Six architectures, one recombination chain: how silicon solar cells closed in on their physical ceiling.
Every silicon solar cell efficiency record of the past fifty years traces to the same question: where are photo-generated carriers being lost before they reach the external circuit? Al-BSF, PERC, PERT, TOPCon, HJT and back-contact aren't six unrelated inventions. Each one fixed the previous generation's dominant loss mechanism, and in doing so exposed the next one. Here's that chain end to end, how close it now sits to silicon's physical ceiling, and a correction we made to our own source material before publishing this.
Every generation fixed one leak and moved the next one somewhere else.
Silicon solar cells lose carriers to Shockley-Read-Hall (SRH) recombination at defects and interfaces, to Auger recombination in heavily doped or high-injection regions, and (negligibly, at 1-sun) to radiative recombination. Every architectural change in this chain is an attack on SRH recombination at one specific physical location in the cell. Suppress it there, and whatever the next-largest loss channel is takes over as the new bottleneck: that's the entire history, generation to generation.
Al-BSF (1970s-2012) drew current off a bare rear silicon-metal interface with a recombination velocity of 10,000-60,000 cm/s: current flowed, but voltage paid for it. PERC (2012-2022) replaced that full-area contact with a dielectric layer (Al2O3 + SiNx) and local laser-opened contacts, cutting rear recombination roughly tenfold: but its boron-doped p-type wafer forms boron-oxygen complexes under light, causing light-induced degradation. PERT and its n-type wafer eliminated that degradation mechanism entirely: no boron, no complexes: but screen-printed metal still touched silicon directly on both surfaces, and that contact recombination became the new ceiling. TOPCon (2017-present) answered that with an ultra-thin tunnel oxide plus doped polysilicon: electrons tunnel through, recombination at the interface drops close to zero: but the front surface stayed a conventional diffused junction, now responsible for roughly half the cell's total recombination. HJT (1992 patent, mainstream since 2020) replaced that diffused junction with a hydrogenated amorphous silicon heterojunction on both surfaces, decoupling passivation from doping entirely: the bottleneck that's left is optical, not electronic, parasitic absorption in the a-Si:H and ITO layers. Back-contact architectures answer that by moving every contact to the rear, eliminating front shadowing outright.
Voltage is the tellEvery fix shows up directly in open-circuit voltage.
Open-circuit voltage (Voc) is the most direct readout of recombination control: every reduction in recombination current density (J0) shows up as higher Voc, with no change to the silicon absorber itself. The climb from Al-BSF to back-contact is roughly 0.55 W of additional output per cell from voltage alone.
Chart data
| Series | Lower bound of typical range | Low | Typical Voc range base | High |
|---|---|---|---|---|
| Al-BSF | 640mV | 640mV | 650mV | 660mV |
| PERT | 680mV | 680mV | 690mV | 700mV |
| PERC | 685mV | 685mV | 692mV | 700mV |
| TOPCon | 720mV | 720mV | 732mV | 745mV |
| HJT | 740mV | 740mV | 750mV | 760mV |
| Back-contact (IBC-SHJ) | 760mV | 760mV | 765mV | 770mV |
Our own source material was already a year out of date.
The technical reference this piece draws on cited 27.81% as the current back-contact silicon record: LONGi's HIBC cell, which it attributed to NREL certification in April 2025. Checking that figure before publishing this turned up two problems. First, that cell was certified by ISFH (Germany), not NREL: NREL's April 2025 certification that month was for a different LONGi result, the 34.85% perovskite-silicon tandem cell. Second, and more materially, 27.81% is no longer the record. LONGi pushed it to 28.04% in January 2026, then to 28.13% (ISFH-certified) in April 2026: the same figure already published on our own technology frontier page. A document dated the same month as this one was already citing a superseded number.
The same pattern holds elsewhere in the chain. JinkoSolar's TOPCon record (27.79%) is genuine but was ISFH-certified, not TÜV SÜD as commonly cited. And the tandem frontier is now a near-tie: LONGi's 34.85% (NREL-certified, April 2025) is still the higher figure, but JinkoSolar certified 34.82% in June 2026 on a TOPCon-based tandem platform, less than a year later and 0.03 points behind. Neither of those changes what this piece argues: it changes which number you'd be wrong to quote six months from now.
How close is closeSilicon is within 1.3 points of a wall physics puts there, not engineering.
The Shockley-Queisser detailed-balance limit for a single-junction silicon cell is 33.5%: radiative recombination only, under the standard 1-sun AM1.5G spectrum. Real silicon can't reach that, because Auger recombination is intrinsic to the material, not an engineering defect: correcting for it (Richter/Hermle/Glunz parameterisation) puts the real single-junction ceiling at 29.43%. The corrected current record, 28.13%, sits 1.3 percentage points below that: not 1.6, which is what you'd calculate from the superseded 27.81% figure.
Once a technology's remaining loss mechanism is Auger recombination in the silicon bulk rather than surface or contact recombination, no further contact innovation or passivation layer can move it: that's the defining boundary of the silicon single-junction era. Back-contact architectures are close enough to that boundary that the next real efficiency gain has to come from somewhere else entirely: a wider-bandgap top cell that captures the high-energy photons silicon currently wastes as heat. That's not a preference. It's why perovskite-silicon tandem cells already exceed the single-junction Shockley-Queisser limit (34.85%, more than a full point past 33.5%) using a completely different physical mechanism, and why the theoretical two-terminal tandem limit sits at 43.2%: an Auger-limited silicon bottom cell paired with a Shockley-Queisser-limited perovskite top cell.
What this means for the products on this benchmarkThe cell architecture field is a scored spec, not a marketing term.
Every panel on review.solar carries a sourced cell architecture spec: TOPCon, HJT, back-contact, or one of the older generations. That field is what this chain explains. A TOPCon panel's efficiency ceiling is structurally different from a back-contact panel's, for the specific physical reasons above, not because one manufacturer tries harder than another. Full generation-by-generation detail, including the exact passivation chemistry and remaining bottleneck for each architecture, sits in our technology frontier reference.
Full technology frontier reference: review.solar/roadmap.
Sources8 references
- Bonilla, Hoex, Hamer & Wilshaw: Dielectric surface passivation for silicon solar cells: a review (2017)
- NREL: Solar Cell Efficiency Tables, Version 66
- Richter, Hermle & Glunz: Reassessment of the limiting efficiency for crystalline silicon solar cells (2013)
- pv magazine: LONGi announces world-record 28.13% back-contact cell efficiency (Apr 2026)
- PV Tech: LONGi develops 28.13% efficient cell, 26.4% efficient module (Apr 2026)
- pv magazine: JinkoSolar achieves world-record 27.79% TOPCon cell efficiency (Nov 2025)
- pv magazine: LONGi achieves 34.85% two-terminal tandem perovskite cell (Apr 2025)
- pv magazine USA: JinkoSolar achieves 34.82% perovskite-silicon tandem cell (Jun 2026)
Read our ownership, funding and editorial disclosures.
Stay across the benchmark.
New scorecards, score changes and market signals, published as they happen.