A 34% Tandem, a 26.7% Record, and the 4.1x Stability Gain the Headlines Skipped

Between September 2 and September 10, three research teams announced perovskite tandem advances, two of them records. LONGi and Soochow University reported a perovskite-silicon tandem at 34.0% efficiency with an independently certified open-circuit voltage of 2.014 V (Science Bulletin). Korea's KIER reported a 26.7% perovskite-CIGS tandem certified by Germany's Fraunhofer ISE, a new world record for that thin-film pairing. A second KIER team reported a dual-molecule passivation treatment that lifted an inverted perovskite cell from 21.21% to 24.6% efficiency. Efficiency numbers made the headlines, but the number that should have is 4.1.

LONGi's paper carried a stability result that most coverage treated as a footnote: the new cell retained 84% of its efficiency after 2,000 hours, while a control without the new interfacial layer sagged to 70% after 1,000 hours. Fit both data points to the standard exponential decay model and the treated cell projects a T80 of roughly 2,560 hours against about 625 for the control: four times slower death. In a field where durability has been the commercialization bottleneck for a decade, that ratio matters more than any record.

The week's scoreboard

Device	Stack	Reported efficiency	Certification	Durability data
LONGi + Soochow University	Perovskite (inverted p-i-n) on silicon heterojunction	34.0% (Voc 1.997 V, Jsc 20.36 mA/cm2, FF 83.62%)	Voc 2.014 V independently certified	84% after 2,000 h; control 70% after 1,000 h
KIER	Perovskite on CIGS (thin-film)	26.7% (27% lab-measured)	Fraunhofer ISE	Not reported
KIER / UNIST / Kunsan National Univ.	Inverted single-junction perovskite	24.6% (control 21.21%, PDAI-only 23.17%)	Reported, not independently certified	Not reported
LONGi (July 2026, context)	Perovskite-silicon	35.5%	ESTI certified	Current world record

The math nobody ran

Model efficiency retention as exponential decay, R(t) = e^(-lt). For the treated cell, 0.84 = e^(-l x 2000), so l = 8.72 x 10^-5 per hour; T80 is the time t where R = 0.80, t = -ln(0.80) / l, about 2,560 hours. For the control, 0.70 = e^(-l x 1000) gives l = 3.57 x 10^-4 per hour and a T80 near 625 hours, so the ratio is 2,560 / 625 = 4.1.

Next, a consistency check on the headline number itself. Multiply the three reported cell parameters: 1.997 volts x 20.36 milliamps per square centimeter x 0.8362, divided by the 100 milliwatts per square centimeter of standard test illumination, and the answer is 34.00%: the paper's headline checks out, a verification worth doing because efficiency claims at this precision have a history of arithmetic optimism.

The voltage checks out too: a certified 2.014 volts minus a typical silicon heterojunction subcell's 0.74 leaves roughly 1.27 for the perovskite top cell, exactly the deficit its bandgap predicts.

Now the scale penalty the records obscure. KIER's 26.7% was measured on a device smaller than one square centimeter. In the same perovskite-CIGS class, the record for devices larger than one square centimeter stands at 25.5%, held by Helmholtz-Zentrum Berlin. Leaving the lab dot costs 1.2 percentage points before you even reach module size. A May 2026 result makes the gap concrete: 34.18% on a small tandem cell versus 31.00% on a large-area multicell string, a 3.18-point haircut for scaling up.

What the efficiency actually buys is concrete: a 34%-efficient panel produces 340 watts per square meter under standard illumination; a typical commercial silicon panel near 22% produces about 220. That is 54.5% more power per rooftop or per acre. A gigawatt of capacity needs roughly 2.94 square kilometers at 34% versus 4.55 at 22%, about 35% less land.

Why the buried interface is the whole game

All three September results attack the same enemy: the buried interface, where charge carriers recombine instead of flowing into the circuit. LONGi and Soochow slipped discrete monoclinic zirconia nanoparticles between the transparent conductive oxide and the self-assembled monolayer, with X-ray photoelectron spectroscopy confirming covalent zirconium-oxygen-phosphorus bonds that anchor the monolayer twice over. The layer is porous and discontinuous, so holes still extract cleanly while leakage pathways close, which is exactly why the voltage reached 2.014 certified volts.

KIER's CIGS team did the analogous job on the thin-film stack: new interface materials plus processing tweaks that protect the perovskite layer during integration, and a redesigned top electrode that wastes less light to parasitic absorption.

From the chemistry side, the dual-molecule team worked the same problem. Small PDAI molecules plug voids deep at grain boundaries; then larger 4TF molecules bind the undercoordinated lead atoms left exposed on the surface. Two molecules, two defect populations, one complementary effect, and the numbers confirm it: 24.6% for the pair versus 23.17% for PDAI alone and 21.21% untreated. Architecture is the link: inverted perovskite cells fabricate at relatively low temperatures, the commercialization-friendly route, and it is the same inverted p-i-n structure LONGi used for its record tandem's top cell. Treat the toolbox paper and the record paper as chapters of one story.

Limitations

This analysis has real blind spots, starting with the central number: the 4.1x comes from fitting two retention data points to an exponential model. Real degradation curves bend, accelerate, and sometimes plateau, and the test conditions behind those two points, illumination, temperature, tracking, encapsulation, were not disclosed in the reporting I could verify. Treat 2,560 hours as a rate comparison between two cells tested side by side, not as a lifetime prediction. Either way, about 107 days of continuous operation.

One more caution: the scoreboard mixes certified and reported numbers measured on wildly different areas. KIER's 26.7% is a sub-one-square-centimeter device, and the dual-molecule 24.6% has no independent certification. Lab hours are not field years, and no perovskite tandem carries a bankable 25-year field warranty. Perovskite absorbers contain lead, so end-of-life sequestration is an unsolved problem at scale. As a caution, an indium-free 31.0%-certified tandem mini-module kept only 65% of its output after 105 days outdoors: the outdoors is harsher than the lab. And the 34.0% does not beat LONGi's own 35.5% July record: this paper's contribution is voltage plus stability.

The strongest case against tandems

Record treadmills have promised commercialization before and been wrong. Perovskite efficiency announcements have outrun deployment for a decade, and each new record arrives on a device too small, too unsealed, and too short-lived to sell. Tandems multiply the interfaces, and every interface is a potential failure mode across 25 years of thermal cycling, humidity, and hail. Damp-heat degradation remains the technology's open wound: no bank finances a 25-year warranty on a technology without 25-year field data, so bankability, not efficiency, decides what gets built.

Meanwhile the incumbent refuses to stand still: commercial silicon keeps getting cheaper and more efficient, so every tandem gain is scored against a moving, cheapening target. That skepticism is the correct default, and it is also why the 4.1x durability move, if it replicates under published stability protocols, is the number to watch instead of the next efficiency point.

What you can do

If you are buying solar panels this year, do not wait for tandems. All three record devices are laboratory cells; commercial tandem modules are years from your roof, and today's inexpensive silicon panels will have paid for themselves by the time tandems arrive.

If you allocate capital to clean energy, stop ranking teams by record efficiency; ask for three numbers: T80 under published ISOS stability protocols, damp-heat results per IEC 61215, and the device area the record was measured on. Back the team that answers all three without flinching; that is the team worth a second meeting.

If you work in the field, the durable moat is the interface toolkit: self-assembled monolayers, scaffold layers, dual passivation chemistries. Records get broken within months; a manufacturable interface stack compounds for years.

If you make policy, write the end-of-life rules now. Perovskite modules contain lead, and gigawatt-scale deployment without sequestration and recycling standards defers a cleanup bill to someone else. Those standards belong in place before the first commercial tandem warranty is signed, not after.

The Bottom Line

Three records in eight days would be a story about speed; the better story is about survival. A 4.1-fold slowdown in degradation, projected from published data, attacks the bottleneck that has kept perovskite tandems in the laboratory for a decade. Efficiency records are scored in percentage points; commercialization is scored in decades. The week's most important number was not 34.0 or 26.7 but 4.1, the factor by which a better buried interface slows a solar cell's death. That is the number that decides whether any of these records ever leave the lab.

Sources: pv magazine on the LONGi/Soochow 34.0% tandem (Sep 10, 2026); pv magazine on the KIER 26.7% perovskite-CIGS record (Sep 2, 2026); TechXplore on dual-molecule passivation, Akyaw et al., ACS Applied Materials & Interfaces (Sep 2, 2026); TechXplore on graded-dielectric tandems at 34.18% (May 2026); CleanTechnica on LONGi's 35.5% world record (Jul 2026); pv magazine on the indium-free 31.0% mini-module (Jul 2026).