A single-junction silicon cell has a ceiling. It is called the Shockley–Queisser limit, it sits near 33% under one sun, and it exists because one bandgap can only politely harvest one slice of the spectrum — the blue photons waste their surplus energy as heat, the deep-red ones slip through uncaught. So when a lab clears 30% on a cell built to beat that limit, the interesting question is not how fast. It is how.

The answer this month comes from Lausanne and Neuchâtel. EPFL’s PV-Lab and CSEM, as pv magazine reported, have certified a power conversion efficiency of 30.02% in a triple-junction cell — two thin-film perovskite layers stacked on one silicon bottom cell — with the measurement validated by the Shanghai Institute of Microsystem and Information Technology. That edges past the previous triple-junction mark of 27.10% held by a National University of Singapore team, and it is the first device of this architecture to break 30 on the number that pays the bill.

What they actually fixed

Stacking bandgaps is old physics; making three of them agree is the hard part. The team named two failure modes and closed both. First, the top perovskite was leaking voltage, so they introduced a molecule that steers crystal growth and passivates defects, pushing the top cell to 1.4 V under sunlight. Second — and this is the elegant bit — the middle layer was starved of current, so they added silicon-oxide (SiO_x) nanoparticles between the silicon base and the middle perovskite to bounce near-infrared light back up into the cell that needed it. That is pure photon management: no new material heroics, just sending the right wavelength to the right layer.

Lead author Kerem Artuk put the frame plainly: this approach reaches toward efficiencies once reserved for III–V multi-junction cells flown in space, which touch 37% and cost roughly 1,000× more per watt. PV-Lab head Christophe Ballif noted the first demonstrator, back in 2018, managed only 13%. The stated headroom for triple junctions is above 40%.

←TODAY: 30.02% certified by SIMIT — the first perovskite-silicon triple-junction over 30%, out of EPFL + CSEM. →3012: a façade that out-earns its own embodied carbon inside its first decade, not its third. Fulcrum: the record only matters if you can also see the century the wall has to survive — a champion cell is not yet a specified module.

Why an architect should keep one number

Here is the desk translation. A building has a fixed surface budget — so much south glass, so much spandrel, so much roof. At 20% module efficiency, a BIPV façade is a nice supplement. Push the delivered figure toward 30% and the same square metres cross from garnish to genuine generator, which changes the economic case for wrapping a tower in active glass rather than dead cladding. PAZ has walked this thread before — our own next-generation-PV read flagged perovskite tandems as the one to watch and bifacial as already-shippable. This result nudges the tandem line one column to the left.

But keep the arithmetic honest: 30.02% is a certified lab cell, typically around a square centimetre, not a certified module on a pallet. The gap between champion cell and warrantied product is where perovskite has always struggled — UV, heat and moisture stability over the 25-to-40-year life a façade actually needs. Meanwhile LONGi has just claimed 35.5% for a two-junction silicon-perovskite tandem; the numbers are climbing on every axis, which is exactly when a specifier should get more careful, not less.

Atelier: The Büro pressure this quarter is the client who read the same headline and now wants “the 30% cells” on the competition render. Do not spec a lab record into a hundred-year envelope. This Monday, add one clause to your BIPV performance spec: the module — not the record cell — must carry an independently certified efficiency and an accelerated-ageing (IEC 61215) datasheet, and the tender must state delivered module efficiency, not press-release cell efficiency. That single line converts awe into something a QS can price.

Hack: Feel why the middle-layer fix mattered. Series-stacked junctions share one current path, so the whole cell is throttled by its thirstiest-starved layer — the weakest link sets the ceiling, while the voltages simply add. Run this and watch the stack current snap to the minimum:

J = [20.1, 19.4, 20.3]      # mA/cm^2 photocurrent per sub-cell (top, mid, bottom)
J_stack = min(J)            # series constraint: weakest link caps the stack
V_stack = 1.40 + 0.62 + 0.68  # volts add down the stack
P = J_stack * V_stack       # mW/cm^2 -> the number that moves
print(round(P, 1))          # nudge J[1] up to 20.2 and re-run

Raise J[1] — the middle cell — and the output jumps; raise the already-strong bottom cell and nothing changes. That is the SiO_x reflector’s whole job in three lines.

Bring one question to your next envelope meeting: are we buying a headline or a datasheet? Ask for the certificate and the ageing curve before the cell goes on the drawing.

Source: pv-magazine.com