<!-- ARTICLE_BODY_START -->
A year outside can be an unusually useful result
A solar-cell paper published on June 9, 2026 reported a difficult outdoor year. The experimental device combined two perovskite junctions with silicon. Its measured performance reached 80 percent of its original level after about four months and 50 percent after about six. All three subcells remained operational at the end of the year, but that did not mean the original output had been preserved. The researchers also observed an initial packaging-related loss that partly recovered outdoors. Manshanden and colleagues' outdoor study
Those findings concern an early research sample, not every perovskite product or a representative fleet. Their value lies in showing what an outdoor experiment can reveal that an initial efficiency measurement cannot. The device changed over time, and different parts of its electrical behavior changed with it. A single before-and-after headline would leave much of that story unexplained.
Perovskite photovoltaics attract attention partly because their material properties offer ways to divide the sunlight spectrum among different absorbers. In a tandem, a perovskite layer can be paired with another photovoltaic material, often silicon. The aim is to convert more of the incoming light into useful electricity than either component would deliver in the same arrangement alone. Achieving that aim once under a defined measurement is a scientific accomplishment. Preserving it outdoors is another problem.
For the owner of a roof or solar installation, the useful product is electricity delivered over time. That makes efficiency, degradation, operating conditions and maintenance parts of the same question. A device can begin with a substantial advantage and lose it quickly. Another can start less impressively but retain more output. Neither observation can be reduced to a contest between the highest initial percentages.
This article examines how to read the evidence connecting those two achievements. Sources were checked on October 31, 2026; research dates and repository dates are distinguished where they differ. Numerical scenarios are original teaching examples, not forecasts for a named manufacturer. The discussion is not a purchasing recommendation, electrical installation guide or instruction for handling damaged modules containing lead.
Why stacking absorbers helps—and complicates measurement
A photovoltaic absorber responds to light according to its electronic properties. A tandem divides the conversion task between layers with different responses. The upper layer must absorb useful light while allowing an appropriate portion to reach the layer below. The electrical design must then collect the resulting power. The attraction comes from making better use of the spectrum, not from creating additional sunlight.
An analogy with two workers can help, provided its limits are clear. If one worker handles a class of tasks poorly and another handles those tasks well, dividing the workload can improve the combined result. But a badly designed handoff can erase the benefit. In a tandem device, the handoff is optical and electrical: light travels through layers and charges must reach suitable contacts. The layers, interfaces and interconnections all belong to the device being evaluated.
A two-terminal tandem electrically connects its subcells in series, so their operation is coupled through the current. A four-terminal arrangement allows separate electrical access to the component cells. These architectures create different opportunities and constraints. A result obtained with independently measured subcells should not be silently presented as the performance of a series-connected device under all conditions. The architecture is part of the result, rather than a minor label in the methods.
Consider an idealized circuit illustration. Two series-connected elements operate at the same current; their voltages contribute to the combined voltage. If one element cannot support the current available from the other under a particular illumination, that mismatch constrains the combined operating point. With independent electrical outputs, the useful powers can instead be collected separately, but doing so introduces its own system choices. This is basic circuit reasoning, not a detailed model of any cited device.
The practical question is therefore broader than whether each absorber performs well individually. It is whether the combination performs well under the light, temperature and electrical conditions it will encounter. A tandem's efficiency advantage and its durability cannot be evaluated entirely independently because deterioration in one part can change how the other part contributes.
A laboratory percentage needs its measurement conditions
Efficiency is the fraction of incident optical power converted into electrical power under the specified measurement. The word specified matters. The light spectrum, intensity, device temperature, illuminated area and electrical measurement procedure affect what the number means. A comparison between two percentages is informative only when the relevant conditions and area definitions are compatible.
Suppose two fictional samples each produce three watts under a test. If one receives ten watts of light across the defined area, its efficiency is 30 percent. If the other receives twelve watts, its efficiency is 25 percent. The equal electrical output does not imply equal efficiency. Conversely, equal efficiency does not imply equal total output if the tested areas differ. These simple distinctions become important when a small research cell is compared with a packaged module.
Area can be particularly confusing. An active photovoltaic region, an aperture through which light reaches the device, and the total area occupied by a packaged product need not be the same. None is an inherently dishonest choice if it is appropriate and clearly reported. Trouble begins when the denominator changes between a scientific claim and a commercial comparison without explanation.
The electrical procedure matters too. A quick sweep through operating points can characterize a device, while sustained operation at a power-producing point answers a different question. If the device changes during measurement, the path and timing of the measurement can influence the result. Readers should look for evidence that a celebrated value corresponds to reproducible useful output rather than treating every reported measurement as interchangeable.
Independent measurement can strengthen confidence in the stated efficiency, but its scope remains the stated measurement. It does not automatically certify manufacturing yield, field durability, environmental containment or a warranty. A certificate can be valuable without answering those other questions. The fairest reading gives it credit for the uncertainty it reduces and does not make it responsible for claims it never tested.
The weather changes more than the amount of light
Outdoor sunlight varies in intensity and spectral composition. A tandem designed around one reference spectrum may see a different distribution of photons at another time, location or atmospheric condition. Temperature also changes as sunlight, wind and ambient conditions vary. A device's output can therefore move even if no permanent degradation occurs. Separating environmental response from lasting change is central to outdoor analysis.
Imagine a fictional tandem whose top and bottom cells are well matched at midday under a particular clear sky. In a different spectral condition, the balance of their generated currents changes. A lower combined output could then reflect current mismatch rather than damage. If an analyst attributes every departure from the reference measurement to degradation, the resulting lifetime estimate will be misleading.
The opposite mistake is possible as well. An analyst might explain a declining output entirely through weather while overlooking a genuine change in the device. A good outdoor study measures enough environmental context to test both possibilities. It asks whether the observed output is consistent with an unchanged device under the conditions actually experienced, and how that relationship evolves over time.
A 2025 study of a perovskite–CIGSe tandem reported year-round outdoor monitoring with maximum-power-point tracking. Its authors described a peak in performance followed by degradation and compared its instantaneous output with a single-junction reference. CIGSe is a different bottom-cell technology from silicon, so the study is evidence about that particular architecture rather than a direct forecast for perovskite–silicon modules. Farias-Basulto and colleagues' primary study
The useful inference is methodological: an outdoor time series can contain both changing environmental advantages and deteriorating device behavior. A maximum advantage on a favorable day and the total energy generated during a year answer different questions. Reporting both is more informative than allowing the best moment to represent the entire season.
Energy yield is an integral, not a photograph
A homeowner does not consume an efficiency percentage. The useful quantity is energy delivered over a period, after relevant system losses and interruptions. That energy is obtained by accumulating power over time. The shape of the power curve matters, not merely its peak. This is why a midday photograph of a meter or a single laboratory measurement cannot establish annual performance.
For a hypothetical comparison, imagine two equal-area modules exposed to the same conditions for three equal-duration intervals. Module A delivers 100, 80 and 60 power units. Module B delivers 85, 85 and 85. A has the higher initial output, but its total over those intervals is 240 unit-intervals compared with B's 255. The example is deliberately simple; it demonstrates how a beginning advantage can differ from accumulated output without predicting any real degradation rate.
A lifetime model needs further distinctions. Calendar age is not the same as hours under illumination, and neither is the same as accumulated exposure to a particular damaging condition. A panel can experience a long dark period, a short intense heat episode and repeated daily changes. Treating all elapsed hours as equivalent can hide the mechanism responsible for change.
Discounted financial value adds yet another calculation. Electricity produced earlier, later or during particular demand periods may have different value in a specific project. That does not change the measured physical energy. A responsible analysis keeps those quantities separate: what the module generated, what the system delivered and how that delivery is valued under stated assumptions. Otherwise a favorable financial scenario can be mistaken for superior device physics.
The practical consequence for a new tandem technology is not that initial efficiency ceases to matter. Higher power from limited area can be valuable. It means the advantage must be carried through a time-dependent comparison that includes degradation and availability. A durable advantage is a stronger proposition than a briefly higher number, and it requires evidence gathered on the relevant time scale.
Recovery can conceal deterioration
Some emerging photovoltaic devices exhibit changes that reverse partly after a change in conditions. If performance falls during one period and recovers during another, the choice of measurement time becomes important. A test taken only after a recovery period can give a different impression from continuous observation during useful operation. Neither should be mislabeled as the other.
The 2020 research-community consensus on stability assessment for perovskite photovoltaics explicitly addresses features including ion redistribution, reversible degradation and separating ambient effects from other stresses. It proposes practices based on ISOS procedures and states that these are not replacements for existing qualification standards. The primary abstract provides a useful statement of purpose without supplying a shortcut from a brief test to a service-life guarantee. Consensus statement in MIT's repository
A fictional daily pattern makes the distinction visible. Suppose a device begins each morning at 100 normalized units, falls to 90 during operation and returns to 100 after a dark period. Another device remains at 96 throughout. A morning-only check favors the first device. Depending on the duration and shape of the daytime change, accumulated energy may favor the second. The morning recovery is real, but it does not retroactively generate electricity that was unavailable during the day.
Now add a slow permanent decline to the first pattern. The morning baseline might become 99, then 98, while the daily reversible change continues. A sparse set of measurements could confuse the two processes. A well-designed analysis distinguishes the changing baseline from the reversible excursion and explains how each affects useful output. The distinction is more informative than a single line connecting two selected points.
Recovery also complicates the meaning of a threshold such as retaining 80 percent of initial performance. Initial can mean the first measurement, a stabilized value or the best value reached after conditioning. Those definitions can produce different threshold times from the same observations. A report should state its choice and show enough of the early trajectory for readers to understand its effect.
The chamber clock is not automatically a calendar
Accelerated testing exposes devices to selected stresses to learn about failure mechanisms or compare designs more quickly. The word accelerated does not mean every hour in a chamber has a universal outdoor equivalent. An acceleration relationship must connect a specific mechanism, stress condition and range of behavior. If the mechanism changes, a simple conversion can become invalid.
Consider an invented material whose measured change proceeds twice as fast under one elevated-temperature condition as under a lower-temperature condition. That relationship would not, by itself, establish that every outdoor damage process is accelerated by the same factor. Moisture exposure, illumination, electrical state and mechanical effects might respond differently. A single factor can be useful within a validated model and misleading outside it.
The order of stresses may matter as well. A component exposed first to mechanical damage and then to moisture might behave differently from an undamaged component exposed to moisture alone. A device that survives each separate test could still have a weakness under a sequence or combination. That possibility is a reason to investigate relevant interactions, not a claim that all single-stress tests are worthless.
The IEC's official description of IEC 61215-1:2021 makes an important boundary explicit: qualification test results are not quantitative predictions of module lifetime. It describes qualification for terrestrial photovoltaic modules and notes that service life depends on design, environment and operating conditions. This article relies on the publicly accessible scope statement, not an inspection of the full purchased standard or a claim that any featured device has passed its requirements. IEC's official scope
Qualification and field evidence can therefore complement one another. Controlled tests can reveal vulnerabilities and permit disciplined comparison. Outdoor studies can reveal interactions and operating histories that the selected tests do not reproduce fully. A compelling durability claim explains how these kinds of evidence fit together instead of demanding that one replace the other.
A good package can damage the device it protects
Encapsulation is intended to protect the photovoltaic stack from its environment. But the process of making the package can itself affect the stack. Heat, pressure, contact materials and subsequent mechanical behavior become part of the finished device. It is insufficient to test an unencapsulated research cell, attach a package and assume the original performance survives unchanged.
The first comparison should distinguish performance before packaging, after packaging and after any specified stabilization. If output falls during packaging and later partly recovers, both events belong in the account. Reporting only the best value from each stage can conceal an integration problem. Reporting only the immediate loss can also miss a reproducible later recovery. A time-resolved explanation is more useful than choosing whichever snapshot supports the preferred story.
The package also has optical functions. Light must pass through the front materials and reach the intended absorbers. Reflection, absorption or physical changes in those materials can affect output even when the underlying absorber remains functional. A decline in electrical performance does not identify the failing layer by itself. Diagnostic work must separate changes in incoming light from changes in charge generation and collection.
Interfaces create mechanical questions. Materials joined together can respond differently to changing temperature. Repeated changes can challenge adhesion or electrical connections. A small, carefully prepared sample may experience a different mechanical situation from a larger module with longer edges and interconnects. Scaling therefore changes the object whose reliability is being tested, rather than merely repeating the small device more times.
For manufacturers, packaging is consequently part of device development, not a final cosmetic operation. For readers, the essential question is whether the tested object resembles the object being offered. A headline about a bare cell may be scientifically meaningful while providing little direct evidence about a framed, interconnected product exposed outdoors. Both measurements belong in the development story, with their relationship explained.
Lead containment is a separate performance claim
Many high-performing perovskite photovoltaic research devices use lead-containing absorbers. Improving electrical durability does not automatically establish how a damaged device retains that material. Likewise, reducing release in a particular test does not establish that power output remains stable. These are related design objectives with different measurements and consequences.
A November 2025 study by Wang and colleagues investigated a self-healing encapsulant for perovskite solar cells. The researchers reported reduced lead leakage in their tests alongside retained electrical efficiency after specified damp-heat and thermal-cycling exposures. These results support further investigation of the combined protective approach. They do not establish that every damaged commercial module would contain all lead indefinitely, and the article does not convert their laboratory conditions into a universal outdoor safety claim. Wang and colleagues' primary paper
A containment result needs a clear denominator. Does a reported percentage compare release with an unprotected control, with the total lead present or with another reference? A large reduction relative to a control can be useful while leaving a nonzero release. The amount, test conditions and measurement limits determine what the result establishes. The percentage alone cannot answer every environmental question.
Damage assumptions matter too. A package can be tested after one defined kind of damage without representing every possible failure. A controlled exposure helps isolate mechanisms; it should not be described as an exhaustive simulation of storms, fire, transport damage or years of weathering. The relevant question is which credible situations have been investigated and how the evidence connects to the intended application.
Containment also has a time dimension. A protective material can be effective initially while changing through aging. An evaluation should consider whether its protective function persists, how that persistence is measured and what happens when the module reaches the end of service. This is not an argument for abandoning a promising material. It is an argument for treating its protective function as something that must be demonstrated alongside its electrical performance.
Manufacturing can change the reliability question
Research improvements often begin with a small number of carefully fabricated devices. Manufacturing asks whether the same useful properties can be obtained repeatedly over larger areas and many production runs. A process can raise the best measured efficiency while widening the distribution of ordinary results. A useful commercial assessment therefore needs the distribution, not only the champion.
A February 23, 2026 Nature Materials paper investigated all-vacuum-deposited perovskite devices and reported perovskite-on-silicon tandem cells along with outdoor testing in Italy. The authors described approximately 80 percent retained initial performance after eight months for the tested tandem arrangement. That is an architecture- and experiment-specific observation, not a lifetime warranty or a claim that all vacuum-deposited products behave identically. The primary vacuum-deposition study
The broader engineering question is how a fabrication change affects multiple properties at once. A process may alter the absorber, interfaces, uniformity and compatibility with the substrate. It may also change throughput, equipment requirements and the consequences of defects. Evaluating only initial efficiency can miss an advantage in reproducibility or a disadvantage in long-term behavior.
Imagine a fictional production trial with two processes. Process A yields one outstanding sample and several mediocre ones. Process B yields consistently good samples without matching the best result from A. The preferred process depends on the intended product and the cost of rejects, rework and field failures. A record table alone cannot decide that comparison. The sample count and selection rules are part of interpreting the result.
Manufacturing changes also create versioning questions. If a module's absorber formulation, contact layer or encapsulation changes, earlier durability evidence may not transfer completely. A company should identify which construction was tested and which construction is being produced. The aim is not to prevent improvement. It is to preserve the connection between a claim and the physical product that earned it.
Five samples can teach a lot without representing a fleet
A conference-paper record deposited on Zenodo on June 25, 2026 describes outdoor testing of five four-terminal perovskite–silicon mini-modules in Cyprus. The record links to a 2024 conference DOI, so the deposit date should not be presented as the date the experiments first occurred. Its abstract describes substantial early change followed by stability or improvement and notes that analysis used selected clear-sky days. This is valuable limited-sample field evidence, not a population-wide reliability estimate. The authors' conference record
The small sample count does not make the observations useless. Early field work can identify a mechanism, reveal a measurement problem or show that a laboratory assumption fails outdoors. Those are important outcomes. It becomes misleading only when evidence about a few research objects is stretched into a precise estimate of failure rates across future mass production.
Sample selection deserves equal attention. Were the tested modules chosen randomly from a production run, selected for high initial performance or built specifically for the experiment? Each approach can answer a legitimate question. They differ in what they reveal about the distribution a customer might receive. A field study of the best devices is not automatically a field study of a typical manufacturing batch.
The observation period also matters. A full year samples seasonal variation at one site, but it does not expose the device to every climate or every rare event. Multiple years add different information from multiple devices observed simultaneously during one year. A durability program benefits from both temporal and manufacturing diversity, with the limits of each dataset stated plainly.
A useful research report can therefore be modest in its statistical claim and ambitious in its scientific contribution. It can explain a failure mechanism convincingly without estimating a global failure rate. That distinction allows early outdoor work to guide development without forcing every study to pretend it has the scale of a mature industry qualification program.
Filtering weather data changes the question
Outdoor measurements can be noisy for legitimate physical reasons. Clouds pass, temperatures vary and sensors have limitations. Filtering can help answer a focused question by comparing more similar conditions. But the filtered result describes the selected conditions. If a study uses clear-sky periods to investigate device change, it has not automatically measured energy delivery across all weather through that filtered analysis alone.
Suppose a fictional dataset contains 100 days: 60 clear and 40 variable. An analyst selects the clear days to compare midday behavior. The resulting trend may be excellent evidence about that behavior. Multiplying its average output by all 100 days would require additional assumptions about the excluded days. The clarity of the trend does not supply those assumptions.
Data gaps can be selective too. If measurements fail most often during difficult weather or low output, excluding the gaps can make a device look better than the complete operating history would. A report should identify how much data are missing and whether the missingness is related to conditions of interest. A long test interval and a dense-looking plot can conceal a surprisingly incomplete record.
A useful publication can offer two views. One presents carefully filtered comparisons to investigate mechanisms. Another accounts for all relevant energy delivery, interruptions and missing periods using a stated method. The views need not produce the same impression because they answer different questions. Showing both prevents a mechanistic analysis from being mistaken for an unqualified yield statement.
The same care applies to normalization. Correcting output for irradiance or temperature can help reveal changes, but the correction model has assumptions. If the device's temperature response changes as it ages, a fixed correction may distort the trend. The model should be examined as part of the analysis rather than treated as an invisible step that converts raw data directly into truth.
Find the failing part before changing everything
When a tandem loses output, the total power measurement is an important symptom but an incomplete diagnosis. Changes in current, voltage and the shape of the electrical response can point toward different limitations. Optical or spatial observations can add information about where the device changed. No single diagnostic is guaranteed to identify the cause, but complementary observations can constrain the explanation.
Consider a fictional sample with a visible change in its front package and a simultaneous decline in output. The timing suggests a possible connection. It does not prove that the package caused the entire decline. An internal electrical change could occur at the same time. A useful investigation asks which additional observation would distinguish reduced light reaching the absorber from a change within the electrical stack.
This distinction matters because the proposed remedy depends on the cause. Improving an absorber may not solve a contact problem. A better barrier may not repair an incompatible interface. Replacing several features simultaneously can produce a better device while making it difficult to determine which change helped. That can slow later optimization or conceal a new trade-off.
Comparative experiments should therefore be designed around plausible mechanisms. A control that differs in one relevant feature can be informative, provided the rest of the fabrication and measurement is sufficiently comparable. Repeated observations across multiple samples help distinguish a reproducible response from a one-off defect. The result should preserve uncertainty when several explanations remain consistent with the data.
There is a practical advantage to this restraint. If a team knows which failure mode limits a device under particular conditions, it can direct development toward that mode and then test whether a different limitation becomes dominant. Durability improvement is often an iterative process. Declaring a device stable in the abstract can obscure that sequence and make the next failure seem inexplicable rather than informative.
The same retention percentage can hide different losses
Normalized performance is useful for comparing changes, but normalization removes information about the starting point. If one fictional device begins at 30 percent efficiency and another at 20 percent, retaining 90 percent of initial efficiency leaves them at 27 and 18 percent respectively. The retained fraction is identical; their absolute output under equal illumination and area is not. A chart of normalized performance should therefore be accompanied by the underlying values and their measurement conditions.
The distinction becomes more subtle when a tandem contains independently accessible subcells. Suppose an idealized four-terminal arrangement initially supplies 60 power units from its upper device and 40 from its lower device under a particular illumination. If the upper contribution falls by half while the lower remains unchanged, the combined output becomes 30 plus 40, or 70. The upper component has retained 50 percent, while the combination has retained 70 percent. Either statement can be accurate, but they describe different objects.
This arithmetic should not be transferred mechanically to a two-terminal series-connected tandem. Its subcells share an operating current, and a change in one can shift the combined operating point. Nor does the four-terminal example account for optical changes that might alter the light reaching the lower cell. Those limits are precisely why the architecture and measured boundary need to accompany a retention number. A component percentage is not automatically the percentage for the finished device.
A comparison with a reference module introduces a third quantity. In the invented example, if a reference provides 65 power units under the same conditions, the degraded tandem still provides five more. It can therefore outperform that reference at the measurement time while having lost a substantial fraction of its own initial output. The two findings are compatible. Describing only the advantage conceals degradation; describing only degradation conceals the remaining comparative output.
A complete comparison should also account for what the reference itself does over time. A reference that deteriorates, becomes dirty or is measured differently can distort the apparent advantage. It needs its own characterization and maintenance record. The label reference does not grant immunity from physical change, and a ratio between two changing devices can remain stable while both become less productive.
For a reader, the practical interpretation is to ask for three views when relevant: absolute output under defined conditions, change relative to the device's own stated baseline, and change relative to an independently characterized comparison. Each view reveals something the others can hide. They are complementary descriptions of performance, not competing ways to select the most favorable headline. Keeping them together makes a complex outdoor result easier to understand without flattening it into a single percentage.
The service decision includes replacement
A module's useful life is partly a technical property and partly a decision about acceptable service. A device may continue producing electricity after its output has fallen below the level expected in a project. Whether to keep it, replace it or repair associated equipment depends on costs, access, safety requirements and the value of the remaining output. A threshold used in a research paper is not automatically the owner's replacement rule.
A hypothetical economic example makes the distinction concrete. Suppose one installation produces an additional 20 energy units per year for its first five years compared with an alternative, but then requires a replacement costing the equivalent of 150 units of value. Ignoring discounting and other differences, the early advantage contributes 100 units and does not by itself cover the replacement. Different prices, lifetimes or service arrangements could change the comparison. The example is arithmetic, not a forecast of tandem economics.
Replacement has physical consequences beyond its purchase price. Access to a roof, compatibility with remaining modules, interruption of operation and management of removed material all matter. A technology that delivers higher initial power may still be attractive, but its assessment should include these consequences rather than treating a declining panel as if it vanished harmlessly from the calculation.
End-of-service arrangements should be considered while the product is being developed. The ability to collect, transport and process removed modules under applicable requirements is separate from their initial electricity performance. Protective packaging that is effective during operation may influence later disassembly. Those choices involve trade-offs that should be investigated rather than assumed away by calling the product recyclable.
A warranty can allocate some financial responsibilities, but it does not change the physical degradation process. Its usefulness depends on its terms, the evidence required for a claim and the continuing ability of the responsible organization to honor it. Scientific durability evidence and a commercial promise are complementary kinds of assurance, not substitutes for one another.
What deserves confidence now
The inspected studies show why outdoor work should be treated as a central part of perovskite development rather than an inconvenient final exam. Some observations reveal substantial degradation. Others show useful retention over the tested interval or complex recovery behavior. These are not contradictory verdicts on a single uniform technology. They concern different materials, architectures, packages, climates and generations of devices.
A strong claim identifies those particulars. It connects a specific construction to a measurement, keeps the test duration visible and distinguishes electrical output from containment performance. It explains whether the quoted result describes a small cell, mini-module or larger product. It avoids transforming a chamber exposure into an unsupported number of outdoor years or a research sample into a statistically representative fleet.
The central opportunity remains significant: extracting more useful electricity from limited illuminated area. But the value of that opportunity depends on preserving performance through the conditions that real devices encounter. The most useful next advance might be a higher initial efficiency, a more reproducible manufacturing process, a better package or a clearer explanation of a field failure. Different advances remove different obstacles.
For readers following the field, the decisive question after a new efficiency announcement is straightforward: what happened when the device kept operating? The answer should include the weather, the electrical operating state, the measurement method and the changes that occurred. That longer answer is where a promising absorber becomes an understandable product—and where durable solar progress can be distinguished from a percentage that survived only long enough to make a headline.