end.
Longer term, more efficiency gain will be hard as low-hanging fruits are picked. Further gain must be achieved through technology breakthrough (or rather commercialization breakthrough), which I will not speculate at this time.
Current Costs
The true cost of c-Si modules is hard to define. There are a multitude of numbers floating around from various companies and analysts. Given the under-utilization and some fluctuation in input materials, it is even harder to determine the true cost. Of course, one common term module manufacturers like to use is “processing cost”, which typically means “non-silicon wafer-to module processing cost”.
To get around the mess, I will go through Canadian Solar’s latest earnings report to find out its costs and the ASP it needed to break even at 2012-Q2. Why do I choose Canadian Solar? It is because CSIQ is one of the lowest-cost manufacturers if not the lowest. It has very good cost control and buys most wafer from GCL. Buying from GCL provides an advantage at the moment since GCL is supposedly the lowest-cost polysilicon producer and wafer maker. For most module makers, wafers from GCL cost less than self-made. In addition, the debt level of CSIQ is modest among Chinese tier-1 players. Another reason is that Canadian Solar just released quarterly report - the first among U.S. listed Chinese solar companies.
At 2012-Q2, CSIQ’s module shipments were 412 MW, net revenue was $348.2 million. Therefore the average ASP is about $0.81 (334M2/412MW after one-time item deduction). Gross profit from selling the modules was $29.2 million, implying a cost of 74 cents (per watt). However, the operating expenses were $46.2M including selling costs at $24.4M, G&A expenses at $18.4M. The research and development costs were $3.5 million and interest payments stood at $15.1 million. All these costs are relatively fixed, so math from above is that CSIQ needs revenue to be $380M to break even with one-time items excluded. That will translate to a break-even ASP of $0.92.
CSIQ’s capacity is about 2GW, so the utilization in the quarter is 82%, not too shabby among its peers. Even if one considers the slight under-utilization, the break-even ASP ought to be around 90 cents. Or can I say its true cost is 90 cents instead of frequently tossed-around numbers such as 67 cents?
In the first quarter CSIQ shipped only 15.7% modules to North America. No percentage of shipment to U.S. was reported this time around so with a rough estimate of 10%, CSIQ’s costs are not particularly skewed by the extra expenses incurred from the shipments to the US.
The Trend in Costs
Another important question would be - how much cost reduction can c-Si module makers achieve going forward? They indeed had a very impressive track record in driving down costs in the last 12-months. Silicon costs were ~$0.3-0.35 while non-silicon processing costs were around $1.1-1.2 a year ago, now they are at $0.12-0.14 and $0.65-0.75 (not including operating, R&D, and interest expenses) respectively. So can they keep the momentum going in the next 12 months?
It is easier to answer the question by separating the silicon costs and non-silicon processing costs.
Polysilicon used to be quite expensive due to limited supply. During the last PV-boom in late 2010, the spot price reached $100+ per kilogram. Now it has been hovering in the lower 20s for quite a while since the price collapse in the second half of 2011. Given the overcapacity in polysilicon supply and the break-even costs for the major producers are about $24-30 per kilogram, the polysilicon price will probably fluctuate within a narrow range in the next year or two. Hence the polysilicon cost should remain in a range of $0.11-0.15 per watt, meaning little savings going forward.
Estimating non-silicon processing costs is harder because there are much more variables. However it is much easier now because each cost component has largely stabilized. The total processing costs typically include 3 parts: polysilicon-to-wafer (or simply wafer), wafer-to-cell (or cell), cell-to-module (or module). Currently processing costs for efficient companies are: $0.18 for wafer, $0.20 for cell and $0.27 for module.
To better understand the forces behind the processing cost decline, one needs to go back and find out how the costs savings from the past 12 months were achieved. For each individual processing cost, it has following components: auxiliary input materials; costs directly related to processing such as labor, electricity; overhead; and depreciation.
The fall in the costs of auxiliary input materials is the primary driving force behind the rapid drop in the processing costs in the past 12 months. In other words, the suppliers cut their prices dramatically at the demand of wafer, cell and module manufacturers. Prices of some components and consumables were down more than 50%, a few even dropped 75% or more. Since the cost of these input materials depend on other more basic materials, it is impossible for them to decrease at the same pace as the selling price. The direct result is that many suppliers, who used to be profitable, were plunged into losses. Now only a handful of suppliers with high technological niche like DuPont and Heraeus can enjoy decent margins. Suppliers of components such as glass, aluminum frame, junction box, back panel, copper strip etc either break even or lose money. Their financial health is no better than the wafer, cell and module manufacturers. Therefore, there is little room from the supplier side to help the PV manufacturers.
For the other cost components within processing, since the processing flow remains unchanged, there is not much the manufacturers can do to reduce the costs as they are pretty efficient already. Certainly they all examined the labor costs and overhead, making some progress there. But there is a limit on how labor and overhead can be cut without a major reorganization.
Overall, there are only limited means on the table for the PV manufacturers to look for savings. One hope is that some key component makers like DuPont can significantly cut their prices. My guess is that DuPont etc may give limited concessions which could translate 2-3 cents savings. The other choice would be using cheaper replacements for certain expensive components such as EVA and silver paste. Of course, using cheaper, less-tested components is a risky bet which could compromise the module quality. By the original design, a c-Si module can last 20 or more years with only slight efficiency degradation. However, using less-tested substitutes may cut module’s lifespan and/or result a much higher degradation rate.
In summary, the rapid cost decline started last year has entered its final stage. The room for further cost decrease is very limited. The retirement of old production lines (which boosts the average efficiency and lower the per watt cost) and some reduction in input material costs may yield savings of no more than 10 cents by the end of next year. My guess is more likely 5 cents or so.
In the long term, those currently-in-red suppliers will have to be profitable to survive, thus at a certain inflection point, the costs shall rise instead of going one way down.
Conclusion
In this article the current efficiency and costs of both First Solar’s CdTe and c-Si modules are examined. In addition, the trend of both type of modules are analyzed.
While c-Si modules will still have higher efficiency than CdTe modules in the future, the efficiency gap between the two is likely to narrow slightly in the next two years. By the end of 2014, average CdTe efficiency should reach ~14% while c-Si module efficiency is going to slightly over 16%. In the meantime, the BOS cost will continue to drop, together with the narrowing of the efficiency gap, the so-called ‘efficiency penalty’ should go down for CdTe modules.
On the cost front, by using the numbers from latest quarterly report of Canadian Solar, I showed that the so-called c-Si catching up and surpassing CdTe is largely a product of misusing the term “costs”. Even for one of the lowest cost manufacturers such as Canadian Solar, First Solar’s true costs are still lower. Going forward, it appears that First Solar will have better chance to whack down costs faster than c-Si manufacturers. On the one hand, First Solar can keep investing and upgrading/optimizing its production lines. On the other hand, cash-strapped c-Si manufacturers can direct limited resources in upgrading (the expansion of YGE and CSIQ remains to be seen). Further decline in material costs will be minimal as many suppliers are struggling as well. In the end those suppliers will have to raise prices for their own survival.
Fundamentally, since c-Si modules have longer manufacturing chain and require way more types of input material, CdTe modules should have lower costs as long as the yield problem is solved. The proprietary technology that First Solar has developed over the years ensures its lead in terms of the manufacturing costs. Recent GE’s setback in its CdTe module ambition shows the technological barrier is too tough to overcome even for a giant like GE.
The solar cell tariffs imposed by the U.S. have already cut heavily into the module shipments from China to the U.S. market. There is also the ongoing anti-dumping investigation in Europe which could potentially limit Chinese module exports to Europe. The potential duties at Europe effectively add costs to the Chinese imports and make First Solar’s CdTe module appealing again.
The strategic scale-back of First Solar’s module capacity has by and large shielded First Solar from the ugly price war in a way that it does not need to sell modules at a loss. In the meantime, it can patiently wait on the sideline until enough c-Si capacities are destroyed and module prices will start to rise.