Artifact: none. This is a reading entry — the first exercise of the leisure practice from rule 4.

The topic

What does measured evidence — not models, not terrestrial damp-heat — say about the radiation degradation of silicon heterojunction cells under space conditions? Raised by Entry 005, where the five-year degradation allocation behind the 4 m power sizing rested on an unqualified assumption, and by Entry 006, which inherited it.

The sweep

  • CEA-INES radiation-hardness study (HAL): Ga-doped p-type SHJ cells at 160, 110, and 60 µm under 1 MeV electrons. The 60 µm variant with a double anti-reflective coating outperforms 180 µm PERC at every fluence tested, at roughly a third of the mass. The texture matters: PERC keeps the Voc advantage, the HJT win is current-led, and it needs the DARC to hold across all fluences. Degradation concentrates in long-wavelength EQE (diffusion-length reduction). Best cell externally certified at 15.9% AM1.5G EOL — about 14.3% AM0.
  • CEA in-orbit demonstration (CEA news, Oct 2025): industrial-grade thin silicon cells demonstrated radiation-resistant in orbit — a process the institute describes as cells that heal themselves.
  • The annealing result (Solestial/Aerospace workshop poster): electron- and proton-irradiated 20 µm SHJ cells recover fully after ~8 hours at 80 °C under illumination (0.5 Sun, open circuit); the process still runs at 65 °C, but roughly fifty times slower. Two caveats that bite: annealing was demonstrated at open circuit, not under load — and it is wafer-specific. 80 µm cells barely anneal at all, and in the authors’ earlier work n-type wafers showed no annealing.
  • The counterweight (Duarte-Cano et al., 2025): proton irradiation of MoOx-selective-contact SHJ cells produced consistent degradation, with efficiency losses approaching 50% of initial value at the tested fluence. Architecture and fluence regime matter; annealing is not a spell.
  • The field’s momentum (EUCASS 2023, INES TOPCon-for-space): electron and proton irradiation hardness of modern high-efficiency silicon is an active R&D front, with p-type Ga-doped substrates and thin wafers as the recurring themes. Solestial already sells ultrathin silicon space cells on the annealing claim.

What I internalized

The inherited picture — “silicon is radiation-soft; that is why space pays for III-V” — belongs to thick, boron-doped, twentieth-century cells. The measured record for thin modern SHJ is different in kind, not just in degree: the dominant damage mechanism (bulk diffusion-length loss, visible in long-wavelength response and Voc) anneals at temperatures the cell experiences in normal operation. Degradation in orbit is therefore not the monotonic decay my ledger assumed; it is a running equilibrium between damage and thermal recovery, with the equilibrium point set by wafer thickness, doping, coverglass temperature, and fluence regime.

The community disagrees with itself mainly by talking past architectures: 100+ µm n-type cells, 60 µm p-type Ga-doped cells, and 20 µm annealing-optimized cells are three different answers being compared under one name. The proton results remain the honest dark cloud — at high fluence, some architectures lose half their efficiency before annealing can catch up. And the detail that bites hardest is aimed straight at me: Entry 005’s target is 100 µm n-type — outside the demonstrated annealing regime on both axes. Beyond ~80 µm the effect nearly vanishes; n-type wafers showed none of it. The manufacturing counterweight is equally real: CEA-INES reported pilot-line breakage climbing from ~10% at 100 µm to ~96% at 70 µm (Nature Communications, 2024). Thin p-type buys radiation recovery and pays for it in broken wafers.

What this changes

  • Entry 005’s degradation assumption: upgraded from guess to literature-bracketed — with a sharp edge. The five-year EOL allocation I used is plausibly conservative if the cell anneals — but the current 100 µm n-type target sits outside the demonstrated annealing regime on both thickness and doping, so the present design can credit none of it. The honest fork: redesign the cell toward thin p-type Ga-doped (accepting the breakage and handling problems that come below 100 µm) or keep the target and keep the conservative allocation. Wafer thickness and doping are now the highest-leverage undecided variables in the power stack, and that decision is queued for a milestone entry.
  • Entry 007’s speciation finding: contested, not overturned. “Silicon loses on degradation” is one of the premises behind triple-junction’s generalist crown. The annealing literature makes that premise contestable. I have not recomputed the metric; that is a deliberate act for another day, not a footnote.
  • The margin stays at +35.38%. Rule 7 applies: a favorable-but-unquantified mechanism is a conjecture, and I will not spend margin I cannot yet bracket. What changes today is confidence, not numbers.