Extraction and microscopic analysis of partial shading‐induced defects in a commercial CIGS PV module

The ever‐increasing instalment capacity of Cu (In, Ga)(Se, S)2‐based photovoltaics calls for a better understanding and control of their reliability. In this paper, we show how using a coring‐based method, small samples can be extracted from full size commercial modules, and prepared for lab‐scale analysis. The method is applied to a Cu (In, Ga)(Se, S)2 (CIGS) module where a non‐reversible, propagating ('wormlike') defect has been created in a controlled partial shading experiment. Through current–voltage, photoluminescence and illuminated lock‐in thermography analyses on an undamaged part of the module, the method used is shown to yield fully functional, undamaged active cells, with a photovoltaic conversion efficiency above the full module efficiency. Where the wormlike defects were present, a typical strong shunting behaviour is observed, as well as an increased sulphur content near the edge of the wormtrails. Furthermore, the wormlike defect propagation is shown to be strongly influenced by the present of specific features near the interconnects, which could be the result of manufacturing. These results demonstrate the potential of coring to analyse module failure with all the laboratory tools available. They also shed some light on how wormlike defects, which are a rare but serious hazard for CIGS modules reliability, can form and propagate in commercial, monolithically interconnected modules.

Keywords: CIGS; coring; illuminated lock‐in thermography (ILIT); partial shading; photoluminescence; reliability; wormlike defects

Small samples are extracted by coring from a commercial CIGS module degraded by partial shading, and unpackaged for microscopic analysis, yielding undamaged cells as demonstrated by IV. The shunting ('wormlike') defects formed by partial shading have sulphur‐rich edges and their propagation pathway is influenced by pre‐existing nodules on the interconnect.

INTRODUCTION

CIGS is a well‐established PV technology, with an estimated 1.9 GWp produced in 20171 and plans for larger production in the future.2,3 Its range of applications includes not only PV field installations but also cladding for the built‐in environment,4 where it has the advantage over crystalline silicon that it can be made sufficiently flexible to fit on surfaces of any shapes (e.g., building facades and roofs).5,6 However, one crucial requirement for the bankability of any PV technology is reliability: CIGS must withstand the various stresses present in outdoor operation, at least within its designed operational lifetime. In this respect, high stability in outdoor conditions7 as well as in accelerated ageing tests8 and field tests9 has been demonstrated for some CIGS modules. However, as the base CIGS material and full device stack evolve over the years to achieve higher efficiencies, the reliability needs to be continuously monitored. The cause behind the degradation of a PV device is usually studied in detail at the laboratory level, on dedicated samples (cells or small modules) which can differ significantly from commercial devices. On the other hand, the characterisation of field‐degraded, full‐size modules is limited to the techniques that can be used at this scale (e.g., I‐V and electroluminescence). Neither of these can draw a complete picture of the degradation mechanisms occurring in the modules.

The most relevant method to overcome this issue is therefore to extract samples from commercial modules exposed in the field ('coring') and prepare them to be analysed in the lab. A similar extraction method has been reported in the past and applied to study local features, for example, by electroluminescence and electron backscatter diffraction,10,11 also on c‐Si.12 Here, we go a step further and demonstrate that the coring and, more importantly, the unpackaging stage, can be carried out without any detectable damage to the active layers, yielding fully functional PV devices. To demonstrate the potential of the preparation method, we apply it to extract, prepare and study samples from a module degraded by partial shading.

Partial shading can occur in PV fields (e.g., from self‐shading or cleaning equipment) and in domestic roof installations (e.g., trees), but is even more likely in integrated applications (e.g., BIPV).13–15 Apart from the temporary power loss that such shading can induce,16,17 it can also lead, in very specific scenarios, to wormlike defects,13–15,18–21 reducing irreversibly the performance of the modules. These defects must be better understood to be mitigated or even eliminated.

Such wormlike defects have been studied in the past. Bakker et al. carried out a comprehensive study on wormlike defects generated on lab‐scale samples by reverse bias.13 Johnston et al. also used reverse bias to simulate partial shading and study breakdown and wormlike defect formation.22,23 Lee et al. combined a macroscopic study of wormlike defects on commercial modules with a lab‐scale study of similar defects on their own samples.14 However, such studies cannot, for example, reveal the interactions between the microscopic features specific to a commercial module (e.g., the interconnects) and the wormlike defect formation and/or propagation.

The present study therefore pursues two aims. First, by applying the extraction and preparation methods developed internally to a commercial CIGS module degraded by partial shading, we aim to study in detail realistic wormlike defect formation and propagation in such a commercial module. Second, this study is also a means to assess the value of the coring‐unpackaging method for failure analysis.

EXPERIMENTAL METHODS

Module

The module used in this study is a full‐size commercial Cu (In, Ga)(S, Se)2 module grown via the sequential route. It was never deployed in the field. The exact time between manufacturing and receiving it (from a third party) is not exactly known, but less than 3 years, followed by storage at our facilities for 12 months, before stressing and testing. The full module with wormlike defects, still framed and encapsulated, was then stored another 2 years before the samples studied here were extracted and characterised. The encapsulant was ethylene vinyl acetate (EVA). Scanning electron micrographs (SEM) suggest that the P1 and P2 scribes were done by laser, while the P3 scribe was mechanical (see Figure 15). Cross section energy dispersive X‐Ray spectroscopy (EDS) is shown in supporting information (Figure S1).

The cells of this module are monolithically interconnected and without any bypass diode.

Generation of the wormlike defects by partial shading

The module studied underwent two shading experiments prior to coring and characterisation. In the first one, an increasing proportion of the module was shaded, in both portrait or landscape direction ('incremental shading'). This is reported in Tzikas et al.24 ('CIGS‐G' module in the paper). However, this experiment did not yield any detectable change in the current–voltage (I‐V) nor the electroluminescence (EL) response of the module. We will discuss in Section 3.4.6 whether those prior experiments might actually have had an effect, only detectable at the microscopic scale.

The shading stress that gave rise to the wormlike defects was carried out at short circuit by covering 10% of the module area (around 14 cells) with an opaque mask extending the whole length of the cells, for 10 min under AM1.5 illumination. This is illustrated schematically in Figure 1.

We must point out that the conditions we simulate here mostly occur in specific field conditions: The shading is more likely to cause wormlike defects if the shading is in the portrait direction, with a strong difference in irradiance between shaded and non‐shaded part. Those conditions can be met during module cleaning, for example, as shown in previous studies.18,25

Characterisation of the module

I‐V measurements on the modules, before and after partial shading, were carried out in an Eternal Sun solar simulator with Rera IVtraQ software. The details can be found in Tzikas et al.24

Electroluminescence of the module after partial shading was carried out by the company SolarTester BV in their custom‐made setup. A current of 0.593A corresponding to around 25% of Isc and a voltage of 74.087 V were applied. The image was acquired with a 5‐s acquisition time.

For the specifications of the electroluminescence before shading, see Tzikas et al.24

Coring and sample preparation

Coring

Using the coordinates of the EL measurements, the locations of interest were determined and the samples extracted and unpackaged. Note that, although the present work focuses on two of those samples (with and without wormlike defects), over 20 samples in total were successfully extracted and prepared.

The coring was undertaken using a waterjet cutter, where the water present in the tray and the waterjet prevented material particles emitted during the coring from becoming airborne.

The coring was carried out from the rear side of the module until the core containing the entire stack rear glass/Mo/CIGS/TCO/front glass came loose. Note that the front glass was made of tempered glass, whose internal stress needs to be relieved in order to limit the cracking of the substrate glass in subsequent drillings, as those can damage the active layers.

By the end of the coring session, the front tempered glass was cracked all over (due to stress release) and a few cracks were also present in the substrate glass.

Removal of front glass and encapsulant

The next step was to remove the front glass and encapsulant, in order to analyse and electrically contact the device ('unpackaging'). This step is done by applying a combination of thermal and mechanical stresses, in a variation to what is described in Moutinho Johnston et al.10 This is the most challenging part of the sample preparation, as commercial modules can have different types of encapsulants with different thermo‐mechanical properties. Additionally, the adhesion between the different layers of the cell stack play a key role in a successful unpackaging. Therefore, this procedure, especially the temperature applied at the encapsulant (here, EVA), needs to be tuned to the different manufacturers and to the encapsulant they use.

The surface of the TCO uncovered by this procedure was clean and without any signs of local delamination or damage. As mentioned before, more samples than those presented here were extracted and prepared, all with similar success.

Mini‐module contacting

The worm‐free core was made into a working mini‐module by applying silver paint on the TCO on the positive side and, after scraping off the CIGS/TCO, on the Mo on the negative side. Note that the commercial module used in this work showed a large amount of MoSe2 at the interface between the CIGS and the Mo, which caused a very high series resistance. This MoSe2 therefore had to be removed before depositing Ag. This was done by laser. The power used was 0.138 W at a wavelength of 1026 nm with a pulse width of 250 fs. The process was carried out in focus, with a Sill S4LFT7163 scan lens. The scan velocity was 500 mm/s with a line distance of 15 um, the spot size ~30 um. The laser induced a periodic surface structure (LIPSS) which is observed as a colour shift when looking at the sample in reflection.

Etching

The etching of the ZnO layer was achieved with a 3% vol. acetic acid solution for various durations, followed by rinsing in deionised water and regularly monitoring the PL image and EDS composition, in a similar procedure to Bakker et al.13

Characterisation of the extracted samples

I‐V measurements on the undegraded cored sample ('NW') were undertaken in a Neonsee setup.

Confocal microscopy mapping was carried out using a Leica DCM 3D microscope, with red LED excitation.

The photographs of cored out samples were acquired using a commercial scanner.

Raman measurements were undertaken in a LabRAM Aramis from Horiba equipped with a 532 nm wavelength (green) diode‐pumped solid‐state (DPSS) laser, at 10% of its total power, yielding a 0.6 mW power at the sample, with a spot diameter between 1 μm and 3 μm. This setup was also used for the optical microscopy of Figure 7.

The photoluminescence of the samples was imaged using a Greateyes setup. The luminescence was measured with a Greateyes camera equipped with a 950 nm long‐pass filter, with 20 s integration time and an aperture of f/2.8. The LED blocks used for excitation are powered using a TDK Lambda Gen‐100 power supply, set at 1000 mA and 220 V. For all measurements reported here, the LED blocks has a peak emission wavelength at 660 nm, yielding an illuminance of 800 W/m2 at the sample.

An Infra Tec setup was used to measure the illuminated lock‐in thermography. The LEDs have an emission peak wavelength of 860 nm. The ILIT measurements are carried out using a TDK Lambda power supply; 16 V and 20 A are applied to the infrared LEDs, which generates an irradiance of around 110 W/m2.

SEM AND EDS MEASUREMENTS WERE CARRIED OUT IN A JEOL INTOUCH SCOPE JSM‐6010LA AT 20 KV ACCELER...

Current–voltage on full module

The current–voltage characteristics before any shading experiment (see Section 2.2) and after the shading experiment studied here is shown in Figure 2. The wormlike defects result in a relatively small (<3% relative) power decrease, driven by an open‐circuit voltage (Voc) decrease. This Voc decrease is consistent with other references in the literature.15,25

Electroluminescence on full module

The electroluminescence imaging of the module before and after shading is shown in Figure 3. Several shunts, visible as darker regions with a brighter corona around them, are already present before shading. The main difference after shading, is the apparition of a strongly shunting area in the shaded edge of the module (boxed in blue). Taking a closer look at this shunted area, some features extending from the interconnects (insert) are visible with the naked eye. Those have already been observed in the past on mini‐modules13 and are usually referred to as 'wormlike defects', owing to their shape. Two cores, shown in Figure 3, are studied in this work: the one extracted from this 'wormy' area ('W', half‐moon shaped), and another core in a location without such wormlike features or indeed any visible EL feature ('NW', round‐shaped).

Worm‐free core (NW)

The ILIT and PL images of NW are shown in Figure 4. At first glance (Figure 4A), the PL image looks homogeneous and no significant defect is visible. Looking more closely, however, one can notice some small point‐like features, dark in PL (Figure 4B) but heating up in ILIT (Figure 4C). Those features are very unevenly spread across the sample's surface and the rest of the cell still emits well in PL. This means that these point‐like features are either non‐shunting or very mildly shunting, that is, only drain the carriers very locally.

To identify these features, the locations where they were visible were analysed by SEM/EDS and compared to featureless areas. Point‐defects similar to the one shown in Figure 5 were systematically present in the areas lighting up in ILIT. Confocal microscopy shows that these features protrude from the surface by 3–5 μm, hence much more than the total thickness of the active layers themselves. EDS (Figure 5A) indicate that these are not contamination (no increased C signal) nor produced during the unpackaging, but rather covered in ZnO like the rest of the bulk.

These features resemble the nodules reported by Palmiotti et al., which they identified as originating from voids and cracks located in the CIGS layer.15 In our module, however, the nodules are protruding by a much larger amount than in this reference and it is unclear whether they result from the manufacturing process or they are an early form of degradation. Note that those defects are also found in large amount on top of the P1 scribe (see discussion in Section 3.4.6).

After SEM and confocal imaging, the core NW was then contacted and a rectangular area including three monolithically interconnected cells was scribed inside. I‐V was then carried out on this isolated device (Figure 6). Some small differences can be observed between the module and the extracted sample, such as a slightly lower fill factor (FF) and slightly higher short‐circuit current (Jsc). However, they are minor and can be explained by the sample preparation carried out on the cored‐out sample: the slight FF decrease (through Rs increase) is likely due to the Mo thinning out after MoSe2 removal by laser, while the Jsc increase can be similarly explained by the reduced reflection losses after removal of the front glass present in the module. It is also possible that slight lateral inhomogeneities in electrical properties partially explain these differences.

These slight differences aside, the efficiency obtained, 15.2%, is consistent with the active area efficiency of the module before the partial shading shown in Figure 2 (14.6%).

This is a very important result, because it means that the coring process does not induce any measurable damage to the device, and can therefore be applied to studying the local opto‐electronic properties of modules degraded in the field.

Summarising these findings:

• the coring and unpackaging procedures did not damage the TCO or the underlying layers.
• the features observed by PL and ILIT are benign with respect to performance

Core with wormlike defects

The half‐moon core 'W' is the main focus of the study. This sample was extracted from an area showing strong shunting in EL induced by partial shading (Figure 3).

Optical and confocal

Since the wormlike features were visible on the surface of sample W with the naked eye, optical characterisation was first carried out at a low/medium magnification, as shown in Figure 7. The wormlike defects can be seen in the lower two cells in the figure, as white swirls near the interconnects. The other cells are free of such features. Although the wormtrails seem to be randomly propagating, they always loop back to the interconnects, as already reported by Bakker et al.18 This is because the interconnects are the dominant source of current, thus any pre‐existing defect close to the interconnect will be fed more carriers and heat up quicker and more than those further away. For a more detailed discussion on the mechanism, see Section 3.4.2.

Confocal microscopy (Figure 8) shows that those features protrude from the surface by over 1 μm, which is comparable to the underlying film thickness. This is consistent with the SEM cross‐section data of Lee et al.14 and Palmiotti et al.,15 where the CIGS had expanded and become porous, leading to the morphology we identify as wormlike defects. The wormlike defects we observe in the present work do not, unlike previous studies,13,26,27 consist of coalescing islands, but instead seem to have a more continuous structure, with a rather smooth ridge.

Bakker et al have shown in the past that, on lab‐scale cells, thicker TCO result in larger islands, more distant from one another.26 Although the NW and W samples have a similar TCO thickness to the 'thick TCO' case described in Bakker et al.,26 the wormtrail morphology in the present work does not match these characteristics, with narrower and more coalescing islands. The reasons for this discrepancy cannot be elucidated from the present data, although it is likely that the local temperature and its propagation are influenced by the much higher currents and additional layers (e.g., encapsulant) present in the module, compared with a lab‐made cell. Another key difference is that S is here present in the absorber, whereas in Bakker et al.,26 the absorber was Cu (In, Ga) Se2. As will be shown in Section 3.4.5, sulphur seems to play an important role in the wormlike defect formation.

Imaging PL and ILIT

To study the shunting behaviour of wormlike features further, the sample cored out from the wormy area (W) was analysed by imaging PL and ILIT. No PL signal is detected from the two cells where ILIT shows presence of wormlike defects (Figure 9A,B). Zooming in on the thermal image (Figure 9C), we can see how the wormlike defects systematically connect to the P1 scribe, in agreement with Westin et al.27

This image also shows that the wormtrails can have very different morphologies, some coiling up tightly and other travelling much longer distances. One possible explanation is that some defects cause significantly more shunting than others. When this happens, the energy density (i.e., heat, current etc.) is concentrated around the hotspot, so that the propagation remains in the vicinity of this strongly shunting defect. When the defects are less shunting, the hotspot hopping can occur over larger distances and the wormlike defect's path uncoils.

To understand the mechanism driving the wormtrails towards the P1, let us look at the effect of a wormlike defect on the band diagram: Figure 10 illustrates, through the voltage profile and electron flow, what happens when a shaded cell is located between two illuminated cells. In the shaded cell (cell 2), there is no electron–hole generation, the only source of current is from adjacent (non‐shaded) cells. This causes a reverse bias. The electrons can only be supplied by cell 1, via the P2, while the holes can be provided, also via the P2, by cell 3 (Figure 10B). When a wormlike defect forms, the locally lower shunt resistance provides a path through the absorber, and the electrons present in the Mo start flowing through the defect to the TCO of cell 2, and finally the Mo of cell 3, where they can recombine with the holes. The wormlike defect is effectively a source of electrons. Since the holes are located in the Mo of cell 3, the flow of electrons is the largest between the original (wormlike) defect and the P2 with cell 3, over the P1. The consequence for the propagation of the wormlike defect of the mechanism shown in Figure 10 is that a potential hotspot (i.e., nucleation point) located further away from the P1 than the original shunt will be supplied with less current than one located closer to the P1. The next nucleation point will therefore be more likely to form closer to the interconnect with cell 3 than the initial worm, that is, towards the P1 side.

Raman inside the worm

The optical images acquired in the Raman setup (Figure 11A) show several distinct regions outside, inside and at the interface with the wormlike defects:

• Interior of the wormlike defect (Region I)
• Edge of the wormlike defect: darker stripe at the interface between the bulk and the wormlike feature (Region II)
• Away from the wormlike defect (Region III): bulk of the sample, without wormlike defect

Raman spectroscopy was carried out along those different regions (Figure 11B–D). Note that the Raman spectroscopy was carried out at regular intervals and not only at the points plotted here, however we focus here on the 'stable' regions, where no two regions overlap in the laser spot.

The three regions mentioned above clearly show a distinct Raman signal. The wormlike defect interior (Region I, Figure 11B) yields a series of broad peaks, especially around 240 cm−1, which could not be unambiguously matched to a known phase. MoSe2 A1g is a potential candidate for this peak,28,29 although that would require the wormlike defect to be so porous that the laser can reach through to the back contact. The large fluctuations in background signal in the wormtrail (Figure 11B) is likely due to the disordered material formed with the wormlike defect. It could arise from melted phases from the constitutive elements of CIGS, in agreement with the results of Guthrey et al.21

Far away from the wormlike defect (Region III, Figure 11C), the characteristic peaks of CIGS are found, with a larger Cu (In, Ga) Se2‐related peak and a smaller Cu (In, Ga)S2‐related peak. We must note at this point that, since the vibration modes observed in Raman are mostly homo‐atomic, in the case of Cu (In, Ga)(Se, S)2 we are looking at the Se–Se and S–S bonds. In regular CIGS, as here in Region III, the main vibration modes visible in Raman are the A1 mode of Cu (In, Ga) Se2 (Se–Se vibrations) and the A1 mode of Cu (In, Ga)S2 (S–S vibrations). The shoulder at 260 cm−1 could be due to Cu‐Se phases,30 although they do not seem to impact the performance of the reference cell NW significantly.

More interestingly, the wormtrail edge (Region II, Figure 11D) contains the same characteristic peaks as Region III, with however a very different ratio between the two main CIGS peak areas: in region II, the S–S peak dominates, whereas the Se–Se peak dominates in Region III. CdS can also give rise to an S–S type peak at the same position as the S–S bond in CIGS. However, CdS is normally associated with another peak at 600 cm−1, not present here. This suggests that the edge of the wormlike defects is enriched in sulphur, but maybe not in the form of CdS.

Etching

To cross‐check the Raman results and quantify by EDS the chemical composition in those regions, the ZnO window layer, which otherwise absorbs most of the electrons from the SEM, had to be removed. This also allows us to study the structure of the wormlike defects further.

Figure 11 shows the photoluminescence (PL) imaging of sample W at various stages of the etching procedure. Looking at the two cells initially affected by wormlike defects, we can see a clear trend. Initially (Figure 12A), the TCO is sufficiently conductive to allow the shunting wormlike defects to drain the photogenerated carriers from the entire cell's surface. As the TCO is etched away, however, its sheet resistance increases. Consequently, the distance from which the wormlike defects are able to draw the photogenerated carriers (or carrier extraction length, LCE) decreases and the parts of the cell sufficiently far away from the wormtrails start luminescing again (Figure 12B,C). Finally, once most of the TCO has been etched away, a uniform photoluminescence intensity is restored in the entire cell (Figure 12D,E). At this stage, the LCE becomes very small in comparison to the cell dimensions, since all that is left is the lateral conductivity of CIGS, which is much smaller than that of the TCO. The effect of a decreasing lateral conductivity (upon TCO etching) is illustrated for a point‐like shunt in Figure 11F–H.

The SEM micrographs of the wormlike features after 180 s etching and 270 s etching are shown in Figure 13. The first image was acquired at 180 s because, based on our previous experience with similar TCO thicknesses, it was the duration for which etching starts having a visible effect. The TCO removal from the wormtrail area proceeds as follows:

• First, the TCO cracks at the wormtrail edge, due to the strong tensile mechanical stress in this region (Figure 13A)
• Then, the TCO is entirely removed, uncovering the bare wormlike defect, while small amounts of TCO persist on the flatter regions around the wormtrail (Figure 13B)

Looking at the bare wormtrail of Figure 13B, one striking feature is the presence of porosity in the underlying CIGS. This explains how the material could accommodate such a drastic expansion during wormlike defect formation (i.e., wormtrails rising by over one micron).

SEM/Raman comparison

Both SEM/EDS and Raman linescans were carried out in the same location, from the wormtrail edge and outwards. Note that, although the Raman measurement was carried out before etching (along with the data shown in Section 3.4.3), the Raman laser beam still interacted only with the underlying material, as shown for example in a previous study.31 This is because the laser wavelength (530 nm) is well above the absorption edge of the ZnO‐based TCO (<350 nm).

One clear trend arises when plotting the atomic ratio [S]/([S] + [Se]) against the distance from the wormtrail edge for the SEM/EDS data (Figure 14C). The sulphur content is increasing sharply in a narrow region of the wormtrail edge, before decreasing again as we get away from the worm.

To compare this result directly to Raman, the same linescan (although less extended) was carried out in Raman, which is a much more surface sensitive technique than SEM/EDS. As mentioned in Section 3.4.3, one can study the evolution of the ratios between the areas of the A1 peaks of Cu (In, Ga) Se2 and Cu (In, Ga)S2 (Figure 14B). Note however that unlike EDS, the Raman analysis is only semi‐quantitative, in the sense that a ratio A1(S)/(A1(S) + A1(Se)) = 2, for example, does not mean that there is twice as much sulphur as selenium. Trends, however, are still meaningful: a ratio increase (decrease) does indicate an increasing (decreasing) [S]/([S] + [Se]) atomic ratio.

The EDS and Raman linescans of Figure 14 show the same trend, thus confirming that there is an increased sulphur ratio in the material located at the wormtrail edge. A similar observation was made by Bakker et al on lab‐scale, co‐evaporated samples, where the authors proposed that some of the S from the CdS replaces Se in the CIGS.13 The module studied in the present paper does indeed include a Cd‐containing buffer layer. However, in the present case, sulphur is also present in the Cu (In, Ga)(S, Se)2 absorber, whereas the co‐evaporated CIGS of Bakker et al.13 was sulphur‐free. This means that we here have two possible sources for the sulphur: the CdS buffer layer and/or the absorber layer. Although this is only circumstantial evidence at this point, the fact that sulphur systematically accumulates at the wormtrail edge could be one of the reasons behind the distinctive morphology of the wormtrails in the present work (see Section 3.4.1), compared to Bakker et al.13,26

We also note that the scale of the variation in Figure 14 is different for Raman and EDS. Indeed, the S/(S + Se) variation (i.e., increasing, reaching a peak and decreasing) occurs over 15 μm in Raman, against 5 μm in EDS. We assign this difference to EDS (especially at 20 kV acceleration voltage) having a wider beam diameter and probing higher sample depths (both values typically ≈2 μm) than Raman (1 μm and 200 nm, resp.).

Point like feature and possible connections with wormlike defects

The strongly protruding nodules observed on NW were also present on W before etching, not only on top of the P1 scribe (see Figure 15B), but also in other parts of the cell. Confocal microscopy showed that the nodules found in NW are protruding by 3–5 μm (Figure 5), against 1–2 μm for the wormtrail ridge in W (Figure 8). They are therefore distinct features, rather than one of the 'islands' making up the wormlike defects, as shown in Bakker et al.26 The I‐V parameters of NW also indicate they do not impact the performance significantly. These two properties (very large size and no impact on the PL or I‐V) suggest that these features might be manufacturing‐related. If these defects are indeed manufacturing defects, they should be avoided during manufacturing, to mitigate the risk of wormlike defect formation. Although the nodules are distinct from the wormlike defects, that does not necessarily mean that they are independent features. They may still play a role in the propagation and even the formation of the wormlike defects.

Bakker et al proposed that the wormlike defects originate in local hotspots and propagate by hopping to the next local hotspot.26 In the present work, we can tie up this model with three observations:

• The wormlike defects always stay close to the P1 scribes, for reasons detailed in Figure 10.
• The nodules, likely manufacturing defects, are found in random locations of the cell.
• The path of the wormlike defects seems heavily influenced by the presence of such nodules, as shown in Figure 15B.

The fact that such nodules light up in ILIT (Figure 4) indicates that they are hotspots, that is, that more current circulates in those features and heat them up. Based on the model of Bakker et al mentioned earlier on hotspot‐based propagation, one can understand why such nodules would affect the propagation pathway of the wormlike defects, as shown in Figure 15B. They could also be the hotspots from which the wormlike defects originate before propagating, although no clear evidence of that can be found in the data presented here. If the wormlike defects do indeed originate from one of these nodules, this is likely in the cell interior, where they are also present (Figure 4), and not on the P1, where shunting is much less likely to occur, since there is no Mo below the P1.

Relevance of the coring/unpackaging for module failure analysis

The question we want to address in this section is more general: what is the value of the coring‐unpackaging method for failure analysis? We have demonstrated in Section 3.3 that the protocol used did not yield any detectable damage to the devices extracted. As already mentioned, all the cores extracted and unpackaged on this module were similarly undamaged.

The main limitation of this method is that it is destructive. Once a coring experiment has been carried out, the module is unusable. This is why, in the present study, we could not extract samples before the static shading experiment presented here, which would have unambiguously answered the question of the origin of the nodules and whether they were partial‐shading related.

However, the present results do demonstrate how valuable this method can be in investigating the origin of a given failure mode in commercial modules, which can differ significantly from the lab‐scale devices. In the present work, the large nodules in the P1 scribe are an example of such a feature.

The method can in principle be applied to study the root cause of any failure in the PV stack, whether manufacturing, installation or operation‐related, using any characterisation device available in the laboratory.

Let us also point out that such methods can be transferred to any thin film or even crystalline technology.

CONCLUSION

We have demonstrated a method to extract and prepare pristine samples from commercial modules, which can then be used for opto‐electronic as well as any other type of material characterisation.

Using this method, we first compared the I‐V data of the original module (before and after partial shading) to that of a sample cored out of a reference (worm‐free) region. The results indicated that the coring/unpackaging procedures did not yield any detectable damage to the active layers.

We then went on to study a sample with wormlike defects induced by partial shading and discovered some original features, including the presence of very large nodules, also present in the reference sample, which steer the propagation of the wormlike defects and could possibly play a role in their formation.

Those results provide valuable insights into partial‐shading induced wormlike defects in modules. They also demonstrate the potential of coring and unpackaging for microscopic study of failure modes occurring in commercial modules.

ACKNOWLEDGEMENTS

The author wish to thank Johan Bosman for the laser removal of the MoSe2 and Frank de Graaf for laying the foundations of the coring procedure.

This work is supported by 'Netherlands Enterprise Agency' (RvO) and the Dutch Topteam Energy via the projects: 'Building Integrated PhotoVoltaic Panels on Demand ‐ in The Netherlands' with grant number TEID215005 and 'Performance and Electroluminescence Analysis on Reliability and Lifetime of Thin‐Film Photovoltaics' with grant number TEUE116203. The Early Research Program 'Sustainability & Reliability for solar and other (opto‐)electronic thin‐film devices' from TNO is also acknowledged for funding.

DATA AVAILABILITY STATEMENT

Research data are not shared.

GRAPH: Figure S1 1. EDS cross section of the PV stack. Note that this EDS profile is not quantitative because of the rough (unpolished) cross sectionFigure SI 2. optical microscopy image of the area mapped by confocal microscopy in Figure 8 of the main document

1 Funding information TNO; Dutch Topteam Energy; Netherlands Enterprise Agency Fraunhofer Institute for Solar Energy Systems. Photovoltaics report. https://www.ise.fraunhofer.de/ 2 PVTECH. Singulus gets €50 million follow‐on CIGS thin film tool order from CNBM. 2020. Accessed March 17, 2020. https://www.pv-tech.org/news/singulus-gets-50-million-follow-on-cigs-thin-film-tool-order-from-cnbm 3 pv‐magazine. The weekend read: CIGS is back, back again. 2018. Accessed March 17, 2020. https://www.pv-magazine.com/2018/07/21/the-weekend-read-cigs-is-back-back-again/ 4 pv‐magazine. Hanergy wraps 460 kW of CIGS around Chinese skyscraper. 2019. Accessed March 17, 2020. https://www.pv-magazine.com/2019/07/10/hanergy-wraps-460-kw-of-cigs-around-chinese-skyscraper/ 5 PVTECH. Hanergy to supply 100MW of thin‐film solar roof tile systems to Japan after JET certification. 2018. https://www.pv-tech.org/news/hanergy-to-supply-100mw-of-thin-film-solar-roof-tile-systems-to-japan-after 6 Shukla AK, Sudhakar K, Baredar P. Recent advancement in BIPV product technologies: a review. Energ Buildings. 2017 ; 140 : 188 ‐ 195. doi: 10.1016/j.enbuild.2017.02.015 7 Del Cueto J, Rummel S, Kroposki B, Osterwald C, Anderberg A. Stability of CIS/CIGS modules at the outdoor test facility over two decades. In: 2008 33rd IEEE Photovoltaic Specialists Conference. IEEE ; 2008 : 1 ‐ 6. 8 Theelen M, Daume F. Stability of Cu (In, Ga) Se2 solar cells: a literature review. Solar Energy. 2016 ; 133 : 586 ‐ 627. doi: 10.1016/j.solener.2016.04.010 9 Jordan DC, Kurtz SR, VanSant K, Newmiller J. Compendium of photovoltaic degradation rates. Progr Photovolt: Res Appl. 2016 ; 24 (7): 978 ‐ 989. doi: 10.1002/pip.2744 Moutinho H, Johnston S, To B, et al. Development of coring procedures applied to Si, CdTe, and CIGS solar panels. Solar Energy. 2018 ; 161 : 235 ‐ 241. doi: 10.1016/j.solener.2017.11.071 Johnston S, Moutinho H, Jiang C‐S, et al. From Modules to Atoms: Techniques and Characterization for Identifying and Understanding Device‐Level Photovoltaic Degradation Mechanisms. Nat Renew Energy Lab (NREL). 2019. https://www.nrel.gov/docs/fy20osti/72541.pdf Moutinho H, To B, Sulas‐Kern D, Jiang C‐S, Al‐Jassim M, Johnston S. Advances in Coring Procedures of Silicon Photovoltaic Modules. In: 2020 47th IEEE Photovoltaic Specialists Conference (PVSC). IEEE ; 2020 : 1449 ‐ 1453. Bakker K, Åhman HN, Aantjes K, Barreau N, Weeber A, Theelen M. Material property changes in defects caused by reverse bias exposure of CIGS solar cells. IEEE J Photovolt. 2019 ; 9 (6): 1868 ‐ 1872. doi: 10.1109/JPHOTOV.2019.2940486 Lee JE, Bae S, Oh W, et al. Investigation of damage caused by partial shading of CuInxGa (1‐x) Se2 photovoltaic modules with bypass diodes. Progr Photovolt: Res Appl. 2016 ; 24 (8): 1035 ‐ 1043. doi: 10.1002/pip.2738 Palmiotti E, Johnston S, Gerber A, et al. Identification and analysis of partial shading breakdown sites in CuInxGa (1‐x) Se2 modules. Solar Energy. 2018 ; 161 : 1 ‐ 5. doi: 10.1016/j.solener.2017.12.019 Herrmann MCAW, in 19th EUPVSEC, Munich, 2004: WIP Renewable Energies. Wendlandt SBS, Weber T, Berghold J, Krauter S, Grunow P. Shadowing investigations on thin film modules. In: 32nd EUPVSEC ; 2016 : 2270 ‐ 2276. Bakker K, Weeber A, Theelen M. Reliability implications of partial shading on CIGS photovoltaic devices: a literature review. J Mater Res. 2019 ; 34 (24): 3977 ‐ 3987. doi: 10.1557/jmr.2019.373 Mortazavi S, Bakker K, Carolus J, et al. Effect of reverse bias voltages on small scale gridded CIGS solar cells. In: 2017 IEEE 44th Photovoltaic Specialist Conference (PVSC). IEEE ; 2017 : 2875 ‐ 2880. Silverman TJ, Deceglie MG, Deline C, Kurtz S. Partial shade stress test for thin‐film photovoltaic modules. In: Reliability of Photovoltaic Cells, Modules, Components, and Systems VIII. Vol. 9563. International Society for Optics and Photonics ; 2015 : 95630F. Guthrey H, Nardone M, Johnston S, et al. Characterization and modeling of reverse‐bias breakdown in Cu (In, Ga) Se2 photovoltaic devices. Prog Photovolt: Res Appl. 2019 ; 27 (9): 812 ‐ 823. doi: 10.1002/pip.3168 Johnston S, Sulas D, Palmiotti E, et al. Thin‐film module reverse‐bias breakdown sites identified by thermal imaging. In: 2018 IEEE 7th World Conference on Photovoltaic Energy Conversion (WCPEC) (A Joint Conference of 45th IEEE PVSC, 28th PVSEC & 34th EU PVSEC). IEEE ; 2018 : 1897 ‐ 1901. Johnston S, Palmiotti E, Gerber A, et al. Identifying Reverse‐Bias Breakdown Sites in CuIn x Ga (1‐x) Se 2. In: 2017 IEEE 44th Photovoltaic Specialist Conference (PVSC). IEEE ; 2017 : 1400 ‐ 1404. Tzikas C, Gómez G, van den Donker M, Bakker K, Smets AH, Folkerts W. Do thin film PV modules offer an advantage under partial shading conditions. 33rd European Photovoltaic Solar Energy Conference and Exhibition. 2017 : 1593 ‐ 1596. Silverman TJ, Repins I. Partial shade endurance testing for monolithic photovoltaic modules. In: 2018 IEEE 7th World Conference on Photovoltaic Energy Conversion (WCPEC) (A Joint Conference of 45th IEEE PVSC, 28th PVSEC & 34th EU PVSEC). IEEE ; 2018 : 3932 ‐ 3937. Bakker K, Åhman HN, Burgers T, Barreau N, Weeber A, Theelen M. Propagation mechanism of reverse bias induced defects in Cu (In, Ga) Se2 solar cells. Solar Energy Mater Solar Cells. 2020 ; 205 : 110249. doi: 10.1016/j.solmat.2019.110249 Westin PO, Zimmermann U, Stolt L, Edoff M. Reverse bias damage in CIGS modules, presented at the 24th EUPVSEC, 2009. Takei R, Tanino H, Chichibu S, Nakanishi H. Depth profiles of spatially‐resolved Raman spectra of a CuInSe 2 ‐based thin‐film solar cell. J Appl Phys. 1996 ; 79 (5): 2793 ‐ 2795. Sekine T, Izumi M, Nakashizu T, Uchinokura K, Matsuura E. Raman scattering and infrared reflectance in 2H‐MoSe 2. J Phys Soc Japan. 1980 ; 49 (3): 1069 ‐ 1077. Witte W, Kniese R, Powalla M. Raman investigations of Cu (In, Ga) Se 2 thin films with various copper contents. Thin Solid Films. 2008 ; 517 (2): 867 ‐ 869. Oliva F, Kretzschmar S, Colombara D, et al. Optical methodology for process monitoring of chalcopyrite photovoltaic technologies: application to low cost Cu (In, Ga)(S, Se) 2 electrodeposition based processes. Solar Energy Materials and Solar Cells. 2016 ; 158 : 168 ‐ 183.

By Rémi Aninat; Klaas Bakker; Laetitia Jouard; Manuel G. S. Ott Cruz; Pelin Yilmaz and Mirjam Theelen

Reported by Author; Author; Author; Author; Author; Author