Abstract:
Chemical compositions of layered hybrid perovskites can be varied easily, and thereby, intrinsic properties like excitonic binding energy and bandgap can be varied over a wide range. We have prepared (PEA)2PbI4 (PEA: phenylethylammonium), (EA)2PbI4 (EA: ethanolammonium), and (CHA)2Pb(Br1–xIx)4 (CHA: cyclohexylammonium) exhibiting distinctly different optical properties. To estimate excitonic binding energy, photoluminescence (PL) spectra are recorded at various temperatures in the range of 300 to 5.6 K. However, the measurements take a long time. To reduce time and cost of experiments, and also avoid the possibility of sample degradation during the measurement, we have employed here a machine learning method, namely, a deep neural network (DNN). It generates hundreds of PL spectra at different temperatures by using only a few (5–7) experimental spectra. The DNN spectra are then used to estimate excitonic binding energies and other optical parameters. This demonstrated method reduces the data collection time from about 13 to 3 h. Furthermore, a DNN is used to generate optical absorption spectra of (CHA)2Pb(Br1–xIx)4 for a large number of compositions “x” by using experimental optical absorption spectra of only a few compositions as inputs. We envisage that the demonstrated method will be helpful for the search of materials with desired optoelectronic properties in the vast compositional space of halide perovskites.