Laboratory test procedures measure photovoltaic cells and modules according to international standards.
Keith Emery, National Renewable Energy Laboratory
The sun’s energy is abundant, and turning it into electrical power is not only good for the environment but also good business. In 2007, solar modules capable of producing approximately 3700 MW of energy were manufactured, resulting in about $30 billion worth of installed systems worldwide. Most installers of photovoltaics will not set up modules that have not passed stringent qualification tests and that have not had their peak watt independently rated. Therefore, representative samples are tested in certified labs around the world.
Standards are set by the International Electrotechnical Commission (IEC) and by domestic organizations such as the American Society for Testing and Materials (ASTM). Performance is tested in accordance with these standards by calibration facilities such as the National Renewable Energy Laboratory in Golden, Colo., by the National Institute of Advanced Industrial Science and Technology in Tokyo, by Fraunhofer Institute for Solar Energy Systems in Freiburg, Germany, and by the European Solar Test Installation in Ispra, Italy.
Figure 1. The workstation of the National Renewable Energy Laboratory’s large-area continuous solar simulator test bed has an optical integrator (on the right) for a 25-kW xenon arc lamp filtered to match the solar spectrum as a Class A solar simulator. The data-acquisition system monitors the light level while the current versus voltage measurements of a module under standard reporting conditions are measured.
Solar cells are measured for their peak watt rating (the maximum power output at 25 °C), 1000 Wm–2 and a reference spectrum. In photovoltaic calibration labs, there is a difference within ±2 and ±3 percent in the current at 0 volts, called “the short-circuit current agreement.” This difference was obtained by blind intercomparison at the cell and module level among the various calibration labs around the world. At the primary reference cell level, the agreement between calibration labs is closer to ±1.5 percent. The agreement should be closer because manufacturers often bin-sort at the 0.2 percent level, meaning that the photovoltaic cells are sorted according to their peak watt rating. Photovoltaic cells are often sold in terms of watts, and the manufacturers rate them at their peak wattage so as not to lose money by selling a product that actually produces more power than its rating indicates. Also, if cells were not sorted according to their power, the added performance of the better cells would not be realized in a module.
Calibration labs use the same standards to rate photovoltaic cells. The standard solar spectrum for photovoltaic applications (ASTM G159 or IEC 60904-3) soon will be changed because it was generated using an obsolete computer model and had a variety of technical deficiencies. The new spectrum (ASTM G173) is based upon an open source well-documented computer model and soon will be adopted by labs around the world.
Photovoltaic cells are typically evaluated using a 1- to 3-kW xenon arc solar simulator that produces a 10- to 30-cm beam diameter. Some simulators can produce a square beam, which is beneficial because some samples and cells are square. Photovoltaic modules, which are packaged cells, are evaluated using continuous solar simulators, pulsed xenon solar simulators or natural sunlight outdoors. A typical solar simulator consists of a light source, an optical integrator to improve the spatial nonuniformity and a custom filter package to better match the reference spectrum. Solar simulators are classified according to IEC standards as A, B or C, depending on their spectral match, spatial nonuniformity and temporal stability.
Most commercial module solar simulators are pulsed to achieve the uniformity and large-beam-area requirements. Large-area continuous simulators also are available and are required for some technologies. Their operating and maintenance costs are substantially higher than those of pulsed simulators because the bulbs do not last very long, so achieving the required spatial nonuniformity and spectral match to the reference spectrum is expensive. For some material systems, the voltage bias rate can be too large, resulting in an over- or underestimation of the power, depending on the bias direction.
Most commercial systems cannot bias in both directions to produce a hysteresis loop as a signature that the bias rate is too large. Pulsed simulators are preferred for production testing of modules because of the long bulb life and negligible sample heating, along with a close spectral match and a low spatial nonuniformity.
Measuring the spectral irradiance of solar simulators is required to calculate the spectral error. A major error source in the relative spectral irradiance is the wavelength-dependent uncertainty in the standard lamps used to calibrate the spectral radiometer. The spectral error accounts for the fact that the natural or simulated light source is not the reference spectrum and that the relative responsivity of the reference device is not the same as that of the test device (Figure 2).
Figure 2. This graphical representation shows the quantum efficiency and spectral irradiance used in calculating the spectral error, which for this example is 0.988.
Most manufacturers and researchers do not calculate the spectral error. For some research cells, this can mean a 50 percent error, although for other groups with reference cells of the same type, this is a <1 percent error. The photovoltaic calibration labs calculate the spectral error because they do not have a reference cell of exactly the same type.
Reference devices such as a pyranometer, cavity radiometer or photovoltaic can be thermal. A thermal reference device used outdoors measures the total irradiance and not the irradiance with respect to a reference spectrum. The spectral error when using a thermal detector can be quite large: >10 percent for multijunction or organic photovoltaic technologies measured outdoors under sunny conditions, or >20 percent for solar simulators. The spectral responsivity of thermal detectors is constant with wavelengths from the ultraviolet to the far-infrared, while photovoltaic devices have an increasing responsivity that drops to zero at the energy gap. The spectral error for silicon modules measured outdoors is typically 2 percent.
For this reason, photovoltaic groups use a photovoltaic reference device to measure the light level. The reference device must be stable over many years of use. Groups testing photovoltaic devices in the research arena, where the stability is questionable, use a silicon reference device. For amorphous silicon, organic or dye-sensitized photovoltaic, researchers use a silicon reference device with an IR-absorbing color glass filter, such as the KG5 from Schott North America Inc. of Duryea, Pa. Most commercial detectors saturate at 1-sun, which means that the photocurrent is not the short-circuit current because the device is resistance-limited at the short circuit. Reference cells are usually sold in packages with reference modules calibrated by a specialty lab.
The spectral responsivity of the test sample and reference device must be measured to calculate the spectral error. Because the relative responsivity can vary with light level, it must be measured with chopped monochromatic light and continuous white light. The white level is adjusted until the short-circuit current of the device under test is close to its 1-sun current. The monochromatic light typically is provided by a grating monochrometer or filter wheel using a set of 10-nm bandpass filters.
Most current amplifiers saturate around 10 mA or less, so an operational amplifier circuit is often used to convert the current to a voltage that is measured by a lock-in amplifier. Commercial detectors often saturate with interference filters because the beam power density is on the order of 10 mW/cm2 — almost 1000 times larger than the beam power in grating-based systems.
Stray light is a major source of error in single-grating systems that do not use bandpass filters, in addition to order-sorting filters for wavelengths less than about 400 nm. For filter-based systems, a blocking of 10–5 is desirable to minimize errors from light outside the passband. However, typical 10-nm bandwidth filters have a blocking of 10?4, which is further complicated because the blocking specification is for 100 percent transmittance, even though the filter often has a maximum transmittance of 50 percent or less.
Figure 3. Shown are the typical current versus voltage characteristics measured under the test bed in Figure 1.
The calibration labs output test results in a standard format (Figure 3) that is used by manufacturers in developing data sheets for use by the customer. Uncertainties in the power rating of modules are typically ±2 to 5 percent, depending on the procedures and equipment used. This represents a variation of 2 to 5 percent in the dollars per watt. Although this uncertainty is understood by the manufacturers and customers, it can be significant in an industry with a multibillion-dollar sales volume.
Meet the author
Keith Emery is manager of the photovoltaics cell and module performance section of the National Renewable Energy Laboratory in Golden, Colo.; e-mail: firstname.lastname@example.org.