PINN-Fotovoltaica: code and data for "Physics-informed neural networks (PINNs) for drift–diffusion simulation of silicon solar cells: a current-based formulation and a comparison with finite volumes"

Code, reference data and results accompanying the article "Physics-informed neural networks (PINNs) for drift–diffusion simulation of silicon solar cells: a current-based formulation and a comparison with finite volumes" (A. J. D. Paxi Silva, V. M. Quilca Umasi, L. A. Montoya Portugal; LabERyA, Universidad Nacional de San Agustín de Arequipa; submitted). Contents: DEVSIM finite-volume reference solver for a 1D n+-p crystalline-silicon solar cell (SRH, Auger and radiative recombination, AM1.5G generation, finite surface recombination), with its verification: method of manufactured solutions (observed second-order convergence), Hovel analytical solution and grid convergence index. PINN formulations: equilibrium, dark and illuminated cells, ohmic and finite-S contacts, and parametric PINNs with two and five parameters, together with the data-driven baselines (splines, RBF, neural networks). Datasets: reference fields (ψ, n, p, J) and J–V curves, the two-parameter grids, and the five-parameter study (1367 DEVSIM runs, Sobol test set). Trained networks, metrics, loss histories and training logs (up to five seeds per experiment). The DEVSIM data can be reused as a benchmark for surrogate models and neural solvers of the same drift–diffusion model; see README.md for the benchmark tasks. Code: MIT licence. Data and results: CC BY 4.0.