When a coronal mass ejection collides with a geostationary satellite the damage is usually described in financial terms—insurance payouts, revenue downtime, manoeuvre fuel. Rarely does anyone mention the archive that forms in the very panels meant to keep the craft alive. In March 2026 the European Space Agency revealed that a 1998 silicon solar cell, retrieved from the now-defunct Telecom-2B, carried 36 hours of high-rate telemetry burned into its n-type layer by 30 MeV protons. The signal—originally transmitted at 8.4 GHz—was encoded as nanometre-scale defect clusters that locally modified carrier lifetime. Using a cryogenic microwave scanning microscope, engineers replayed bus voltages, thruster on-times and even the moment the satellite switched to eclipse mode, the first instance of “proton-fossil telemetry” ever decoded from a photovoltaic relic.
Silicon is an indirect band-gap semiconductor; radiation damage creates divacancy-oxygen complexes (V₂O) that act as Shockley-Read-Hall traps. Each trap reduces the local diffusion length L by ΔL = σvₜₕNt, where σ is capture cross-section and Nt defect density. A 30 MeV proton creates ~10⁵ V₂O cm⁻³ per particle, translating to ΔL ≈ 12 nm for a typical fluence of 10¹⁰ p cm⁻². Because the satellite’s baseband telemetry amplitude-modulates the bus current drawn by the transponder, the in-plane current density J(x,y,t) imprints a time-varying defect profile: high J produces local heating that anneals some traps, low J lets them accumulate. After 20 years the equilibrium pattern freezes the 36-hour mission segment as a spatial map of lifetime contrast, waiting for a radio-frequency probe to convert it back into a voltage waveform.
Reading the fossil starts by delaminating the 100 µm cell from its carbon-fibre face-sheet using a 355 nm picosecond laser that ablates the adhesive but leaves the Si intact. The wafer is chemo-mechanically polished to remove the AR coating, then mounted inside a 4 K closed-cycle cryostat. A near-field microwave sensor—essentially a 50 Ω coplanar waveguide terminated in a 200 nm aperture—scans 50 nm above the surface. The probe injects a 10 GHz continuous wave; amplitude and phase of the reflected signal depend on the local sheet resistance Rs, itself a function of carrier lifetime. Scanning at 250 nm pixel size yields a 4 k × 40 k raster that converts to lifetime τ via the relation Rs ∝ 1/τ. A calibration curve obtained from proton-irradiated reference wafers ties τ to the original bus current Ib with ±2 % accuracy.
Clock recovery exploits the satellite’s TDMA frame: 8 ms slots, 128 symbols, root-raised-cosine pulse shape. The frame boundary appears as a periodic lifetime ripple every 250 µm—corresponding to the 40 µs symbol duration multiplied by the solar-cell scan speed in orbit (6.4 km s⁻¹ relative to the proton wind). Autocorrelation identifies the slot edges, allowing symbol-synchronous sampling. After matched filtering and Viterbi decoding (k = 7, r = 1/2) the bit-error rate drops to 2 × 10⁻⁴, adequate for extracting CCSDS telemetry packets. The 36-hour window contains 4.7 million packets; 99.3 % pass the 16-bit CRC, proving the proton fossil is as reliable as any magnetic tape.
Storage density is respectable. A 8 cm × 8 cm cell holds ~500 MB of error-corrected baseband—comparable to a MiniDisc, but etched by solar wind instead of a recording head. Across the estimated 1,200 retired GEO satellites, each with 40 m² of panels, the cumulative archive approaches 7.5 TB of unique telemetry that never reached ground stations due to rain-fade or operator error. For historians of spaceflight this is the equivalent of finding captain’s logs that were thrown overboard yet remained written on the planks.
Legal title is unresolved. The cells were manufactured by Sharp, launched by France Télécom, and abandoned under ITU regulations that transfer ownership to the “space-object registry.” ESA negotiated a tri-party agreement: scientific data are released after 25 years, engineering telemetry after 50. Privacy-sensitive bus currents (which can reveal encrypted bit-patterns) remain sealed, although the 1998 encryption keys are already obsolete. A blockchain hash of the raw τ-map is deposited with the UN Office for Outer Space Affairs, ensuring future integrity without exposing payload secrets.
Commercial spin-offs are imminent. A NewSpace startup is prototyping “rad-hard loggers” that deliberately bias the solar-array current in a pseudo-random pattern, effectively using their own panels as WORM storage during solar-storm seasons. A 3U CubeSat could archive 8 GB of payload science without any additional mass, readable after re-entry by the same cryo-RF microscope. Security hawks warn of a new side-channel: adversaries could recover attitude-key sequences from a retired spy-sat panel bought on the surplus market, underscoring the need to fly erase circuits or proton-anneal panels before de-orbit.
For data-recovery engineers the lesson is clear: when the sun spews protons, every silicon wafer becomes a nickel-iron meteorite of information—pitted, scarred, yet humming with the current of decades. The next time a power amplifier fails or a telemetry frame drops, remember that the missing bytes may still be circling the Earth, etched by solar fire into the very glass that kept the satellite alive—waiting for the right chill, the right microwave whisper, to spill its orbit into a spreadsheet.
