Perovskite Solar Cells: Promise and Problems

Ask ten perovskite researchers what the field's main problem is, and you'll get two answers. One camp says the efficiency race is basically won and the real work is making cells survive a desert summer for twenty-five years. The other says stability will follow, as it did for silicon, once the materials science matures, and that efficiency and stability gains are coupled anyway. Both camps publish in the same journals. They often read each other's papers with mild irritation.

I've spent long enough around photovoltaics to have sympathy for both. I also think the second camp is underestimating how different perovskite degradation is from anything silicon had to overcome. Below I walk through why these materials caused such excitement, what the efficiency record actually tells us, and where the durability evidence is strong and where it's thin. Then I'll say what I'd watch over the next couple of years.

Why a mineral structure from the 1830s became a solar material

"Perovskite" names a crystal structure, ABX3, not a single compound. In the photovoltaic versions, A is a monovalent cation (methylammonium, formamidinium, cesium, or a mix), B is usually lead or tin, and X is a halide. The structure tolerates a wide range of substitutions, which is both its gift and its curse. You can tune the bandgap from roughly 1.2 to over 2 eV by changing composition. Every one of those knobs also changes how the lattice responds to heat, light, moisture and electric fields.

The first perovskite photovoltaic device reported 3.8% efficiency. It used the perovskite as a sensitizer in a liquid-electrolyte cell that degraded within minutes (Kojima et al., Journal of the American Chemical Society, 2009). Few people paid attention. The turning point came in 2012, when solid-state devices crossed 10%. Snaith's group then showed that the perovskite didn't need a mesoporous TiO2 scaffold to transport electrons; it could do the job itself (Lee et al., Science, 2012). That was the moment the material stopped being a dye and became a semiconductor in people's minds.

The physics explained the excitement. Lead halide perovskites absorb strongly, so a film a few hundred nanometers thick does the work of much thicker silicon. Carrier diffusion lengths in mixed-halide films exceeded a micrometer (Stranks et al., Science, 2013). That was surprising for a polycrystalline film deposited from solution at low temperature. The usual explanation is "defect tolerance": many of the easiest defects to form sit shallow in the bandgap or in the bands themselves, rather than deep in the gap where they would kill carriers. That account is broadly accepted, though how far it extends to real, messy films is still argued over.

What the efficiency record actually shows

This is the best-evidenced part of the story, and it's worth being precise about why. Record efficiencies are certified by independent test labs and compiled on the NREL best research-cell efficiency chart. That's about as close as photovoltaics gets to a pre-registered, externally adjudicated outcome. Single-junction perovskite cells went from under 4% to the mid-20s in roughly a decade, and certified values now sit around 27%. Silicon took about half a century to reach comparable territory.

The more consequential numbers are for tandems. A wide-bandgap perovskite stacked on a silicon bottom cell can, in principle, beat the single-junction limit that caps silicon at about 29% in practice. Monolithic perovskite–silicon tandems passed 29% in 2020, partly thanks to self-assembled monolayer hole contacts that cut interfacial losses (Al-Ashouri et al., Science, 2020). Certified tandem records have since climbed past 34%, with industrial labs as well as academic groups trading the lead.

A few caveats that experts know but press releases don't mention:

• Record cells are typically small, often well under 1 cm². Efficiency drops on scale-up because of film non-uniformity, interconnection losses and the resistance of transparent electrodes.

• A certified efficiency is a snapshot. It tells you nothing about what the cell does after 1,000 hours under light, or after a year outdoors.

• The early literature had a real measurement problem. Current–voltage hysteresis meant the reported efficiency depended on scan direction and rate (Snaith et al., Journal of Physical Chemistry Letters, 2014). Stabilized power output reporting is now expected, but you should still check for it when reading older papers.

My view: the efficiency question is settled enough that a new 0.3% record, on its own, shouldn't change anyone's research priorities.

Why perovskites degrade

The same soft, ionic lattice that makes perovskites easy to process makes them easy to disturb. The mechanisms are reasonably well mapped, even where the remedies aren't.

Moisture, oxygen and heat

Methylammonium lead iodide hydrates in humid air and decomposes, and methylammonium is volatile at module operating temperatures. Much of the compositional progress since 2015 has involved replacing or diluting methylammonium with formamidinium and cesium. That approach is close to consensus now. Oxygen combined with light can also drive photo-oxidation, especially in films with many iodide vacancies.

Ion migration

Halide ions and vacancies move under modest electric fields at room temperature. This is the root of hysteresis. It also contributes to reversible and irreversible losses during operation, and to chemical reactions with contacts. Mobile iodide can reach metal electrodes, and metal can diffuse inward. In my experience, people outside the field underestimate this. Silicon has nothing comparable.

Light-induced phase segregation

Mixed iodide–bromide compositions, which you need for the wide bandgaps used in tandems, can separate under illumination into iodide-rich and bromide-rich domains. This lowers the effective bandgap and the voltage (Hoke et al., Chemical Science, 2015). The effect is often reversible in the dark. Its drivers (polarons, strain, defect density, carrier gradients) are still debated. That matters a great deal for tandems, which depend on exactly these compositions.

Interfaces and contacts

Many degradation pathways turn out to be interface problems rather than bulk ones. Examples include lithium-doped organic hole transporters, UV-activated TiO2, reactive metal electrodes, and defect-rich grain boundaries at buried surfaces. That's why so much recent high-impact work is really interlayer engineering. One example is coherent interlayers between the perovskite and SnO2, which improved both efficiency and operational stability (Min et al., Nature, 2021). Papers like this are the strongest support for the "coupled problems" camp. Fixing interfaces often helps on both fronts.

Grading the durability evidence

The brief I usually give PhD students is to sort stability claims by how they were generated, not by the journal. Here's roughly how I'd tier the current literature.

What's established

Encapsulated perovskite devices can pass individual standard stress tests that would have been unthinkable a decade ago. One widely discussed example used tailored 2D/3D heterojunctions to survive extended damp-heat exposure at 85 °C and 85% relative humidity (Azmi et al., Science, 2022). The field has also converged on consensus testing protocols, the ISOS framework adapted for perovskites, which makes cross-study comparison far less chaotic than it was. Composition choices that reduce volatile components and interface layers that block ion and metal diffusion reliably improve stability.

What's emerging

Lifetime prediction is the frontier, and it's contested. Some groups have used temperature-accelerated aging with Arrhenius-style analysis to extrapolate multi-year operating lifetimes for interface-stabilized, all-inorganic cells (Zhao et al., Science, 2022). This is careful work, and it's the right direction. My hesitation is that extrapolation assumes a single dominant, thermally activated failure mechanism. Perovskites in the field face light, heat, bias, humidity ingress, daily temperature swings and partial shading together, and those stresses interact. Reverse-bias degradation under shading, for instance, has only recently received sustained attention, and it isn't captured by standard light-soaking tests.

Large open datasets compiled from thousands of published devices are also starting to show which compositions and architectures correlate with longer reported lifetimes. They're the closest thing this field has to a meta-analysis, and they carry the same weakness. Reporting is heterogeneous, and the published record is biased toward devices that worked.

The lead question

Most high-performing perovskites contain lead in a water-soluble form. The amount per square meter is small, and the debate is about whether it matters. One camp points out that silicon modules have historically contained lead in solder, and that the quantities in a perovskite layer are modest. The other notes that lead halide is far more bioavailable than lead metal in solder, so a cracked module in heavy rain is a different scenario.

Mitigation work is promising but young. Lead-absorbing coatings and encapsulants designed to capture lead released from damaged devices have shown substantial sequestration in lab tests (Li et al., Nature, 2020). Tin-based alternatives avoid lead but oxidize readily and lag well behind on efficiency. My honest read is that lead is a solvable regulatory and engineering problem, not a fundamental barrier. But it has to be solved in the product, with end-of-life recycling planned, not argued away in discussion sections.

Where I land on the two camps

The "stability will follow" camp is right that the efficiency and stability literature has converged more than skeptics expected. Better interfaces and cleaner compositions have delivered both. But I think they're wrong to treat silicon's history as a template. Silicon's degradation modes are slow, well characterized and mostly about packaging. Perovskite degradation starts in the absorber itself, through ions that move at room temperature. That's a different kind of problem.

If I were starting a PhD in this area now, I wouldn't chase single-junction records. I'd work on one of three things: failure under combined, realistic stressors; the wide-bandgap compositions tandems actually need; or the gap between small cells and full-size modules. That's where the publishable unknowns are, and where industry will be looking for people.

What to watch next

• Outdoor field data on commercial tandem modules. The first commercial perovskite–silicon tandem products have started shipping. Independent multi-year performance data from real installations will matter more than any lab record.

• Certification under full qualification sequences. Look for modules, not cells, passing the complete IEC test sequence, ideally with results that other labs reproduce.

• Reverse-bias and shading tolerance. Expect more work on device designs that survive partial shading without irreversible damage.

• Validated lifetime models. The key test is whether accelerated-aging predictions match outdoor degradation rates for the same device stack. Until someone closes that loop, extrapolated lifetimes stay in the "emerging" column.

• Wide-bandgap stability. Progress against phase segregation in the 1.65–1.8 eV range will set the ceiling for tandem durability.

• Lead management in practice. Watch for recycling pathways and encapsulation standards that regulators actually adopt.

The literature here moves fast and sprawls across materials, physics and engineering journals. That makes it easy to miss the paper that contradicts the one you're citing. When I'm tracking a thread like reverse-bias degradation, I keep the search, the reading notes and the draft in one place. An AI-native workspace like PACR handles that kind of literature tracking well, because it links the summaries to the papers they came from. That way, when I return to a stability claim six months later, I can see which study it came from and how it was tested.

If you're building a review or a grant section on perovskite durability, sort your sources into the tiers above before you write a word. You may find, as I did, that the strong column is shorter than the field's confidence suggests. If it helps, you can set up a PACR workspace and start that sorting from your first search.