1 INTRODUCTION

Photovoltaic (PV) energy systems are one of the least expensive and fastest growing sources of electrical generation in many parts of the world.¹ This major technology shift is largely because of the 85% decrease in the cost of solar modules since 2010.² Economies of scale; use of new, higher efficiency cell designs; automation of production lines; larger modules; and changes to the bill-of-materials (BOM) (thinner glass and frames, new encapsulants, backsheets, etc.) have contributed to cost reductions. The levelized cost of electricity (LCOE) is sensitive to the power degradation rate (Rd); as it increases, LCOE rises, and system lifetime falls. As a note, there is no consensus on PV lifetime definition; this study arbitrarily defines lifetime as the period of time until PV performance drops to 80% of initial. However, this does not mean that a PV power plant should retire once it reaches this state.³

Since many of the degradation processes originate from how modules and their materials interact with the environment, there is concern that cost cutting design and materials changes may result in higher degradation rates, which could cancel out many of the positive results of lower module costs. Studies on modules manufactured between ~1979 and ~2014 conducted by Jordan et al^{4, 5} found module Rd values between −0.8%/year and −0.9%/year with the median between −0.5%/year and −0.6%/year. These studies were based on several thousand modules; however, many of the Rd values were calculated as changes from nameplate power rating based on a single flash test measurement taken after field exposure rather than measuring the modules at the time of installation and retesting over the years. This approach based on only two points in time provides a linear estimate of degradation but may be biased if the nameplate rating does not match the initial performance. Is the average degradation rate for newer modules significantly different because of recent changes to module design and materials?

The PV market is continuously changing, recently transitioning from conventional aluminum back surface field (Al-BSF) designs, which dominated (by up to 90%) the global solar cell production until 2018, to other high-efficiency cell concepts including passivated emitter and rear cell (PERC), passivated emitter with rear locally diffused cell (PERL), passivated emitter rear totally diffused cell (PERT), silicon heterojunction (SHJ), and tunnel oxide passivated contact (TOPCon) designs.⁶ According to the International Technology Roadmap for Photovoltaics,⁷ the market share in 2020 was around 15% for Al-BSF and around 80% for high-efficiency cell concepts, mainly dominated by PERC. There is a lack of long-term field data for these new cell and module technologies.^{8, 9} High-efficiency modules can exhibit open-circuit voltage (Voc) degradation as opposed to the short-circuit current (Isc) or fill factor (FF) degradation common in Al-BSF modules.⁸ Nonlinear degradation modes such as light induced degradation (LID)¹⁰ or light and elevated temperature induced degradation (LeTID)¹¹ are more common in high efficiency modules and can impact project cash flows^{5, 12, 13} and increase uncertainty in lifetime energy yields.

The market is also seeing increases to module size, use of new materials (e.g., anti-reflection and anti-soiling coatings, thinner glass, new encapsulants, and backsheets), or other concepts such as split cells, dense interconnection designs (shingling), or increased number and topology of busbars or wires. Such concepts may increase module efficiency and reduce balance of system (BOS) costs, but it remains unknown how these changes affect degradation rates and long-term reliability.

To investigate this question, our team purchased and fielded 834 modules from the open market representing 13 different module types from 7 manufacturers and deployed them in the field at 3 locations (New Mexico [NM], Colorado [CO], and Florida [FL]) representing a range of climates. We measured performance under Standard Test Conditions (STC) at the start of the study, examined initial power stabilization, and periodically retested the modules over the following 5 years to monitor degradation rates over time.

The Sandia National Laboratories (Sandia), the National Renewable Energy Laboratory (NREL), and the University of Central Florida's Florida Solar Energy Center (FSEC) partnered for this project in 2016.¹⁴ The objective was to measure PV degradation rates over time for technologies with significant US market share and release onymous (not anonymous) results publicly to inform relevant stakeholders and inform best practices. More than 2000 separate flash tests were analyzed in this study.

2 DESCRIPTION OF FLASH TEST CAMPAIGNS

2.2 Module stabilization and testing protocols

Upon arrival at Sandia, NREL, and FSEC, at least 25% of the modules were sampled for flash testing prior to any light exposure. Before flash testing at STC, Jinko260, Jinko265, Trina255, and Trina260 modules were light soaked for at least 4.3 kWh/m² in 2016, whereas the remaining modules (post-2016) were light soaked for at least 20 kWh/m², consistent with an update to IEC61215-1:2016.¹⁶ The modules were then installed in the field, connected to inverters and the grid. A sub-sample of modules (25%) was retested annually, and light-soaked control modules were stored in climate-controlled dark rooms and tested with each batch of fielded modules to help distinguish between field-related degradation and non-field-related material degradation. Additional process control modules (different types) were tested before each flash test session at all laboratories to monitor the long-term stability of the flash simulators. It should be noted that the FSEC modules were purchased by Sandia but directly shipped to Florida by the vendor. As such, Sandia and FSEC modules were sourced from the same batch and might differ in BOM with the respective modules at NREL. A flowchart describing the module testing protocol is shown in Figure 1.

2.3 Indoor flash testing and stability

Sandia, NREL, and FSEC harmonized the flash testing procedures for this project. All laboratories included control modules for each module technology under test, which were stored indoors, in the dark, in a climate-controlled room. All modules were light soaked at the beginning of the project before baseline flash testing, whereas a subsample (typically 50%) was flashed before light soaking to assess initial power stabilization. For all flash testing, the modules were placed at the same location on the simulator test planes to reduce uncertainties because of light non-uniformity. For each test campaign, all modules retrieved from the field were brought indoors the night before flash testing and allowed to thermally equilibrate. However, there are some differences between the methods used at each facility because of different site constraints; these are listed in Table 2.

TABLE 2. Summary of characterization differences among the laboratories

Laboratory	Temperature control	Method*	Repeatability	Mean 2 standard deviations**	Flash tester
Sandia	±0.5 °C	Isc to Voc	±0.5% on normalized Pmp	±1.2% on normalized Pmp ±1.1% on normalized Isc ±0.2% on normalized Voc	Spire 4600 SP
NREL	±0.5 °C	Voc to Isc	±0.1% on Pmp	1.3792 W 0.0122 A 0.0783 V	Spire 5600
FSEC	±1 °C	Isc to Voc	±1% on normalized Pmp	±1.55% on normalized Pmp ±1.5% on normalized Isc ±0.3% on normalized Voc	Sinton Instruments FMT-350

• * Although the absolute measured values differ from one sweep direction to another,^{17, 18} this study reports on relative changes over time. As such, the IV curve sweep direction should have a negligible effect as long as each laboratory is consistent over time.
• ** Standard deviation values for NREL were taken from Ndione et al.¹⁹

With respect to temperature corrections, Isc, Voc, and power at maximum power (Pmp) are corrected for temperature using the temperature coefficients from the specification sheets. Quantifying the stability and repeatability of solar simulators at each site is important because degradation studies require monitoring small changes in module power ratings over long periods of time (up to 10 years or more). All laboratories have identified a set of performance monitoring modules that were carefully stored indoors and flash tested regularly to monitor the stability of the simulator. Figure 2 shows the repeatability in normalized Pmp for 10 of these modules at Sandia and 11 modules at FSEC since mid-2017 and 15 since mid-2018; NREL's repeatability has been published elsewhere (±0.1% around 2 sigma¹⁹ based on the module self-reference method²⁰), and it will not be repeated here. The dashed red, green, and blue lines represent ±3%, ±1% and ±0.5% around the median, respectively. Most of the data (9 out of 10 modules at Sandia and 11 out of 15 modules at FSEC) fall within this uncertainty bound, which verifies the current repeatability of the measurements made at Sandia, FSEC, and NREL (for additional information on NREL's repeatability/uncertainty, see Ndione et al.¹⁹). Changes to FSEC's system introduced short-term spread on the stability plot because of the removal of the lamp filter, improvements made in non-uniformity, and a change in calibration procedure. Overall, uncertainties of relative power changes over time (±0.5%) are much lower than the absolute power accuracy.¹⁴

3 INITIAL PERFORMANCE AND POWER STABILIZATION

Selected modules were flashed before any outdoor light exposure, and deviation from the nameplate power rating is shown in Figure 3. Median percentage differences ranged from around −3.6% to +4%. However, the uncertainty in measured power is approximately ±2.3%, based on interlaboratory comparisons. Because all module manufacturers claim positive bin tolerances (up to +10% for Panasonic, +5 W for Canadian Solar, and +3% for the remaining manufacturers), most modules appear to be ‘under-rated’ (i.e., measured power > nameplate), whereas others are near or even below nameplate. This is important because degradation in the context of a warranty is relative to the nameplate power. Modules that are ‘under-rated’ can degrade further before a warranty claim can be made.

LID can cause initial degradation in poly-crystalline (or multi-crystalline) and mono-crystalline silicon modules within the first 10 kWh/m² of light exposure through formation of boron-oxygen defects.²¹ Examples of the temperature-corrected power stabilization process are shown in Figure 4 for four representative modules. The largest initial power loss of −3.3% was for the mono-silicon module-type CSmono275 compared with a similar poly-Al-BSF (CSpoly270) from the same vendor that saw only −1.5% power loss. This is to be expected since LID is known to more strongly affect mono-Al-BSF. Two PERC module types had comparable performance loss of −1.2% (Qmono290 and Qpoly280). The observed stabilization of the PERC modules is similar to a recent study by Chen et al.²²

Initial power stabilization deviations for the other module types are given in Table 3. It is worth noting that the initial power improvement of the Panasonic325 at NREL is in agreement with a study by Kobayashi et al.²³ where silicon heterojunction solar cells demonstrated a light-induced performance gain prior to long-term degradation. Kobayashi et al.²³ attributed this improvement to an increase in surface passivation.

4 OBSERVED DEGRADATION PATTERNS

Studying performance degradation patterns on a large group of diverse modules in a well-controlled lab environment allows us to observe both general trends and variations in behavior indicative of modules being produced on different production lines including typical variations in the bill of materials due to complex supply chains. A systematic analysis of all the module results is presented in supporting information sections A and B. The most interesting highlights are summarized below.

LeTID-like behavior was observed on certain PERC and Al-BSF modules in this study. Jinko260 (Figure 5) and Jinko265 (see supporting information section A1) modules exhibited a steep drop in performance followed by unstable behavior indicative of LeTID. These modules measured at NREL every 6 months showed that performance was lower in the summer and increased in the winter, consistent with this degradation process. LeTID in Jinko260 modules was split between current and voltage loss, which can be seen in the Isc and Voc curves in Figure 5B and C. Although typically associated with PERC cells, LeTID has been identified in some Al-BSF multi-crystalline cells,²⁴ and these modules were shown to be LeTID susceptible in IEC 61215 module quality tests 23.²⁵ In this case, LeTID was also verified through another study where a sample of the same Jinko260 modules was used to compare LeTID progression against an LeTID prediction model.²⁶ Furthermore, one outlier Trina260 module at NREL also appears to exhibit LeTID-like performance.

Some modules showed differing patterns over time at different sites. Trina260 modules at Sandia showed a modest drop in Pmp, Isc, and Voc in the first year, but these values remained relatively unchanged thereafter (see Figure 6). In contrast, these same modules at NREL showed continual drops in Pmp, Voc, and FF from 2016 to 2020 followed by an increase in 2021. Furthermore, the outlying Trina260 module at NREL (see purple color in Figure 6) points out that even when the same module is purchased from the same vendor, there is a possibility of ending up with different BOM and thus, a potential different behavior.

LG320 modules displayed moderate overall power degradation at Sandia (−0.69%), whereas at NREL, these modules on average remained relatively stable in performance over time (−0.06%); however there was significant variation in individual modules, with some increasing and others decreasing (see supporting information section A3). One explanation for why some results varied between sites could be because of differences in duration of the initial light soaking at each site. LG320 modules were exposed for 20 kWh/m² at NREL and 356.8 kWh/m² at Sandia. It is possible that the performance of the Sandia modules improved during this extended light soak period, whereas the NREL modules did not have sufficient time to do so. Also, the control modules at the two laboratories also showed contradicting performance with −0.32% and +0.43% power change at Sandia and NREL, respectively. This might suggest that the BOMs may also be different.

The power loss of the Panasonic325 modules (−1.87% at Sandia and −0.68% at NREL; see supporting information section A4) was large because of decreases in voltage, and, in fact, Isc increased over time for these modules at both sites, overall changes in Isc (i.e., ΔIsc) of +0.24% and +0.57% and ΔVoc of −1.48% and −1.03% for Sandia and NREL, respectively.

5 DISCUSSION

Module Rd is calculated using two methods that differ in how the initial power rating of the module is defined. The first uses nameplate rating as the starting point and the second uses the first post-stabilization flash test. Figures 7 and 8, show the evolution of Rd for all module types across the different laboratories based on these two methods. In Figure 7, because the first flash test is at ‘time zero’ (i.e., there is no time information between the first flash and nameplate), it is only possible to calculate a ΔPmp in percent and not an Rd in percent/year. For this reason, first measurements are excluded from Figure 7. A summary of the reported degradation rates and differences among the two methods are summarized in Table SB13.

Nameplate-based Rd values ranged from −2.79%/year to 0.72%/year, whereas first flash-based Rd values ranged from −1.94%/year to 0.33%/year. Trina260 and CSmono275 exhibited positive nameplate-based Rd values at Sandia (+0.29%/year, +0.72%/year), whereas the Rd of Panasonic325 and LG were +0.30%/year and +0.60%/year at NREL. However, first flash-based Rd values were positive only for CSmono275 at Sandia (+0.33%/year); this module exhibited the greatest initial power deviations. Therefore, most positive nameplate-based Rd values could be because of initial measured power > nameplate (i.e., under-rating) and not necessarily because of module performance gains. The overall mean and median Rd values were found to be −0.61%/year and −0.58%/year, respectively.

6 SUMMARY AND CONCLUSIONS

Our team purchased over 834 modules representing 13 different technologies from 7 manufacturers and deployed them in the field at 3 different locations representing a range of climates. We measured performance indoors under standard test conditions at the start of the study, examined initial power stabilization, and periodically recharacterized the modules over the following 4 years to monitor degradation rates over time.

Nameplate power rating differences ranged from around −3.6% to 4% with initial power stabilization varying from −3.3% to +0.6%. Flash-over-flash measurements showed variable performance with seasonality amplifying the variations. Furthermore, the module power loss for higher efficiency technologies (e.g., SHJ and PERC) was driven by Voc degradation indicating that the cells are still changing, whereas LeTID was observed in PERC and Al-BSF modules confirming previous studies. It was also observed that the BOM makes a big difference in field performance even when the modules were purchased from the same vendor.

The Rd values were evaluated based on nameplate values and first post-stabilization flash power ratings and showed differences of up to −2.39%/year depending on the reference used. We found that post-stabilization flash measurements help to ensure that a module performs according to expectations and avoids energy yield and financial bias. Furthermore, we observed that the Rd ‘stabilizes’ after ~3–4 years.

The analysis found mean and median Rd of −0.62%/year and −0.58%/year, respectively, with some modules exhibiting improvements over the test period. Even though the median Rd values at Sandia (New Mexico; hot climate) and NREL (Colorado; cooler climate) were very close (−0.56%/year and −0.61%/year), this was not the case for the systems at the hot and humid subtropical climate of Florida (FSEC) exhibiting twice the median Rd value of −1.12%/year. However, more samples are required to verify such weather dependent findings.

Overall, the degradation rates found in this work are within the values observed in the past (from ~1979 to ~2014 module technologies) for more expensive (by up to 85%) conventional PV technologies. Therefore, although the costs declined sharply in the last decade, module degradation rates do not seem to be affected, at least for the sample investigated in this work. This is a very encouraging result, but more opportunities exist to reduce Rd to levels that enable longer PV module lifetimes. Finally, with respect to module warranties, the analysis so far showed that 26.1% of the systems are exceeding the warranty limits, whereas 56.5% of the systems demonstrated the potential of achieving lifetimes beyond 30 years assuming that Rd trends are stabilized.

CONFLICT OF INTEREST

All PV modules used in this study were purchased by the Sandia National Laboratories and National Renewable Energy Laboratory from independent vendors with the financial support of the US Department of Energy's Office of Energy Efficiency and Renewable Energy (EERE) under the Solar Energy Technologies Office. As such, the results are reported in an onymous (not anonymous) manner. The authors declare no competing interests.

AUTHOR CONTRIBUTIONS

Conceptualization: MT and JSS; Data curation: MT; Formal analysis: MT; Funding acquisition: BHK and CD; Investigation: CR, WS, AA, and DJC; Methodology: MT and JSS; Project administration: BHK, CD, JW, and HS; Software: MT and JSS; Supervision: JSS, BHK and CD; Validation: MT, JSS and CD; Visualization: MT and JSS; Writing – original draft: MT and JSS; Writing – review and editing: MT, JSS, CD, DJ, CR, WS, DJC, and HS. All authors read and approved the final manuscript.