Perovskite–silicon tandems have a geometry problem that lives at the bottom of the stack. The silicon cell underneath is etched with microscopic pyramids that trap light, but the perovskite layer that must coat them is thinner than a micrometre — thinner, in places, than the pyramids are tall. A paper published October 31 in Advanced Energy Materials reports a fix that works on the silicon itself rather than on the perovskite: shrink the pyramids to submicron size, then round their tips and steps with a three-minute ozone etch. The resulting tandem reaches 33.25% on a 1.014 cm² cell. That is not a record — certified tandems have passed 35% — but the number is not the point. The point is that the texture trade-off, which has forced tandem makers to choose between light trapping and a clean coating, can be engineered away at the surface.

The study, “Chemical Rounding Suppresses Morphology-Induced Interfacial Losses in Perovskite–Silicon Tandem Solar Cells,” comes from a team led by Nan Sun at Forschungszentrum Jülich and the University of Duisburg-Essen, with colleagues including Binbin Xu and Gaosheng Huang. What makes it worth a full read is that the team diagnoses two distinct ways the texture hurts the device — one optical, one crystallographic — and shows a single surface treatment addressing both.

The trade-off: pyramids against coating

In a monolithic perovskite–silicon tandem, the silicon’s surface texture does two jobs at once. Optically, the pyramids scatter incoming light, lengthening its path through the wafer so the silicon absorbs more of the infrared that passes through the perovskite. Mechanically, the same pyramids are the substrate the perovskite must coat: a solution-deposited film, well under a micrometre thick, spread over peaks and valleys.

Micron-scale textures — the 2–3 µm pyramids of industrial heterojunction cells — trap light well but demand evaporation-based perovskite processing for adequate coverage; a spin-coated film cannot drape itself over pyramids several times its own thickness. Submicron pyramids suit solution-processed perovskites better, but scatter less infrared light unless the morphology is carefully engineered.

The team’s first move is to quantify the trade-off. Optical simulations show a flat silicon surface giving a bottom-cell photocurrent of 18.9 mA/cm², texturing lifting it above 20 mA/cm², and micron-scale texture pushing it past 21 mA/cm² — but only at the expense of perovskite coverage. Infrared absorption rises with pyramid height up to roughly 500 nm, then levels off, and a symmetric, double-side submicron texture pushes the pathlength enhancement factor above 7, against 2.1 for a planar surface. The design window that emerges is a symmetric submicron texture of moderate pyramid height — and the rest of the paper is the attempt to build exactly that.

The hidden defect: micro-terraces and partial epitaxy

Building a controlled submicron texture is harder than building a micron one. Micron texturing is a fast, anisotropic alkaline etch; submicron texturing needs slower etching in the reaction-limited regime, and the slow kinetics introduce a defect micron pyramids rarely show: nanoscale irregularities and micro-terraces.

The team controls pyramid height by adjusting the potassium hydroxide concentration, at 80 °C for seven minutes with a commercial texturing additive. At 0.23% KOH the pyramids average 248 nm tall; at 0.28% KOH the mean height reaches 363 nm, close to the simulated optimum; at 0.33% KOH they grow larger but the distribution broadens, raising the risk of incomplete coverage and shunting. Average height alone, the paper notes, is not an adequate criterion: the distribution matters as much as the mean.

But the slow etch also leaves the pyramid facets imperfect: because the etching depth is shallow, the (111) facets do not fully develop, and small local variations show up as facet truncation, shallow depressions, and low-angle micro-terraces — locally flattened regions with extremely small miscut angles.

That matters because of what happens next. The silicon heterojunction bottom cell is built by depositing hydrogenated amorphous silicon (a-Si:H), and a-Si:H growth is exquisitely sensitive to the microscopic structure underneath. The micro-terraces act as atomically favourable templates for ordering, so the amorphous film starts to grow in partial registry with the crystalline substrate. Atomic-resolution microscopy shows lattice fringes extending continuously from the crystalline silicon into the a-Si:H layer — partial epitaxy at the interface. Those epitaxial features introduce structural defects and deteriorate interface passivation, limiting carrier lifetime. The texture is not only a coating problem for the perovskite on top; it is a crystallographic problem for the amorphous silicon underneath, costing the bottom cell its passivation before the perovskite is ever deposited.

The fix: a three-minute ozone round

The treatment is deliberately unexotic: a three-minute ozone-assisted chemical rounding step in an isotropic solution of hydrofluoric acid, hydrochloric acid, and 20 ppm ozone. Ozone oxidises the exposed silicon to a thin dioxide layer, which the hydrofluoric acid dissolves; because formation and dissolution occur preferentially at protruding regions, the texture evolves toward rounded, symmetric pyramids with cleaner (111) facets.

Two things decide whether the rounding is a laboratory trick or a process step: it preserves the overall submicron pyramid geometry while modifying local curvature — the pyramids stay pyramids — and it is wet chemistry of exactly the kind silicon fabs already run, a texturing bath followed by an acid-based rounding bath. The fix rides on the industry’s existing texturing infrastructure instead of replacing it.

On rounded textures the a-Si:H layer is fully amorphous — diffuse diffraction halos where the irregular sample showed crystalline spots — and rounded bottom-cell precursors show markedly higher minority carrier lifetimes, signifying enhanced passivation.

The benefit carries one layer up. Steady-state photoluminescence on the perovskite top cells — a triple-cation composition with a 1.68 eV bandgap — shows the rounded sample with a higher quasi-Fermi level splitting of approximately 1.25 V, and spatial maps show a brighter, more uniform splitting across the film. Cross-sectional microscopy finds more frequent crack-like discontinuities in films on irregular textures, and a more continuous film on rounded ones, while X-ray diffraction shows no pronounced change in the bulk crystal structure. The improvement is a film-integrity and buried-interface effect, not a change in the perovskite itself.

The device numbers, handled in order

The devices are monolithic two-terminal tandems: a silicon heterojunction bottom cell on a 280 µm float-zone wafer, an ITO recombination layer, a 4PADCB hole-selective monolayer, the spin-coated perovskite, and C₆₀/tin-oxide/indium-zinc-oxide contacts with a magnesium fluoride coating.

Rounded-texture devices show an open-circuit voltage exceeding 1.96 V, approximately 20 mV higher than irregular-texture devices, and reproduced in a second, independently fabricated batch. Short-circuit current density is comparable between the two configurations — the rounded textures reflect slightly more light, but the light trap survives the rounding — and the rounded devices exhibit a higher fill factor, consistent with reduced morphology-related electrical losses. The champion rounded device reaches 33.25% on a 1.014 cm² cell.

Three qualifications travel with that number. First, the 33.25% is the team’s own laboratory measurement, not a certified figure. Second, it is a 1.014 cm² cell, not a module. Third, it is not a record — certified tandems have passed 35%. What the Jülich result offers is not a leaderboard position but a surface-shaping route that preserves light trapping while fixing the buried interface.

Why a surface treatment travels further than a new layer

The field has converged on the buried silicon surface as the place where tandem manufacturing will be won or lost. A Shanghai team reported in Advanced Functional Materials in April that chemical polishing of submicron textures lifted wide-bandgap tandems from 30.04% to 31.83%; a Zhejiang–Aiko team reported in Nature Communications in August that silica nanospheres filling the valleys of industrially textured silicon produced iceberg-like pyramids and a certified 33.15% on 1 cm²; a Matter paper this autumn selectively passivated only the pyramid peaks with aluminium oxide and reached 33.33%, certified at 32.89%.

The Jülich paper’s distinguishing claim is the mechanism: the polishing paper smooths grooves, the nanosphere paper fills valleys, the peak-passivation paper insulates the peaks — while the rounding paper shows, with atomic-resolution microscopy, why the surface matters at all. Micro-terraces template partial epitaxy in the amorphous silicon, and epitaxy degrades passivation. If partial epitaxy is a general failure mode of shallow-etched submicron textures, any group in that regime has a reason to check its own interface — and a known wet-chemical remedy to try. Rounding also subtracts sharpness from a surface the fab already etches — though no cost, throughput, or yield data accompany the paper, and the demonstration uses float-zone wafers and a spin-coated perovskite.

What this is not: a record, a module, or a production line

First, it is not a record: 33.25%, uncertified, sits below the certified tandem figures above 35% the paper itself cites, including the Soochow–LONGi team’s certified stabilised 35.13%. The value is the fix and the mechanism, not the leaderboard position.

Second, the demonstration is a 1.014 cm² laboratory cell on a float-zone wafer with a spin-coated perovskite top cell. Nothing in the material consulted addresses large-area uniformity, coating throughput, or yield, and the submicron texture here is not the micron-scale texture of standard industrial silicon. Whether ozone rounding transfers to micron pyramids, and whether it survives production deposition methods, are open questions.

Third, no stability result appears in the material consulted — no maximum-power-point tracking, damp-heat, or thermal-cycling data. The improved passivation is a reason to expect better durability, not evidence of it, and the gap stands out beside the Matter peak-passivation result, which pairs its 33.33% with 1,000 hours of tracked operation.

Fourth, the mechanism evidence comes from one laboratory, one texture regime, and one perovskite composition. The generality claim rests on a single materials system.

What to watch

The most informative follow-up would be the industrial texture: if ozone rounding works on standard micron-scale pyramids without giving up their light trapping, the idea becomes a general surface-preparation step. Second, stability data — passivation is usually where tandem degradation starts. Third, replication on a different stack: if a second group finds partial epitaxy at its own a-Si:H/c-Si interface and the same rounding step suppresses it, the micro-terrace diagnosis becomes a design rule rather than one team’s observation.

The bottom line

A Jülich-led team has shown that the tandem texture trade-off can be narrowed from the silicon side: etch controlled submicron pyramids, round their tips and micro-terraces with a three-minute ozone treatment, and the rounding suppresses partial epitaxial ordering at the interface, improves passivation and carrier lifetime, and gives the perovskite a more continuous film — without giving up the bottom cell’s light trapping. The tandem reaches 33.25% on a 1.014 cm² cell, with a voltage above 1.96 V and a higher fill factor than the unrounded control.

It is not a record, not a module, not a stability result, and not yet a production step. It is a surface-shaping route to a problem every tandem maker faces: how to keep the light trap while giving a sub-micrometre film somewhere smooth to land. The number to watch is not 33.25%. It is the first rounded texture on an industrial wafer — the day this stops being a laboratory result and becomes a step a fab can run after its existing texturing bath.

No investment recommendation is made or implied anywhere in this article.

Sources and further reading

  • Sun, N., et al. “Chemical Rounding Suppresses Morphology-Induced Interfacial Losses in Perovskite–Silicon Tandem Solar Cells.” Advanced Energy Materials (2026), published October 31, 2026. DOI: 10.1002/aenm.71784 — the primary source for all figures in this article: the trade-off framing, optical simulations, texture control (KOH 0.23% → 248 nm; 0.28% → 363 nm), rounding chemistry (HF, HCl, 20 ppm O₃, 3 min), microscopy (partial epitaxy on irregular textures; fully amorphous interlayer on rounded ones), perovskite composition (Cs₀.₀₅(FA₀.₇₇MA₀.₂₃)₀.₉₅Pb(I₀.₇₇Br₀.₂₃)₃, 1.68 eV) and quasi-Fermi level splitting (~1.25 V), device figures (Voc above 1.96 V; 33.25% on a 1.014 cm² cell), funding and conflicts statement (none declared). The full text is open access; no numeric champion fill factor or short-circuit current density is stated in the text consulted.
  • CiVL (November 1, 2026), “Rounded silicon texture helps stacked solar cells turn a third of sunlight into power” — independent summary corroborating the 33.25% efficiency, the 1.014 cm² cell area, the uncertified status, the submicron pyramids plus ozone rounding, and the team leadership.
  • Excitonic.com, “Twist the Molecule, Not the Process: Conjugation Control Solves the Tandem SAM Trade-Off” (November 10, 2026) — the paired tandem story: the Soochow–LONGi self-assembled monolayer, certified stabilised 35.13% on a 1 cm² cell, and the certified tandem figures above 35% that put the Jülich 33.25% in context.
  • Xia, S., Wang, F., et al. “Advanced Sub-Micron Pyramid Texture Enables Approximately 32% Efficiency in Perovskite/Silicon Tandem Solar Cells.” Advanced Functional Materials (2026), published April 14, 2026. DOI: 10.1002/adfm.202525903 — the chemical-polishing route on submicron textures, lifting wide-bandgap tandems from 30.04% to 31.83%.
  • Zhang, D., et al. “Iceberg-like pyramids in industrially textured silicon enabled 33% efficient perovskite-silicon tandem solar cells.” Nature Communications 16, 7331 (2025). DOI: 10.1038/s41467-025-62389-3 — the silica-nanosphere route on industrially textured silicon (pyramids above 2 µm), certified 33.15% on 1 cm².
  • Matter (2026), “Selective passivation of pyramid peaks for 32.9%-efficient perovskite/silicon tandem solar cells” — peak-selective Al₂O₃ passivation, 33.33% (certified 32.89%), 90% retention after 1,000 hours of tracked operation.

A note on sourcing: the full text of the Sun paper was read directly, and all device figures, texture parameters, and mechanism claims above are drawn from it or from the CiVL summary where noted. No numeric champion fill factor or short-circuit current density is quoted, because none is stated in the text consulted. No stability, large-area, cost, yield, or industrial-wafer result is stated, because none appears in the material consulted.