Predictive Modelling and Experimental Analysis of Radiation-Resistant MXene–Silicon Heterojunction Solar Cells

The MXene–Si heterojunction retains more than 84% of its initial PCE under simulated space irradiation, whereas conventional Si devices retain only about 55%. Prior MXene–Si work focused mainly on beginning-of-life efficiency (for example reported PCE above 17%); this work moves the metric to end-of-life retention at mission-relevant fluences and provides a bare-Si reference under identical conditions. Based on twelve 1 cm^2 devices fabricated in an ISO Class 5 cleanroom, measured under a calibrated AM 1.5G solar simulator with a Keithley 2400 source meter, with fluence verified by calibrated solid-state dosimeters (±5%) and ±4% model uncertainty.

Interface analysis shows the MXene layer remains continuous after irradiation, with atomic intermixing confined to an interfacial mixing zone of roughly 6–8 nm. Three-dimensional APT reconstructions plus elemental depth profiles (C, Si, O, P) are applied to the MXene–Si interface evolution, providing atomic-scale interfacial evidence that has been less commonly reported for this system. APT shows a sharp interface before irradiation and interfacial broadening after exposure above 10^14 particles/cm^2; XRD retains Si(111)/(220)/(311) and MXene(002) peaks, while Raman shows a D band near 1350 cm^-1 with reduced 2D intensity, indicating defects and lattice disorder.

A three-layer digital twin combining LAMMPS/OVITO atomistic simulation, physics-informed degradation equations and a Random Forest surrogate predicts VOC, JSC, FF and PCE with R^2 > 0.96. The authors state that no prior study has combined mission-relevant irradiation of MXene–Si cells with atomistic simulation and machine-learning lifetime prediction, which is the gap this work addresses. A synthetic degradation dataset of N = 5000 samples spans dose 10^4–10^7 Gy, time 0–200 h and temperature 31–49 °C, with an 80%/20% stratified split, 200 trees, maximum depth 12, minimum 2 samples per leaf, grid search under 5-fold cross-validation, and external validation against measured post-irradiation device parameters.

Degradation differs by parameter: VOC declines gradually and approximately linearly with dose (starting near 0.65 V), JSC is most dose-sensitive and decays exponentially (about 36 mA/cm^2 unirradiated, falling below 10 mA/cm^2 at the highest doses), and FF and PCE degrade faster with dose and time and are temperature-sensitive. VOC, JSC, FF and PCE are each modelled as functions of dose, time and temperature, with a parallel-coordinate visualization to identify safe operating envelopes and critical degradation thresholds. Based on the physics-informed degradation equations (1)–(4), with coefficients calibrated from atomistic simulations and experimental irradiation data and error bars indicating ±4% model uncertainty; the modelled PCE range spans 0.0%–21.1%.

The results apply to vertical heterojunction devices using n-type monocrystalline silicon as absorber and Ti3C2Tx MXene as transparent conductive front electrode and interface modifier, evaluated under simulated near-vacuum (about 10^-5 Torr) 50 MeV proton and 1 MeV electron irradiation at fluences of 10^12–10^15 particles/cm^2, irradiation temperature 20–25 °C and AM 1.5G measurement; the authors point future work toward wafer-scale MXene deposition, flight-telemetry validation of the digital twin, broader proton–electron–thermal degradation spectra and MXene–Si tandem architectures.

A careful reader would still watch: the digital twin is trained on synthetic data generated from calibrated degradation equations (N = 5000), so its extrapolation capability awaits flight-telemetry validation; MXene termination oxidation and restacking under ambient humidity leave long-term stability and encapsulation as open questions; how uniformity and Schottky-barrier quality scale from 1 cm^2 devices to wafer-scale areas remains to be verified; and because this is a fast parse of the text, the explicit forms of Eqs. (1)–(4) and some figure details are not fully reproduced here, so reproducing the coefficient calibration would require consulting the original figures and tables.