ECD: Better or Worse? (ENER)

ECD has created a shiny new presentation to introduce interested parties to the company. It is an excellent illustration of what Energy Conversion Devices are all about. Let us focus, specifically, on the "better real-world performance" claim made on slides 10 and 12.

According to these slides, Unisolar's "amorphous silicon triple-junction" product dramatically outperforms virtually any other PV panel on the market - from "dual-junction amorphous silicon" to monocrystalline silicon. It is all bogus, of course. In the real world, on real roofs, Unisolar's glued-on laminates should rarely outperform optimally-tilted rack-mounted glass PV panels (whether thin-film or crystalline), and the initial energy-yield advantage, if any, should typically be gone within a few years, mostly due to long-term degradation issues.

The first thing one should notice on Slide 10 is that the yellow bar labeled "Avg. a-Si Double Junction" appears too tall. That bar is actually only 29% shorter than the red UNI-SOLAR bar, not 41% as the label claims. Turns out that the bar labeled "Avg. Poly-Si" is also a bit taller (it is only 15% shorter than the Unisolar bar, not 18% shorter as labeled). Thus, the bar chart drawing appears to visually penalize Unisolar's "relative performance." So was this some kind of a Freudian slip or simply an innocent alteration to make the chart look nice?

Slide 12 supplies the "supporting data" for the chart. According to that slide, "USO [that is, United Solar Ovonics] outperforms: Mono-Si by 11%, Poly-Si by 18%, CIS/CIGS by 24%, DJ a-Si by 41%." The source is indicated as "Tucson Electric Power, Arizona, USA," the site, as "Tucson, Arizona, USA." And, specifically, the average annual yield (2004-2007) of Unisolar's "a-Si TJ" is shown as 1,757 kWh/kWp, while the "avg. mono-Si" is 1,588, and the "avg. poly-Si" is 1,491. So far, so good.

Now, go back and carefully read the Safe Harbor (slide 2). All the disclaimers there are about forward-looking statements. But what about past ("historical") information and statements? What about slide 23 presented at the Investor Day last year, and a subject of the first post in the "kWh/kW myth" series in this blog?

On that day, investors were told that "Tucson Electric Power report[ed]" that Unisolar had 1,757 kWh/kW average annual yield, mono-Si, 1,686, and poly-Si, 1,492.

Wait!

The 1,757 Unisolar number is the same in both the "Introduction" and the "Investor Day" slides, but the mono-Si number is different (1,588 vs 1,686) and so is the poly-Si number, by just a tad (1,491 vs 1,492). So when did ECD tell the truth, on October 2nd, 2008 or, a year later, on October 12th, 2009? Let us look more closely at the data source.

Tucson Electric disclose, on their web site, the underlying raw monthly and yearly kWh data (in pdf files) for many systems they have monitored at their Springerville station (SGS), their Operating Headquarters (OH), as well as some customer locations, but there is no specific report there comparing Unisolar's performance with those of other systems for the period 2004-2007. In fact, that raw set is plagued by some incorrect entries (for example, the Shell/Siemens ST40 panels that comprise the 1440W system at the OH are not a-Si as stated - they are CIGS, and the Astropower system at OH is not a round 1500W but a bit smaller - 1485W or 9 x 165W). In addition, there have been various panel replacements and inverter changes and modifications (the Trace inverter of the Unisolar system, for example, was apparently replaced by a Fronius inverter sometime in 2005). Finally, it should be noted that only the SGS systems were truly monitored on a real-time basis (while data was collected manually for the OH and the rest of the systems, see pages 33 and 36 of the 2006 Annual Report). The table below is an attempt to provide a clean dataset (showing only systems that had non-zero kWh generation in each of the 48 months in the period 2004-2007, although an exception had to be made for the mono-Si systems).

So, what do the data reveal? The "average" 2004-2007 kWh/kW energy yield number for small systems (basically, at OH) was 1,590 for mono-Si (one AstroPower and one Shell system, but the data are not clean), 1,532 for poly-Si (three BP/Solarex, one Sharp, and one Kyocera system), 1,445 for CIS/CIGS (one Global Solar and one Shell/Siemens system), and 1,273 for DJ a-Si (one BP/Solarex system). The respective "average outperformance" by Unisolar is 11%, 15%, 22%, and 38% - these percentage numbers are not much different from what ECD show in their slide. Great!

But the data from the large systems, which are monitored more reliably, show the poly-Si's average as 1,723 (ASE systems), and DJ a-Si's as 1,524 (BP/Solarex systems). Unisolar's "average overperformance" is just 2% and 15%, respectively, and, nowhere near the 18% and 41% numbers in the "Introduction."

Moreover, take a look at the yearly outperformance percentages for each small OH system, separately. Do you notice how they jump (7% to 18%) between 2005 and 2006 for each OH system, except Sharp, and then continue to increase in 2007 as well? The change of the Unisolar inverter sometime in 2005 might have had something to do with it (Sharp was the other system that had an inverter replacement that year - notice the 60% jump in 2006?). In year 2005, Unisolar outperformed mono-Si by 5%, but UNDERPERFORMED large-system poly-Si by 2%. And it performed on par with First Solar's CdTe large system (no wonder ECD, who lost the thin-film race, never mention First Solar in the entire "introductory" presentation).

But that's not all. The 2003 Tucson Annual Report reveals on page 40 that "modules are facing due south at an angle of 34 degrees from horizontal, which is latitude angle" (describing the SGS large-system set-up, but the OH set-up at Tucson is probably similar, with possibly slightly-lower tilt, as the latitude there is 32 degrees). It appears that the Unisolar modules used in this "study" were the tilted, framed US-64 panels, discontinued in November of 2007. Why would ECD's introductory presentation rely on performance data from modules they no longer make and sell? Couldn't they find some independent, side-by-side long-term performance data for the PVL systems they have been selling since 1997?

Luckily, there might be a way to estimate the hypothetical performance of Unisolar's PVL products which typically lie glued to a flat roof without any ventilation. The analysis of another dataset (Santa Cruz, CA, located at 37 degrees latitude, 600 miles west, but just 5 degrees north), again a subject of a previous post in this blog, reveals that in the period 2005-2008, the 30-degree-tilted Unisolar framed modules (US-116) outperformed the no-tilt glued-on Unisolar laminates (PVL-128) there by 24% in 2005, 26% in 2006, 27% in 2007, and 27% in 2008 (yielding 1,607, 1,573, 1,647, and 1,628if Tucson Electric had installed a system with Unisolar's current product line (the PVL series) glued to a flat roof, then, in fact, the Unisolar system could have UNDERPERFORMED all the tilted rack-mounted panel "average" systems (mono-Si, poly-Si, CdTe, Sanyo's HIT, and, yes, even CIGS and the large BP/Solarex a-Si system), except the small OH BP/Solarex a-Si system (but note that Tucson Electric complained on page 42 of their 2003 Annual Report that the "BP a-si modules have been challenging" due to issues with voltage mismatch at the inverter, while in other studies, those modules had been plagued by corrosion and delamination). kWh/kWp, respectively, or underperforming the Tucson set-up by just 3% in 2005, explained by the lower solar irradiance 5 degrees north, but by 11% in 2006 and 2007, adding further evidence that something could have gone wrong after the inverter change in 2005 at Tucson). Isn't this interesting? It turns out that

Specifically, assuming a 25% outperformance of Unisolar's tilted framed modules vs the glued-on laminates, and even ignoring the possible data contamination from the inverter change, the "average underperformance" by Unisolar's flat-roof "BIPV and BAPV" products would come to 11% vs tilted mono-Si, 8% vs small-system tilted poly-Si, 2% vs tilted CIGS, 18% vs large-system tilted poly-Si, and 8% vs large-system tilted DJ a-Si, based on the Tucson, AZ data. In other words, a glued-on BIPV/BAPV Unisolar system would be underperforming (in terms of energy yield) virtually any optimally-tilted rack-mounted glass PV rooftop system, even before the long-term degradation sets in.