Solar absorbers live under a rule that reads like a law of nature. The more strongly a semiconductor drinks in sunlight, the faster the charges it creates tend to vanish again. Absorption and recombination are two faces of the same quantum-mechanical coin: a material whose electronic transitions readily absorb photons will, by the same logic, readily emit them back, letting electron and hole meet and cancel before either can be collected. A paper published October 28 in Nature argues that this trade-off is not inevitable — and it names a material that appears to break it.
The material is bismuth sulfide, Bi₂S₃: a non-toxic, earth-abundant semiconductor that this desk has watched mainly as a cousin of the bismuth-based absorbers — silver bismuth sulfide, sodium bismuth sulfide — pitched as lightweight, printable alternatives to silicon. The new result is not a better bismuth sulfide cell. There is no cell in the paper at all. What the Oxford-led team reports is a mechanism: a rare electronic structure, which they call pseudo-direct, in which sunlight is absorbed through a strong high-energy pathway while a lower-energy state just beneath it quietly holds the excited charges, slowing their loss. At room temperature, the team calculates, that arrangement suppresses the intrinsic bimolecular recombination rate by about 75% relative to a purely direct semiconductor that absorbs just as strongly.
The trade-off, stated precisely
The trade-off has a formal statement. The generalized Planck’s law, and the van Roosbroeck–Shockley relation built on it, ties a semiconductor’s absorption spectrum to its radiative recombination rate: absorption and emission are inverse processes running through the same electronic states. A direct-bandgap semiconductor — gallium arsenide and the lead halide perovskites are the canonical examples — absorbs strongly because its band-edge transitions are momentum-allowed, needing no phonon assist. But those same allowed transitions make electron–hole recombination fast. The paper quotes bimolecular recombination coefficients around 10⁻⁹ cm³ s⁻¹ for lead halide perovskites and around 10⁻¹⁰ cm³ s⁻¹ for gallium arsenide. An indirect semiconductor like silicon recombines slowly, because its band-edge transitions require a phonon to conserve momentum, but it absorbs weakly for the same reason and needs a thick wafer to catch the light.
Previous escape attempts have all paid a price. Natural photosynthesis, the paper notes, trades roughly 0.7 eV of photovoltage to stretch charge lifetimes by six orders of magnitude. Strategies that park charges in polaron or defect states in photoelectrochemical cells similarly disrupt charge transport, adding losses of their own. Rashba splitting — a spin–orbit effect proposed to give perovskites an indirect character — has, in the authors’ account, turned out to play a negligible role in bulk halide perovskites. The quest was for a way to delay recombination without sacrificing voltage and without wrecking transport. Bismuth sulfide, the authors argue, does it by accident of band structure.
The trapdoor: a direct transition with an indirect basement
Bismuth sulfide crystallizes in the orthorhombic stibnite structure, built from parallel quasi-one-dimensional (Bi₄S₆)ₙ ribbons held together by van der Waals forces. Its bandgap character has been debated for years, with studies claiming both direct and indirect behavior — a confusion the new paper explains rather than merely resolving. First-principles calculations from Professor Marina Filip’s group in Oxford’s Department of Physics show that several bands and valleys sit within a few tens of milli-electronvolts of each other near the band edges, producing a cluster of closely spaced direct and indirect transitions. Different measurements kept giving different answers because the material genuinely contains both.
The crucial geometry is this: the lowest-energy transition in Bi₂S₃ is indirect, but it sits less than 40 meV below the lowest direct transition — an energy difference comparable to thermal energy at room temperature (k₍B₎T ≈ 26 meV). Absorption is therefore dominated by the strong direct transitions, with an absorption coefficient above 10⁵ cm⁻¹ for visible light and an absorption onset around 1.4 eV, close to ideal for a single-junction absorber. But the charges, once created, rapidly relax toward the band edges and settle into thermal equilibrium across both sets of states — including the lower-lying indirect ones, which recombine only weakly because they are momentum-forbidden.
The experimental signature is a temperature dependence that runs backwards. In ordinary semiconductors, bimolecular recombination gets faster on cooling, because the narrowing Fermi–Dirac distribution concentrates carriers where they can recombine. In Bi₂S₃, the team found the opposite: recombination slowed on cooling, from room temperature down toward 4 K, as more carriers condensed into the low-lying indirect states and starved the fast channel. Fitting that trend gives ΔE ≈ 38 meV, matching the ≈40 meV from first-principles calculations.
The measurements are contact-free and therefore clean of device artifacts. The team excited ~150 nm Bi₂S₃ films — thermally evaporated onto quartz, annealed at 250 °C in nitrogen — with 35-femtosecond laser pulses at 1.55 eV, then tracked the surviving charge population with optical-pump terahertz-probe conductivity, supplemented by millimetre-wave conductivity and time-resolved photoluminescence, from 4 K to 295 K. Separating monomolecular (defect-mediated) from bimolecular (intrinsic) recombination by varying the excitation fluence, they extracted a room-temperature k₂ ≈ 5 × 10⁻¹¹ cm³ s⁻¹ — lower than gallium arsenide’s and the perovskites’, despite the higher absorption coefficient. Feeding the measured band structure into the van Roosbroeck–Shockley model then gives the headline number: at ΔE = 40 meV, the pseudo-direct configuration suppresses k₂ by approximately 75% at room temperature relative to a purely direct semiconductor.
What 75% is — and what it is not
Three qualifications should travel with that number wherever it goes.
First, it is a modelled suppression of the intrinsic rate constant, not a measured device improvement. The 75% comes from applying the van Roosbroeck–Shockley relation to the calculated band structure, validated against the temperature dependence of the measured k₂. It compares Bi₂S₃’s configuration against a hypothetical purely direct semiconductor with the same absorption strength. It is a statement about physics, not about a product.
Second, intrinsic recombination is not the only loss in these films, and on the timescales that matter for a device it may not even be the dominant one. The same terahertz transients show that, on nanosecond timescales, the charge population in these Bi₂S₃ films almost entirely recombines within about 100 ns through a defect-mediated, monomolecular channel. The pseudo-direct mechanism lowers the intrinsic floor; it does nothing, by itself, about the trap ceiling. Any real device must still solve the defect problem, and Bi₂S₃’s device history gives reason for caution: solution-deposited films have historically delivered poor solar conversion efficiencies, with sulfur vacancies identified in earlier work as deep trap states. A mechanism that lowers the intrinsic recombination rate only matters if the extrinsic rate can be brought down to meet it.
Third, no device exists to test the translation. There are no contacts, no transport layers, no current–voltage curves, and no efficiency — certified or otherwise — in this paper. The authors are explicit that the study demonstrates a physical mechanism rather than a finished solar cell, and that further work is needed to translate the findings into improved solar technologies. Anyone quoting a Bi₂S₃ efficiency alongside this result is importing a number from a different paper.
None of this diminishes the result; it locates it. The pseudo-direct idea matters as a mechanism unless and until a device proves otherwise.
Why the mechanism generalizes — on paper
The most consequential part of the paper may be its final section, where the authors turn from Bi₂S₃ to a search strategy. If the trick is an energy offset ΔE of roughly two to three times thermal energy between an indirect transition and a nearby direct one, then other materials should qualify, and the authors propose a two-phase screen: first, a coarse trawl of large materials databases — they name the Materials Project — for semiconductors and insulators with ΔE ≈ 2–3 k₍B₎T; then refinement using hybrid-functional bandgap calculations and sifting for the other traits a good absorber needs, including high absorption coefficients, low and isotropic effective masses, and favorable defect chemistries.
The model also names candidates already in the literature. In layered van der Waals materials, where interlayer coupling tunes the direct–indirect splitting, the authors’ calculations suggest that tungsten diselenide, with ΔE ≈ 60 meV, could show k₂ suppression around 88%, and indium selenide, with ΔE ≈ 50 meV, around 81%. And because Bi₂S₃’s ribbons are van der Waals bonded, the splitting itself may be tunable: the team’s calculations show that pulling the (Bi₄S₆)ₙ ribbons apart along the crystal’s a- and c-axes can push ΔE up to around 100 meV, via exfoliation, strain, or intercalation. Larger splittings could extend carrier lifetimes by up to two orders of magnitude, at the cost of a larger voltage sacrifice — roughly 100 meV in the paper’s example. The trade-off reappears, but now as a tunable dial rather than a fixed penalty.
There is a second application lane, and it may be the one where longer-lived charges matter most. In solar-to-chemical fuel conversion — photoelectrochemical cells that use sunlight to drive reactions such as water splitting — charges must survive long enough not just to reach an interface but to wait for slow, multi-electron chemistry. The paper’s introduction frames this explicitly: extended carrier lifetimes at semiconductor–electrolyte interfaces are vital for solar energy conversion to chemicals. A reservoir state that holds charges without immobilizing them is, in principle, well matched to chemistry that cannot hurry. That remains a prospect, not a demonstration: no catalytic turnover is measured here.
One reason Bi₂S₃ keeps reappearing in the absorber conversation is that it is easy to make and hard to object to: earth-abundant and non-toxic, unlike the lead and cadmium in incumbent thin-film absorbers, with films producible by accessible routes including low-cost solution processing, thermal evaporation — the route used here — and chemical bath deposition. But the manufacturing story cuts both ways: the same literature records Bi₂S₃’s device disappointments and the annealing tricks, such as sulfur-vapor annealing, needed to fill vacancies before carrier lifetimes become respectable. The process question is not whether Bi₂S₃ can be coated cheaply — it can — but whether it can be coated well.
What to watch
The first test is replication in a second material. The paper’s screening proposal is specific enough to be falsifiable: a database search for ΔE ≈ 2–3 k₍B₎T candidates, followed by the same temperature-dependent terahertz signature — recombination that slows on cooling — in a compound that is not Bi₂S₃. The tungsten diselenide and indium selenide predictions give the search a head start. A second pseudo-direct material would convert a curiosity of one band structure into a materials class.
The second test is a device — any device. A Bi₂S₃ solar cell, or a photoelectrochemical cell, that shows the pseudo-direct advantage surviving contact formation, transport layers, and interfaces would answer the question this paper cannot: whether a reservoir state that slows recombination in a bare film still slows it once charges must actually leave. Watch for the control comparison that matters: the same device stack with a direct absorber of comparable absorption, so the pseudo-direct gain can be isolated from ordinary process improvements.
The third test is defect control in Bi₂S₃ itself. A film in which the monomolecular, trap-mediated decay no longer swallows the population within ~100 ns — through sulfur-vapor annealing, stoichiometry control, or passivation — would reveal how much of the intrinsic advantage is reachable. Until that result appears, the 75% figure describes a floor being lowered in a room whose ceiling is still traps.
And the screening effort itself is worth tracking as a program rather than a single result. If the Materials Project trawl the authors propose yields candidates with the full package — pseudo-direct splitting, strong absorption, light effective masses, benign defects — then the paper’s lasting contribution will be a search filter, not a material.
The bottom line
An Oxford–Wuhan collaboration led by Professor Laura Herz, with electronic-structure calculations from Professor Marina Filip’s Oxford group, has shown that bismuth sulfide thin films combine strong direct absorption — above 10⁵ cm⁻¹, with an onset near 1.4 eV — with an indirect transition sitting less than 40 meV below the direct one. That small offset creates a reservoir state in which charges linger, and the team’s van Roosbroeck–Shockley modelling puts the resulting suppression of intrinsic bimolecular recombination at about 75% at room temperature versus a purely direct absorber of equal strength, a figure corroborated by the unusual observation that recombination in Bi₂S₃ slows on cooling.
It is a mechanism paper, not a device paper: no solar cell efficiency is claimed, no cell is built, and the films’ own defect-mediated losses still dominate on nanosecond timescales. Its value is the design principle it names — scan for, or engineer, the same trapdoor electronic structure, with an indirect–direct splitting of a few times thermal energy — and the two lanes it opens: absorbers that keep strong absorption while shedding part of the recombination penalty, and solar-fuel chemistry that gains time for slow reactions. The number to watch is not 75%. It is the second material: the first compound found by this logic, outside Bi₂S₃, that shows the same backwards temperature dependence and the same waiting room.
No investment recommendation is made or implied anywhere in this article.
Sources and further reading
- Herz, L. M., et al. “Delayed electron–hole recombination in a pseudo-direct semiconductor.” Nature (2026), published October 28, 2026. DOI: 10.1038/s41586-026-11223-x — the primary source for the pseudo-direct mechanism (indirect transition less than 40 meV below the direct transition; fitted ΔE ≈ 38 meV), the film preparation (thermal evaporation onto quartz, ~150 nm, annealed 30 min at 250 °C in nitrogen), the measurement approach (optical-pump terahertz-probe conductivity with millimetre-wave and photoluminescence probes, 4–295 K, 35-fs 1.55-eV excitation), the extracted room-temperature bimolecular coefficient (k₂ ≈ 5 × 10⁻¹¹ cm³ s⁻¹), the ~75% modelled suppression of k₂ at room temperature versus a purely direct semiconductor, the backwards temperature dependence, the nanosecond-timescale defect-mediated decay (population largely recombined within ~100 ns), the absorption coefficient (above 10⁵ cm⁻¹) and onset (~1.4 eV), and the screening and tuning proposals (Materials Project screen for ΔE ≈ 2–3 k₍B₎T; WSe₂ ≈ 60 meV → ~88% and InSe ≈ 50 meV → ~81% modelled suppression; ribbon separation pushing ΔE toward ~100 meV). Corresponding author Professor Laura Herz, Department of Physics, University of Oxford; calculations by Professor Marina Filip’s group; collaboration with Wuhan University. The article is open access.
- University of Oxford / TechXplore (October 28, 2026), “Semiconductors could sidestep a key solar-energy trade-off through a newly identified mechanism” — the announcement coverage carrying Herz’s quotes (“this trade-off is not inevitable”; “a roadmap for discovering and potentially engineering other materials”) and the reservoir framing, and confirming the October 28 Nature publication date.
- Interesting Engineering (October 28, 2026), “Oxford researchers identify new semiconductor for more efficient solar cells” — trade coverage confirming the non-toxic, earth-abundant framing, the solar-to-chemical fuel application, and the accessible processing routes (low-cost solution processing, thermal evaporation, chemical bath deposition); the study is described there as proving a mechanism rather than a commercial product.
- Huang, Y.-T., et al. “Enhancing the solar energy conversion efficiency of solution-deposited Bi₂S₃ thin films by annealing in sulfur vapor at elevated temperature.” Sustainable Energy & Fuels (2017). DOI: 10.1039/C7SE00398F — background on Bi₂S₃’s device history: a favorable ~1.3 eV bandgap, sulfur vacancies as deep trap states, and sulfur-vapor annealing raising carrier lifetimes from sub-picosecond to ~30 ps; cited for the caution that Bi₂S₃ devices have historically underperformed despite good absorption.
No market-size figure is quoted, because none is needed for this mechanism result, and no device efficiency, module result, or catalytic turnover figure is stated, because none appears in the sources consulted.