A solar cell bent 5,000 times around a curve tighter than a thumb kept 91 percent of what it started with, while the flexible tandem inside it reached 33.6 percent certified efficiency on a silicon layer thin enough to roll

The jig takes both ends of the strip, folds it toward itself, holds it, and lets go.

Then it does it again.

It ran that way for days, and the film in the clamps looks like nothing at all. Not a panel, not glass, barely a component.

Voltage is the giveaway, and it came out at 2.015 volts open circuit, a record for a bendable device and level with rigid cells of the same design.

Yet the interesting part of the test is not whether the thing survives the folding.

It is why folding should cost it anything in the first place.

Why folding costs a stacked cell anything at all

Two semiconductors are sitting on top of each other in there.

The upper one, a perovskite, takes the high energy blue and green light. The one below it, ordinary crystalline silicon, takes the red and infrared that slips through.

Between them sits a layer most people never hear about, a few atoms of oxide whose only job is to let electrons arriving from above meet holes arriving from below and cancel out cleanly.

That seam is the whole problem.

When the sandwich bends, the two materials do not stretch by the same amount. They want to slide against each other, and where they slide, gaps open at the boundary.

Carriers fall into those gaps. Voltage drops. The cell does not crack, it just quietly stops adding up.

So the seam is where the work went. The team built that middle layer out of indium oxide co doped with cerium and hydrogen, laid down by reactive plasma, which grips both faces tightly enough to take the movement.

Voltage is the giveaway, and it came out at 2.015 volts open circuit, a record for a bendable device and level with rigid cells of the same design.

Two layers of light and a wafer thin enough to roll

A single material has a ceiling.

Tune it to one slice of the spectrum and it wastes the rest, which caps a one junction cell at roughly 33.7 percent in theory.

Stacking two materials with different appetites is the only way past that line, and it is why every efficiency record of the last few years has been a stack.

The bending is a separate trick entirely.

An ordinary solar wafer runs 150 to 180 micrometers thick and snaps rather than bends.

Thin it down far enough and silicon becomes flexible, and a comparable flexible tandem announced weeks earlier was built on a wafer of 60 micrometers, thinner than a lot of human hair.

What the certified numbers are and where they come from

The result was published in Nature on 10 November 2025 by a group at Soochow University in Suzhou with collaborators in Saudi Arabia.

The efficiency is 33.6 percent certified, not a self reported bench figure.

After 5,000 bending cycles at a radius of about 0.69 inches, roughly the curve of a thumb, the cell held 91 percent of where it started.

Not 97. The 97 percent figure belongs to a different flexible tandem, bent 43,000 times around a gentler curve.

It also kept 90 percent after 1,000 hours of damp heat, and ran past 2,000 hours of continuous illumination before dropping to 80 percent of its initial output.

Where the gap is wider than it reads

The comparison everyone reaches for is the rigid record, and that number moved.

It was 34.85 percent, certified by the American national laboratory in the spring of 2025, which puts the flexible cell 1.3 points behind.

But a European test institute certified 35.5 percent in July of this year, so the real distance is closer to 1.9 points, as the paper and the record announcements together make clear, and the newer figure came with no active area disclosed at all.

Then there is the jump from a fingernail sized coupon to a factory. Perovskite is grown from liquid chemistry, manageable at that size and stubborn across a full sheet, where pinholes and thickness gradients eat the yield.

The commercial reference point makes that concrete. The first tandem modules sold anywhere shipped in September 2024 at 24.5 percent, about nine points below this cell, and they are rigid glass.

Verification is its own story, as a Finnish battery cell showed when independent tests refused to agree, and the road from a laboratory win to a product is the same one an iron catalyst is still walking.

What a bendable record actually changes

Not rooftops. Glass is cheap, flat and already there.

What changes is the argument.

Flexibility used to mean accepting a serious efficiency penalty, and that penalty has now shrunk to almost nothing in the laboratory.

That moves the question from whether it can be done to whether it can be made in volume, which is a more tractable kind of problem.

The uses that follow are the awkward ones. Curved building surfaces, vehicle skins, equipment carried into places with no grid, all of them shapes a glass panel simply cannot take.

What is still missing is a single outdoor season of data on a bendable stack, and until that exists the strip in the jig is a very good argument rather than a product.