Photovoltaic devices are usually characterised using simulated sunlight. A solar simulator consists essentially on a light source, often a Xe arc lamp, and a set of filters with the aim that the spectral distribution of the emitted light is as close as possible to the standard AM1.5g spectrum, and its integrated intensity is 1000 . Figure 1 shows the AM1.5g solar spectrum at an integrated intensity of 1000 , together with an idealised solar simulator spectrum using the irradiance spectrum of a black body at 5700 K.
In the day-to-day lab life, setting up your experiment so it matches these two requirements (spectral distribution and integrated intensity) is done by first, getting an spectral-class A solar simulator to have the peace of mind that the spectral distribution of the emitted light is quite close to the AM1.5g spectrum and, second, using a calibrated reference solar cell to set the intensity of the lamp in your simulator and the distance between the simulator and your devices so the light intensity that reaches the cells is 1000 .
Spectral class A means that the integrated intensity of the light emitted by the solar simulator falls within 25% of the integrated intensity of the AM1.5g standard spectrum when considered in wavelenegth intervals. That is, as the integrated irradiance of the AM1.5g spetrum in the wavelength interval from 500 to 600 nm is 19.9% of the total spectral irradiance (i.e. of the 1000 ), the integrated irradiance of the solar simulator in the same wavelength interval can have any value from 14.9% to 24.9% of its total integrated irradiance. If this condition is met for a set of wavelength intervals in the 300 to 1400 nm range the solar simulator can be considered Class A in terms of spectral match to the AM1.5g. Note that a Spectral Class A solar simulator, even though it is in the top-tier of the solar simulator classification, can still be significantly different to the AM1.5g spectrum.
To verify that, in addition to having the correct spectral shape, the light intensity falling on the solar cell under test is actually 1000 , a reference solar cell is usually employed whose yield in terms of power conversion efficiency (PCE) and short-circuit current () is calibrated by the manufacturer against the AM1.5g spectrum (or as close to that as humanely possible). In other words, when you buy a reference solar cell, the manufacturer tells you how much and PCE it should yield under AM1.5g at 1000 . Accordingly you adjust the distance between reference cell and simulator, as well as the intensity of the solar simulator lamp so the reference cell yields the appropriate and PCE.
Most often, Si solar cells are used as reference cells and that is mostly fine whenever the devices under study have a spectral response similar to that of Si solar cells, especially in terms of energy band gap , which for Si is around 1100 nm. However, when characterising organic cells, which tend to have larger energy band gaps, using a Si reference cell can lead to large measurement errors. In the following paragraphs I will try to illustrate how these errors can arise.
Let’s assume that we want to compare the results be obtain with the ideal solar simulator of Figure 1, with what two other research groups A and B get using their solar simulators, which have slightly different emission spectra in the range of interest (300-1100 nm for a Si reference cell). For the record, I have produced these two spectra by adding a straight line with a positive/negative slope to the emission spectrum of the ideal simulator (Figure 2).
In a real situation, we would set the lamp intensities and simulator-cell distances so the Si reference cell yields in all three labs the same provided by the reference cell manufacturer. This is equivalent to multiplying the solar simulator spectrum with the external quantum efficiency (EQE) of the reference cell (an example of which is shown with a black line in Figure 3), and integrate over the whole wavelength range to obtain . If we do this process with the spectra of simulators A and B in Figure 2, we end up with the relative intensities of the three spectra shown in Figure 4. Simulator A has a weaker intensity from 400 to, say, 750 nm that is compensated by its more intense emission from 750 to 1100 nm, and viceversa with Simulator B, so the reference solar cell gives the same or PCE in both systems.
Now that the systems are set up, let’s characterise our organic solar cell. Let’s assume, for example, that it is based on PTB7 polymer, whose energy band gap is something around 750 nm. Our organic solar cell only sees light from 300 to 750 nm, and is transparent from roughly 750 nm onwards. This is illustrated with a yellow-shaded area in Figure 4. How close to each other (and to the ideal simulator and the AM1.5g standard, for that matter) are both simulators A and B in terms of the emission spectrum in the relevant range? Figure 4 shows the spectrum of Simulator A is below that of Simulator B on the whole range considered, i.e. there is a clear bias in favor of Simulator B even though all simulators were calibrated with a Si cell to give 1000 . In the example of Figure 4, there is indeed a 12% difference in integrated area between both curves in the yellow-shaded range. So, our friends from Lab B will consistently get 12% better results from their PTB7-based cells than our friends from Lab A, and also better than what they would get under the actual AM1.5g spectrum (or, equivalently, our ideal solar simulator).
To solve or at least minimise this discrepancy, there are several strategies that can be followed. One of them is the use of the so-called mismatch factor. This is a factor that results from comparing the short circuit current that the device under test and the reference cell would give when illuminated by the AM1.5g standard on one hand and by a particular solar simulator on the other. This factor can be multiplied to the resulting from the experiment to get a corrected value. The closer the resulting mismatch factor is to unity, the closer the obtained results would be to what we would have obtained if the cell had been measured under the AM1.5g standard itself. On the contrary, the more different the spectral distribution of the emitted simulated sunlight and the AM1.5g are, and/or the more different the spectral responsivity of the reference cell and the device under test are, the more different from unity the mismatch factor would be, and correspondingly, the larger the correction that would have to be made to .
Another, quite practical way of reducing the mismatch errors consists on using a reference cell for the calibration of the solar simulator lamp intensity that is better suited for the particular device under test. As said above, the mismatch factor compares the light sources and also the spectral responsivity of the reference and test cells. The former is something about which labs normally cannot do much: once you buy your Class A solar simulator, that’s about it unless you can get a new filter that makes its emission spectrum even closer to the AM1.5g standard. However, if the spectral responsivity of your Si solar cell is not similar to that of the device you are trying to characterise, you can always use a different reference cell whose spectral responsivity matches better that of your device. This results in a experimental setup whose mismatch factor is closer to unity.
A usual alternative reference cell employed often -but by no means always- in the organic photovoltaic community is a Si reference cell with a KG5 optical filter that basically elminates any light of wavelenghts above 750 nm, the light not seen by PTB7-based devices. Using such a cell to set the intensity of our simulators A and B, again by multiplying the emission spectra by the EQE of the KG5-filtered reference cell (red line in Figure 3), integrating, and adjusting the intensities until all simulators give the same , we end up with the results of Figure 5. The light intensity for wavelengths above 750 nm is very different in both systems now, but we don’t really care because our PTB7-based solar cell does not see that light at all. However, in the wavelength range of interest, the spectral match is much better now, and the integrated intensity difference between A, B, and the ideal simulator (which is very, very similar to the AM1.5g spectrum) is now only 1%.
Is using a calibrated, KG5-filtered reference cell the optimal solution? Well, depends on the particular solar simulator and device under test. If your simulator is excellent and matches perfectly the AM1.5g spectrum, just like the ideal one of Figure 1, it would not matter what reference cell you use, although finding such a solar simulator is not a trivial task. If your simulator is not perfect, using a KG5-filtered Si solar cell will improve your mismatch factor, but may not be enough to make it exactly equal to unity. On the other hand, a KG5-filtered calibrated solar cell is useful if its spectral responsivity is close to that of the device under test. Otherwise, another reference whose spectral responsivity matches the device under test has to be found (like, for example, a Si cell filtered with a different optical filter).
References and further reading.
Toward reliable and accurate evaluation of polymer solar cells based on low band gap polymers, Long Ye et al., J. Mater. Chem. C, 3, 564 (2015)
Measurement and Characterization of Solar Cells and Modules, by K. Emery, in Handbook of Photovoltaic Science and Engineering, eds. A. Luque and D. Hegedus, John Wiley and Sons, West Sussex (2003)
Accurate Measurement and Characterization of Organic Solar Cells, V. Shrotriya et al., Adv. Funct. Mater., 16, 2016 (2006)