The most reliable method of determining the power conversion efficiency (PCE) of a new solar cell is to send it to an accredited testing laboratory, such as the National Renewable Energy Laboratory (NREL), the Fraunhofer Institute for Solar Energy Systems or the National Institute of Advanced Industrial Science and Technology. However, this is not always a practical solution and these certification laboratories only determine the efficiencies of solar cells that have already been tested in a researcher’s own laboratory and have shown a promisingly high performance. The record efficiency “hero” cells, certified in accredited laboratories, have been tracked and the results have been published semi-annually in Progress in Photovoltaics: Research and Applications²  since 1993.³  A graphical depiction of these published hero cell performances, and earlier data, is recorded on a chart, the May 2017 version of which is shown in Figure 1.1. The most recent version of this graph can be found online,¹  including an interactive version.⁴ 

The solar cell systems described in this book fall into the category of emerging photovoltaics and include dye-sensitized solar cells (DSSCs), organic photovoltaic (OPV) cells, quantum dots, perovskite solar cells (PSCs) and the new copper zinc tin sulfide inorganic cells. Although these new photovoltaic technology platforms are identified on both the NREL and IEEE charts, the values are tabulated in Progress in Photovoltaics: Research and Applications²  and on charts as “notable exceptions” because they rarely comply with the full set of stringent tests associated with the certification process. Specifically, the active areas measured for these emerging devices are often less than the required 1 cm² and issues associated with device stability during the evaluation process might also be questionable. Regardless, the values cited serve as a valuable target for researchers aiming to improve device performance. The recognition of a researcher’s efforts on the chart is a frequent goal of many in the field of solar photovoltaics.

Small devices are commonly made in the world of emerging photovoltaics for several reasons. Small devices are often constructed on a single larger substrate yielding, for example, six or eight discrete pixel-type solar cells that can be constructed from the same set of depositions, thus providing a good range of performance statistics. The amount of active material available is often restricted or expensive and smaller devices reduce the amount of material required to evaluate the performance of the solar cell. A small device also provides an opportunity to reduce the impact of the sheet resistance on the fill factor. Within the ranks of hero cells for emerging photovoltaics, most devices have active areas, apertured or masked, of ∼0.05 cm², with only the current record DSSCs exceeding the required 1 cm.²  For example, the current record for PSCs as a “notable exception” is 22.1%, whereas at 1 cm² the PSC record is lower at 19.7%.² 

Why is it important to precisely determine the PCE of a new cell? The primary reason is to ensure that a researcher can identify advances in their own work so that new findings are discovered and can be exploited. It is also necessary for different research groups to be able to repeat published research and obtain similar results. To demonstrate and report a record efficiency, it is necessary to have the device certified by a recognized center. Making a good, reliable efficiency measurement helps to pre-screen devices that are suitable for the certification process, although this can be both costly and time consuming. For systems that have the potential to be scaled up to larger device areas or modules, it is important to know that the starting point is good.

Despite repeated efforts to caution the photovoltaic community of the need to make good device performance measurements and descriptions of the exact methods required to avoid mistakes, erroneous reporting still occurs.^5–9  Some journals have attempted to introduce itemized checklists to prevent this,¹⁰  but it is easy to make the wrong measurements and to inadvertently publish incorrect results. A common method used by researchers at NREL when reading or reviewing a new article is to determine the short circuit current, J_sc, predicted from the external quantum efficiency (EQE) data in the article and to compare this with the measured J_sc values under one sun illumination. The latter are occasionally observed to be far higher due to poor device stability or errors in determining the active device area.¹¹  These data are often shown to have J_sc values that exceed the Shockley–Queisser limit.¹²  Figure 1.2 plots the AM1.5 spectrum with three scenarios of short circuit current versus the optical bandgap: (1) EQE = 100%, corresponding to the optimum internal quantum efficiency; (2) EQE = 96%, corresponding to a reflectivity of ∼4% from the device surface; and (3) EQE = 80%, corresponding to a reasonably good device. A device with an optical bandgap of 700 nm (1.77 eV), as depicted in Figure 1.2, would have maximum short circuit currents under these three situations of 20.5, 19.6 and 16.4 mA cm⁻², respectively.

Although the device performance group at NREL is accredited by the American Association for Laboratory Accreditation to ISO-17025 standards, it was necessary to develop additional capabilities to rapidly evaluate the performance of the many new emerging photovoltaic solar cells being constructed prior to identifying hero devices suitable for certification, despite the fact that these hero devices are often certified at one of the other two accredited laboratories. This additional capability is described here, together with the rationale for its design and the protocol used to measure many different types of cells with the goal of obtaining reliable data accurate to ∼95%.

To make a good measurement of solar cell efficiency for high-quality science requires the acquisition of the following:

• a well-measured spectrum of the lamp used in the solar simulator;

• a determination of the spectral mismatch factor between the lamp spectrum, the test cell and the reference cell;

• a measurement of the illuminated device area;

• an EQE spectrum of the cell under test; and

• measurements made with care.

In establishing a new set-up to make efficiency measurements, the most central piece of equipment is generally viewed to be the solar simulator, which can easily cost many tens of thousands of dollars; however, with more care this might not be necessary. When considering a new purchase, and sometimes evaluating others’ work, many researchers focus on the class of the simulator as designated by the International Electrotechnical Commission (IEC) standard 60904-9 (there are also country-specific standards such as the American Society for Testing and Materials (ASTM) E 927-05 and the Japanese Industrial Standard (JIS) C 8912). As the class system is intended to designate how easy it will be to make a good measurement, this is not unreasonable. Using an AAA simulator does not guarantee good results, however, just as using a less well-matched simulator does not guarantee bad results. Understanding the measurements and identifying the sources and magnitudes of any errors is the most important way of reducing these to an acceptable level.

We explain here how simulators are categorized into classes. Each letter designator corresponds to a particular quality. The first letter indicates the spectral match to the reference spectrum, the second letter indicates the quality of spatial uniformity and the third letter indicates the quality of both short- and long-term temporal uniformity (Tables 1.1 and 1.2).

The spectral quality of simulators is officially evaluated from 400 to 1100 nm for historical reasons—the field was, and still is, dominated by silicon solar cells and hence all specifications are designed around the needs of that particular technology. There are many aspects of the characterization of solar cells that have the same silicon-based legacy. There are fewer photons in the UV spectrum below 400 nm, especially after transmission through the front sheet used in photovoltaic modules, and silicon is unable to convert a significant number of photons above 1100 nm. To evaluate the spectral quality from 400 to 1100 nm, six bins are generated in which the AM1.5G reference solar spectral irradiance (e.g. ASTM G173-03, Figure 1.3) is integrated.

Each bin provides a target percentage of the full irradiance. Class A spectra will have the integrated irradiance of each bin within ±25% (relative) of the target level, class B ±40% and class C −60 or +100%. It is important to note that these are integrated values, so if there is a very strong line in the light source, then it can be balanced by a strong dip (Figure 1.4).

Using this system definition, six light-emitting diodes (LEDs) in the correct bins can provide a class A spectrum, which many might argue looks nothing like the reference spectrum, although it is worth noting that many new LED simulators on the market use closer to 20 LED wavelengths and do provide an excellent match to the solar spectrum.¹⁴  In general, if a sample is not especially sensitive to a specific wavelength, this is not likely to be an issue. However, this may not be true for some devices due to, for example, layer thickness effects, narrow bandwidth absorbers such as quantum dots, the optimization of anti-reflection coatings or the presence of trap states. A light source is characterized by its worst bin. For example, if a light source is within ±10% of 400–900 nm, but has +50% variation from 900 to 1100 nm, then this would be designated as a class C spectrum. It is also important to recognize that if a bin is outside the range of class C, it may be designated X or simply not designated. However, there are many technologies that may not require light beyond 900 nm—amorphous silicon, cadmium telluride, DSSCs, OPV cells and PSCs are examples (Figure 1.5)—although PbS quantum dot solar cells do require light up to 1200 nm. Hence it should be recognized that there may be light sources that could generate a reasonable emission spectrum that might be more affordable or more durable (longer lifetimes) for certain technologies, applications or experiments such as stability tests.

The second letter designates the quality of the spatial uniformity over the illumination area. It is important to note that a class AAA simulator has ±2% uniformity. However, any product that is purchased and advertised as such does not guarantee this uniformity for every measurement, but the designation means it should be readily achievable. To achieve this level of uniformity requires calibration. The definition of uniformity varies depending on the standard used. For example, the IEC divides the designated illumination area over which the uniformity is specified into an 8×8 grid, whereas the ASTM requires one that is 6×6 with a varying detector size and packing fractions. In a research environment, cells are typically much smaller than the designated area of a particular light source and thus the spatial uniformity should be evaluated with this in mind. A monitor cell that is smaller than, or at least of similar size to, the test cell can be moved in a region to establish the level of uniformity. For example:

• if fixturing will result in samples always being measured in a fixed location relative to the light source, then a spatial map across the substrate area is advisable;

• if probes are used and a sample can be placed freely in the path of a light source, then a monitor cell can be used to establish the most spatially uniform region;

• if a substrate to be tested is designed with multiple cells, increased uniformity can be achieved by lifting the probes between each separate cell measurement and adjusting the location of the substrate such that each cell under test is at the same location for each test.

In this manner, even a simulator that has a class C (±10% uniformity) or poorer spectrum can have a potential 20% error effectively eliminated and, at a minimum, this method should remove relative errors.

Although the simplest way to establish the spatial uniformity of a light source is to take a small photodiode and scan the active (useful) area by hand, several factors should be considered. If this approach is adopted, it is helpful to ensure that the diode remains with a 0° tilt throughout the course of the measurement. If a photodiode is chosen with a package that is impervious to light, no light reaches the active area from the back or the sides. Examples of such currently available photodiodes include the low dark current silicon photodiodes from Hamamatsu: the 1.3 mm × 1.3 mm S1087 photodiode and the larger 2.8 mm × 2.8 mm S1787 photodiode are both available with cut-off filters, although they can also be added separately for use as reference cells. For most laboratory measurements, this will easily establish the level of spatial uniformity and associated error. If it is desirable to establish a particularly high-quality spatial map of a light source with a larger area (Figure 1.6)—when for instance, building a custom light source and optics system—a few different methods might be considered. Although spatial uniformity mappers can be purchased, they are expensive and such products are normally only specified for the illumination area of interest. A spatial mapper will also typically have at least one off-area cell that can be used to reduce/eliminate the effects of temporal non-uniformity.

Do-it-yourself approaches can also be made using simple x–y stages or even by reconfiguring old digital plotters. Again, if this level of effort is desired, it is wise to have a fixed location diode to help remove the effects of temporal non-uniformity. If a single diode is translated, more reliable results may be realized if this is carried out in concert with random x–y locations than if a sequential scan is used because this partially obviates the effects of cell heating.

Long-term uniformity correlates with lamp drift, whereas short-term uniformity is a measure of lamp flicker. These values are all dependent on the type of light source, the power supply and the age of the lamp. One of the most effective ways of dealing with this is to incorporate a monitor diode for in situ intensity monitoring. After calibrating the monitor diode’s output to the reference cell, which could be one and the same, it is easy to establish and then significantly reduce the effects of temporal drift. Even if the in situ monitor diode is in a region outside uniform illumination, a spatial map/correlation to individual cell locations can be utilized.

Now that we understand the most important elements in classifying a light source for solar simulation, there are a few more factors to be considered in high-quality solar simulation measurements. The next element to address is spectral mismatch. As noted earlier, light sources do not tend to share the same spectrum as any reference solar spectrum. There has been considerable progress in tunable spectra with the advent of new light sources, such as those based on LEDs. Either various colored LEDs can be tuned to match the spectral class, tuned to reduce or eliminate spectral mismatch in the standard calculation, or phosphors can be selected to closely follow the reference spectrum over certain wavelength ranges. Although considerable attention is often paid to the illumination source, an equally important element that is sometimes given little thought is the reference diode and its effect on spectral mismatch.

The role of the reference diode is to set the intensity for a one sun measurement. To accomplish this, the reference diode must have been calibrated from a known source. This can be done in coordination with one of the accreditation laboratories (e.g. NREL, the Fraunhofer Institute for Solar Energy Systems or the National Institute of Advanced Industrial Science and Technology). A calibrated reference diode will have a measured one sun short circuit current and spectral response. Ideally, a commercially obtained cell that has been packaged for enhanced stability is used. Cells intended for calibration measurements will typically include a temperature sensor in the package to enable and promote measurements at 25 °C, which is also part of the standard test conditions. The reference diode should have the same spectral responsivity as the device under test. For example, the Hamamatsu S1087 or the larger S1787 photodiode can both be used with separate color filters, such as the Schott KG and BG series. However, this can be challenging for new material development, such as some of the devices described in this book, as they tend not to be commercially available in a packaged device form and stability can be an issue.

To remove the error of spectral mismatch between the solar simulator lamp, the AM1.5G reference spectrum, the reference diode and the test cell, a spectral mismatch factor can be calculated. The spectral mismatch factor, M, is a convolution of four inputs: the solar irradiance reference spectrum (E_AM1.5); the lamp irradiance spectrum (E_Lamp); and the EQEs of the reference cell (QE_Ref) and the device being tested (QE_TestCell).⁵ 

Where

Only relative values are required because each input appears in both the numerator and denominator. It is important to integrate over the larger of the two wavelength ranges for the spectral response of either the reference diode or the test cell. Each integral corresponds to the short circuit current associated with the lamp and the solar spectrum, each measured by both the reference and the test cell.

As an extreme example, Figure 1.7 uses an OPV cell with a filtered (KG2) silicon diode and a halogen lamp spectrum to explain the importance of the mismatch factor and what is being done when setting the one sun intensity with a reference diode. Normalized curves are shown in panel (a). The filtered silicon reference cell has a similar response profile to the OPV cell, although there are some differences toward the red and blue sides of the response. There is, however, a significantly greater mismatch between the tungsten lamp and the two reference spectra. Panel (b) shows roughly what the lamp spectrum might appear to be when set by the filtered silicon reference cell to the one sun intensity relative to AM1.5G. Because the lamp spectrum is deficient in the UV region, the red overall intensity must be significantly increased to generate the one sun current from the reference cell. If a silicon reference cell was used without a filter, then the intensity of the lamp spectrum relative to the AM1.5G spectrum would be lower. This is less appropriate, however, for the OPV cell because much of the intensity is now in the red to IR region where the OPV cell does not respond, while still deficient in the blue/UV region where the OPV cell would respond.

The spectral mismatch factor is a correction factor used to adjust the current density measured after setting the one sun current on a measurement set-up for a specific test device:

Here, we divide the measured current from the test cell (I_TestCell) by the mismatch factor (M) and multiply by the measured number of suns, which gives the corrected one sun current for the device under test. The number of suns ratio is determined by measuring the actual current from the reference cell (I_{Ref, Meas}) and ratioing against the expected value (I_Ref,1sun) from calibration. This ratio is often set to unity by adjusting the height of the sample stage to make these two numbers equal by taking advantage of the divergence of the solar simulator light.

The mismatch factor (M) should, ideally, be unity. Although it is not uncommon for groups to spend large sums of money on a solar simulator light source, this does not guarantee that M = 1 because no lamp spectrum looks exactly like the reference spectrum. It is also well known that lamp spectra can drift with time.

Figure 1.8 shows the spectrum of a xenon lamp with a commercial AM1.5G filter set over the course of its lifetime. This lamp costs, at the time of publication, between US$500 and US$3000, depending on wattage, and is part of an extremely expensive solar simulator, unlike the inexpensive halogen lamp used to record the spectrum in Figure 1.7.

Notice the relative shift of red to blue over the course of the lifetime of the bulb. This is because there is a clean quartz window on the bulb early in the bulb’s life and xenon arc lamps deposit a thin layer of carbon on the window of the bulb as they age. Shorter (blue) wavelengths are transmitted less through this film than longer (red) wavelengths. Although this may appear to be a relatively minor error, it can have significant effects if there is an improper choice of reference diode.

Figure 1.9 shows the evolution of the spectral mismatch factor using either an unfiltered silicon diode or a KG5-filtered silicon diode and an OPV device.⁶  The unfiltered silicon diode begins with M ∼ 1.18; but in <1000 h it is reduced to ∼1.05, with a particularly large decrease in the region between 100 and 200 h. Hence even if spatial and temporal errors are ignored or eliminated, merely having a class A spectrum, as in this example, can result in a drift of up to 13%, which can be very misleading if not taken into consideration. Examining the curve corresponding to the right-hand axis illustrates the effect of using an appropriate choice of filtering. Choosing an appropriately matched reference cell can therefore reduce these spectral errors. In this instance, the mismatch factor ranges between ∼1.0275 and ∼1.02 with the same lamp spectral shifts.

It is generally good practice to occasionally re-measure the lamp spectrum to track the evolving mismatch factor. It is worth noting, however, that many inexpensive spectrometers are not well calibrated in terms of intensity. Although the wavelength information can be fairly close when examining calibrated spectral lines, the relative intensity at each wavelength can lead to considerable errors if the system is not carefully calibrated for intensity as well as wavelength. A spectral mismatch factor of unity would be achieved if the lamp irradiance spectrum was identical to the solar irradiance spectrum (e.g. AM1.5G) or if the spectral response of the reference cell was identical to that of the test cell. Although it is desirable to match both as best as is feasible, it is arguable that it is easier and less costly to focus on matching to the reference cell’s spectral response. For laboratories studying different absorbers, it is prudent to have multiple filter sets to pair with the reference cell. Filter sets that match well with different photovoltaic technologies include some of the Schott KG and BG color glass series.

One of the largest sources of error is poor measurements of the illuminated area of a cell under test. This is greatly exacerbated by the propensity of laboratories to make small devices. This happens for several reasons:

• research groups like statistics on relatively small area substrates;

• processes may be highly non-uniform, so small devices may enable research groups to separate dead spots from good regions; and

• higher efficiencies can be reported when using transparent conductors optimized to let more light through at the expense of higher resistance.

This latter point is illustrated in Figure 1.10, where the same spin-cast OPV system of P3HT:PCBM is processed on identical substrates with the same transparent contact and therefore the same sheet resistance; the only major difference is the active area of the two cells. This difference manifests itself primarily as a large loss in fill factor due to the higher resistance pathway of the larger device, with both the short circuit current densities and open circuit voltages remaining almost the same (Table 1.3).

Although the search for higher efficiencies tends to encourage groups toward devices with a smaller area, the potential for error increases, as illustrated in Figure 1.11. It is not uncommon for devices with areas between 1 and 25 mm² to be used. For square devices, an underestimate of the actual side lengths by only 100 µm results in a 21% error for a nominal 1 mm²  device and >6% for a nominal 10 mm² device. Length/area measurements errors can occur for a variety of reasons.

There are several different standard approaches that can be used to define the active area of a laboratory-scale device. Perhaps the most reliable method is to use mesa isolation, where a completed device is etched such that the top contact and the active layer are patterned. In this example, the area of the top contact indisputably defines the area of the device. Doing this generally requires extra processing steps, such as photolithography, and can result in carrier recombination at the edges of the device. If a less reproducible, but cheaper, method than lithography is used, such as a razor blade, measuring a device area precisely becomes far more onerous and prone to error.

An alternative method to mesa isolation is to use a patterned bottom contact and patterned top contact deposited through a shadow mask. The overlap area of the two patterns can then be used to define the area in a cross-bar pattern (Figure 1.12). In addition to avoiding edge recombination, the device can now be contacted without the concern of puncturing the top contact. Although the bottom contact may be pre-patterned in a highly reproducible manner through a process such as photolithography, defining dimensions with a shadow mask is less precise. Evaporating through a shadow mask results in much better edge definition than sputtering, but samples are generally rotated to improve uniformity and coverage, which leads to a layer with a variable thickness wider than the opening of a shadow mask. Even without rotation, the thickness of the mask and the locations in which it does not make intimate contact with the sample results in a non-uniform film that is wider than the mask’s opening. Although it is easy to either use the specified shadow mask’s dimensions or measure the opening with a pair of calipers, this would result in an underestimation of the device area.

Another, typically smaller, error in the other direction occurs from the build-up of deposited films over time. Metal electrode film thicknesses of the order of 100 nm are standard. After 100 such depositions, this can result in a 20 µm reduction in a lateral dimension of the deposition mask (10 µm on each side). It is unlikely that a greater number of depositions would occur, however, without cleaning the mask, or without natural flaking of the mask, returning it to its original dimensions.

The last method of defining the correct active area is a variant of the two previous methods. An unpatterned bottom contact is used in conjunction with an island top contact. Here the same error associated with the deposited area through the shadow mask is generated. In these previous cases, however, there was another potential source of error in the definition of area. In photovoltaic devices, transport, buffer or collection layers are commonly used between the electrode and the active material. These are typically not highly conductive relative to a transparent conductor or metal layer; however, on a length scale of 100s of micrometres they can be efficient. In certain cases, such as an OPV device, an intentionally conductive layer such as PEDOT:PSS is used. Although there is no reason to limit the conductivity of such a layer in a large-scale device, in laboratory-scale devices in which there is a strong desire to limit cross-talk and have well-defined areas suited for the accurate determination of efficiencies, using a low (lateral) conductivity for such layers is a good idea.

It is easy to underestimate the device area by inadvertently exploiting thicker or more conductive layers, which will result in an overstatement of the device’s current density. A standard hallmark of this is voltage-dependent collection in the J–V curve (low shunt resistance) (Figure 1.13) and this mechanism has been explored in detail, with an excess efficiency of ∼61% demonstrated in small area devices (4 mm²).¹⁵  These researchers strongly suggested the use of island rather than cross-bar cells.¹⁵  However, it is important to note that although such a device architecture may work for certain thin film stacks, in others it would not fully address the problem. For instance, in cadmium telluride, it is not uncommon to modify the surface with an oxidizing etch that generates a thin layer of tellurium on the surface of an otherwise highly resistive semiconductor. Such a tellurium layer can easily lead to the collection of charge 100 µm, or even further, away from a metal contact. In other systems, such as PSCs, there may be an unpatterned transport layer directly underneath the metal contact and, just as in the case of PEDOT:PSS, this can result in charge collection around even an island geometry. The “lollipop” geometry (Figure 1.12) is a compromise between island and standard overlap patterns and is intended to minimize charge collection outside the area.

Other than mesa isolation, there is only one agreed way in which a cell area can be well defined: an aperture. An aperture can be made of many materials, with the main requirements being that it is opaque and can be used with cells of any architecture. It is placed in contact with the cell being tested (or as close to it as possible in the case of a superstrate or packaged device). In its simplest form, an aperture can be made by applying an opaque tape to a device. Care is required to measure the area because vinyl tapes (such as electrical tape) can stretch, making curved instead of straight lines. A more reliable method is to use a metal mask, which can either be as simple as a drilled/reamed hole in a thin plate or a photo-etched/laser-cut pattern produced in the same manner as a shadow mask. These may be painted black or anodized to remove the influence of reflections.

In any official accreditation measurement of a sample that is not mesa isolated, an aperture is required to establish that there is no outside area collection of current. The mask containing the aperture should ideally extend fully past the edges of the sample to eliminate wave-guiding effects that might occur in a superstrate or packaged device. In a laboratory setting, however, an aperture may be used to establish that the correct short circuit current density (J_sc) is measured, although if a sample has the same J_sc with and without an aperture, then no aperture is required. If a fixture for measuring a standard patterned sample is used, custom apertures may readily be made that will align with the sample with minimal effort. This can enable a standardized, extremely high-quality measurement to be performed daily. For most technologies, an aperture that is smaller than the active area of the device is usually chosen because this allows the aperture area to be used to calculate J_sc. Manually aligning an aperture with only 100 µm tolerance can be challenging, whereas a tolerance of 400 µm is much more reasonable. If an aperture is much smaller than the device area, the shaded cell area in parallel may start to significantly lower the voltage and keeping this area to <40% of the total device area is advised. It is also worth noting that any bus-bar or metallization covered by an aperture does not adversely count against a device’s area.

Measurement of a device or aperture area may be accomplished in one of several ways. Calipers, used correctly, can provide sufficient precision and accuracy for measurement of the aperture area. Measuring a cell area without an aperture is generally achieved either with a calibrated microscope stage or using a camera in combination with a calibration standard. With a microscope and camera, the appropriate magnification is chosen, the image saved and image processing software is then used to establish the area. Alternatively, in the case of a calibrated stage, the user may focus on the edge of a device and record the stage location. Many points can be recorded to establish the perimeter of a polygon in the event of non-linear features. A simple examination of samples at high magnification and attempts to measure the area will demonstrate the irregularities and, in some cases, potential difficulties in identifying the edge of a device.

To assess the efficiency of a device, the spectral response must be known as this is needed to decide which filter must be placed on the reference diode. An absolute quantum efficiency is ideal in that it can theoretically be integrated to establish the one sun current of a cell. This value should agree closely with the one sun current from a spectrally corrected I–V measurement. Typically, the current density extracted from a spectrally corrected I–V curve is assumed to be more accurate, but the numbers should generally agree within a few percentage points. It is important to note that only a relative EQE is required to define the efficiency of a device. The difference between an absolute and relative quantum efficiency measurement is that in an absolute measurement the excitation beam is fully contained in the device’s active area. For a relative measurement, the beam’s spot is larger than the device, such that the number of incident photons is not known, merely the full spot’s relative intensity at each wavelength. This is typically carried out as a lock-in measurement, in which a monochromatic signal is chopped and used to generate an ac signal, which is then amplified for measurement.

Several considerations should be made when performing these measurements. The chopped frequency should be notably different from the line frequency (i.e. 60 Hz in the USA, 50 Hz in Europe) and prime numbers are generally good choices for the frequency (e.g. 37 Hz). The intensity from the monochromatic signal is typically ≪1% of one sun. Although a white light bias is often not performed, there are material systems for which a white light bias may alter the spectral response. An example of where this may be important is when there is a highly defective layer that may have traps filled by a white light bias. Generally, lock-in measurements of the EQE can take ∼5–15 min depending on the system and wavelength range, a time that offers an explanation as to why they are not recorded for every device being tested. In addition, for samples that are particularly unstable under light illumination, this period of time can lead to degradation during the course of the measurement.

Although technically an EQE should be known to choose the appropriate reference diode, scanning the I–V curve first in these cases and using a representative EQE from a similar, but stable, device will be reasonably accurate. A recent solution to this problem is the “flash EQE” system developed by the NREL, which enables extremely rapid quantum efficiency curves to be measured (∼1 s for each EQE curve). This type of system uses a series of different colored LEDs, each driven at a separate frequency. The signal is then deconvoluted with a Fourier transform to establish the efficiency at each wavelength. Although this is a more limited technique in that the wavelengths are fixed, it has been demonstrated with conventional systems (e.g. Si and copper indium gallium selenide solar cells), including tandem cells, to be fairly effective and reproducible.^16,17 

Although the points discussed so far in this chapter are generally the most crucial elements that are most commonly poorly performed, it is worth discussing a variety of other aspects of the measurements. The standard test conditions state that a cell should be measured at 25 °C, but cells can easily heat up to 40 °C under full illumination at one sun. Depending on the temperature coefficient, this will lower the performance to varying degrees; the temperature can be better controlled by placing the device on a cooled stage. This is a good solution for substrate cells, but superstrate cells (where light illuminates the cell through the growth substrate) are more challenging to maintain at a fixed temperature. Thermal deviations can be significantly reduced for superstrate cells by fan cooling and by closing the light source’s shutter between measurements. To establish the level of heating in a system, a thermocouple or resistive temperature detector may be mounted directly on the sample and a standard test protocol may then be performed and the temperature swings monitored. If no significant light soaking is required, it should be possible to keep even superstrate samples at or below 30 °C.

When there is more concern about temperature effects with simulators that have unused illumination areas away from the test cell, a proxy sample (i.e. a cell with a similar absorber) may be set up with a semi-permanent temperature sensor that can be monitored in real time during a measurement. In certified measurements, where it is often difficult to control the device temperature and scan rates are slow, V_oc might not actually appear on the final measured I–V curve. This is because V_oc can be measured independently of the rest of the curve and can exhibit a strong temperature coefficient. A routine can be written to monitor the rise and fall of the open circuit voltage (V_oc) while rapidly opening a mechanical shutter.

Probes are the most common method used to couple electrical signals to and from a device, although wires may also be attached directly. If a fixture is made to measure patterned samples, standard connecting methods include pogo pins and integrated circuit test clips. Such a fixture can significantly speed up measurements, especially when coupled to a multiplexor. This may ultimately enable more samples to be measured under many more measurement conditions, such as light soaking or voltage bias holds. The data described in this chapter used such an approach. A fixture can also be fitted with a temperature sensor, an in situ light monitor and an aperture. Higher versatility can be achieved with micro-manipulated probes, which are often used in a Kelvin probe configuration to remove contact resistance effects. If a four-wire or remote sensing measurement is performed, then it is important to recognize that the positive and negative probe pairs should be placed close to one another. Typically, they should be within <1 mm, but, more importantly, the resistance between a pair should be kept to <2 Ω. As samples are typically measured with a voltage bias, a non-negligible resistance between one pair of probes means that the measured voltage–current pair may not reflect what is actually seen by the device.

When physically contacting a sample, some device geometries such as the cross-bar (Figure 1.12) may enable probes to be applied with less concern about puncturing a thin film top contact. Gentle pogo pins with rounded tips and low spring forces or gently applied Kelvin probes can typically make contact with island or mesa samples. For more fragile films, placing a soft metal such as indium or gold between the sample and the contact can facilitate the connection. Certain transparent conductors such as SnO₂:F can be difficult to contact; soldering (ultrasonically) indium to the SnO₂:F can greatly ease this process. Placing a bus-bar (e.g. an indium solder line) around the border of a transparent electrode can also significantly reduce the series resistance.

To generate I–V curves, a voltage bias is applied and the current measured. Although there are many pieces of equipment that can do this, the most common choice in photovoltaic measurement laboratories is to use a source-measure unit (e.g. Keithley 2450). The ASTM E 948-05 standard specifies that the current measurement electronics should have a resolution of at least 0.02% of the maximum current (I_sc). Hence a sensitivity of 200 nA is required for a J_sc of 10 mA cm⁻² and a 0.1 cm² cell. This is readily achieved with a coaxial cable and a source meter; indeed, some source meters can measure down to 0.1 pA (a triaxial cable is required). Although such low level measurements are unlikely to affect the sensitivity of a light curve, a nanoamp or lower sensitivity can be useful in analyses of the dark saturation current density (J₀) from the dark I–V curve. Again, the standards specify that the voltage control is within 1% V_oc. This is not to say that the step size must be that small, just the control. In practice, this only requires a resolution of a few millivolts, which is not challenging (a 10 bit control is sufficient for 1 mV from 1 V full-scale).

Scans are generally performed as quickly as possible without introducing hysteresis when swept in both directions. Although some technologies are known to exhibit hysteretic behavior (e.g. DSSCs and PSCs), it is worth noting that most technologies will display some level of difference between forward (V < 0 to > V_oc) and back (>V_oc to <0) scans if a curve is swept quickly enough. With most low-hysteretic technologies, a scan rate of the order of 0.3–1 V s⁻¹ will result in minimal hysteresis.

Another approach has been adopted for highly hysteretic systems, such as some PSCs (Figure 1.14a). A quick curve may be swept in both directions and then a maximum power point region identified. After this, a slower sweep around the maximum power point region may be conducted, a voltage is identified and the current settles with time. After the device is not changing by more than a few per cent, a new voltage is applied and the process is repeated. It has become common to report not only the forward and backward scan rates for PSCs, but also a time versus maximum efficiency to demonstrate that this is a stable operating point (Figure 1.14b). After establishing a non-hysteretic I–V curve, the maximum power point, and hence the efficiency, is calculated by taking current–voltage pairs, multiplying them together and then selecting the highest pair.

There are often stability concerns with new materials that can frustrate good quality measurements of efficiency, but there are several ways of addressing this without having to delve too deeply into an understanding of the degradation mechanisms. If cells are unstable due to the presence of moisture or oxygen, a simulator may be integrated into the inert atmosphere of a glovebox, either through a window or by fully enclosing the system (Figure 1.16). Heat management can become an important issue for a fully enclosed system. Alternatively, various schema can be considered to take a sample out of a glovebox without exposing it to air. One of the more popular methods is to use a UV-curable epoxy and to seal the sample with a glass cover. This has the advantage that it allows many samples to be prepared and removed at the same time. However, it has the disadvantage that it can be messy and there is the potential for chemical interactions of the sealant with the actual test sample. Another approach is to use a re-usable “space suit”, in which a KF window blank allows light in, a KF stub is machined to allow electrical connections and the seal is made with a standard KF O-ring and clamp.¹⁹ 

This section describes the characterization of an emerging photovoltaic technology based on colloidal quantum dots:²⁰  one based on PbS and the other based on CsPbI₃. Detailed reports on the optimized performance of these two device systems have been reported elsewhere.²¹  We explore the data produced during the development of these devices, which eventually led to a hero cell with a device performance that was ultimately transferred to the NREL photovoltaic certification laboratory.

The two device stacks are as follows:

1. PbS:glass/FTO (300 nm)/TiO₂ (70 nm)/PbS quantum dots (∼400 nm)/MoO_x (15 nm)/Al (200 nm)

2. CsPbI₃:glass/FTO (300 nm)/TiO₂ (70 nm)/CsPbI₃ quantum dots (250 nm)/spiro-OMeTAD (150 nm)/MoO_x (15 nm)/Al (150–200 nm)

A schematic diagram of the structure and a cross-sectional scanning electron microscopy image of the CsPbI₃ device²¹  are shown in Figure 1.15; an almost identical structure was used for the PbS device.

The solar simulator used in these studies was originally developed for studies of organic photovoltaic devices and has been systematically improved for performance and usability. The new set-up provides multiplexed acquisition of the J–V curves of the six pixel devices on each substrate using the substrate design shown in Figure 1.12. A schematic diagram of the set-up is shown in Figure 1.16. The sample is positioned inside a nitrogen-filled glovebox with the solar simulator housed externally; the solar cells are illuminated from below through an optical window. Similarly, for the EQE measurements, the sample is located inside the glovebox with the light source and electronics located below.

It is first necessary to examine the type of filter to be placed on the reference silicon cell to more closely match its spectral response to that of the test cell; this requires a knowledge of the EQE spectrum of the new device material. Although this spectrum could be estimated from the sample absorption spectrum, it is preferable to determine the EQE spectrum directly. Figure 1.17 shows that the PbS device has a response with a peak EQE of ∼80% at 500 nm, which extends into the IR with an optical bandgap at 1100 nm. The CsPbI₃ device has a peak EQE at ∼450 nm and extends to the optical band edge at 700 nm. No filter was required to match the response of the silicon reference cell to that of the test cells when measuring the J–V curves for the PbS devices, but the shorter wavelength bandgap of the CsPbI₃ devices required a KG2 filter to be placed in front of the silicon reference cell. It should be noted that the unfiltered silicon reference cell is only just suitable for the PbS solar cell because the response extends significantly into the IR. Having completed this task once, it can be reasonably assumed that these filter combinations can be used for future measurements on the same quantum dot systems.

The mismatch factor (M) requires EQE data for not only the solar cell being tested, but also the EQE for the reference cell with an appropriate filter to set the one sun intensity. Figure 1.18 shows four individual plots of lamp irradiance, E_Lamp(λ), and the AM1.5G irradiance, E_AM1.5(λ), compared with the corresponding product of spectral responsivity for the PbS test cell, S_TestCell(λ), and the reference silicon cell, S_Ref(λ). Each plot also contains the integrated area under the E(λ)S(λ) product, the peak value of which corresponds to the integrals defining the mismatch factor. Similar data for the CsPbI₃ solar cells are shown in Figure 1.19. The key differences are the extended response of the PbS sample in the 700–1150 nm spectral window and the correspondingly larger integrated short circuit currents determined for the silicon reference cell in the absence of the KG2 filter. Relative to determining the J–V curves for new test cells, the EQE spectra take longer to record and the relative spectra for individual cells often change very little, resulting in far fewer measurements of this type. As discussed earlier (Figure 1.8), the effects of lamp aging can induce significant changes in the mismatch factor if the reference cell’s spectral responsivity, S_Ref(λ), is not chosen carefully.

Data points in the EQE spectra are often determined at large wavelength intervals. For example, the data shown in Figure 1.17 has values that are taken every 10 nm, which is acceptable providing that the spectra do not exhibit sharp features. It should be noted that the irradiance spectra of both the AM1.5 spectrum and the xenon lamp both exhibit many sharp spectroscopic features. To avoid errors when determining the irradiance product with spectral responsivity, E(λ)S(λ), the number of data points in the EQE spectrum should be increased to match the resolution of the irradiance data by interpolating between known values in the spectral responsivity data. The sharp spectral features in the E_x(λ)S_x(λ) product shown in Figures 1.18 and 1.19 are evidence of the use of interpolation to retain the sharp features of the irradiance spectra.

The mismatch factor, M, is calculated using eqn (1.1), which can be written as:

where A and B represent the integrated short circuit currents for the silicon reference cell with AM1.5 and xenon lamp irradiation, respectively. Similarly, C and D correspond to the xenon lamp and AM1.5 irradiation for the new test cell. Using these data, the mismatch factors are:

and

Note that both mismatch factors are >1, which will result in a corrected short circuit current (I_corr in eqn (1.3)) that is less than that measured directly using the solar simulator. The 5% correction is only just acceptable and, if the test cell should have exhibited a promising performance, the certification laboratory would probably be required to find a different reference filter or even a different reference cell (e.g. GaAs) to reduce this correction to a mismatch factor closer to 1.

Measuring the J–V curves is the final stage of the process. A silicon reference cell (Hamamatsu S1787-08) of known active area is placed on a simulator stage. Eqn (1.3) (I_corr) suggests that the measured current density is used to ratio against what is expected from a one sun AM1.5 spectrum. However, an approach is to adjust the height of the sample mount stage such that the current obtained from the reference cell matches that expected from one sun. Prior to loading the six pixel substrates into the solar simulator mount, a thin, pre-machined stainless-steel mask with six precision apertures (area 0.059 cm²) defining the active illumination area of each pixel is first placed in the mount. The substrate is then placed on top of the mask with the transparent electrode, FTO in the two examples here, facing down toward the illumination source (a superstrate sample). Metal pogo pins located in a single anodized aluminum block contact with the six metal electrodes with pressure pins, with two additional pins on either end of the transparent conductor ‘race-track’ electrode (Figure 1.12), which are shorted in this detection configuration. The J–V curves for each of the six pixels are measured separately using a source meter. The NREL system has all six devices connected to the source meter via a multiplexer, which cycles through 1–6 in sequence. The data for the PbS and CsPbI₃ QD devices are shown in Figure 1.20.

Tables 1.4 and 1.5 give the data extracted from each of these curves, along with the average of each of the four characteristic parameters: V_oc, J_sc, fill factor and efficiency. The open circuit voltage (V_oc) and short circuit current (J_sc) are obtained by crossing the two axes again using interpolation to identify V_oc and applying a zero voltage to identify J_sc. The fill factor (FF) is identified from the voltage (V_max) and current (J_max) at the maximum power point (P_max), where d(JV)/dV = 0 and FF = P_max/(V_ocJ_sc). The efficiency is calculated from V_ocJ_scFF/P_in, where P_in = 100 mW cm⁻².

Therefore the solar PCE of the PbS device is (correcting for the mismatch factor) 6.6 ± 0.6% and the solar PCE of the CsPbI₃ is 6.8 ± 0.6%. The short circuit currents measured from integrating the EQE data are 21.5 mA cm⁻² for the PbS device and 11.1 mA cm⁻² for the CsPbI₃ device. Both results are in reasonable agreement with the currents measured from the J–V curves. These two solar cell devices—the best optimized devices—have yielded uncertified efficiencies of 8.3 ± 0.6 and 13.4 ± 0.6%²¹  for PbS and CsPbI₃, respectively, although these results continue to improve.

Measuring the PCE or performance of a new solar cell is a relatively simple procedure, but the process can easily lead to erroneous results that can mislead research directions and foil opportunities to publicize champion solar cell performances. These erroneous results can be minimized by using the following prescribed measurement methodology. The protocol is straightforward:

• measure the EQE spectrum;

• measure a calibrated lamp spectrum from the solar simulator;

• determine the spectral mismatch factor;

• identify the correct reference cell and filter;

• use an aperture test cell to define the illuminated active area;

• measure the J–V curves to avoid the effects of any hysteresis;

• take note of sample degradation issues; and

• avoid the temptation to artificially enhance efficiency data.

Obtaining performance data that is 95% reliable can help accurate device optimization and may result in a device being certified at one of accreditation laboratories, with champion efficiencies being published and ultimately referenced on the NREL research cell efficiency chart.

We thank Yang Yang (UCLA), Kai Zhu (NREL) and Joey Luther (NREL) for permission to reproduce their data in Figures 1.9, 1.14 and 1.15, respectively. We also thank Tom Moriarty (NREL) and Moritz Riede (Oxford) for the original lamp spectra in Figure 1.4. ARM acknowledges support from the Center for Advanced Solar Photophysics, an Energy Frontier Research Center funded by the US Department of Energy, Office of Science, Office of Basic Energy Sciences. MR acknowledges support from the US Department of Energy, Office of Energy Efficiency and Renewable Energy, Solar Energy Technology Program. GR acknowledges support from the Solar Photochemistry Program of the US Department of Energy, Office of Science, Basic Energy Sciences, Division of Chemical Sciences, Geosciences and Biosciences. All support was under Contract No. DE-AC36-08-GO28308 to NREL.