A simple optimization of a crystalline silicon solar cell is performed
using a one-dimensional finite element calculation, and a neural netw
ork is subsequently trained to reproduce the data. The quality of this
reproduction is discussed. Considering that translation of the origin
al problem to run on a trained neural network permits a reduction of c
alculation time by three to six orders of magnitude, the scope for con
tinued work on and extension of this method appears appealing.