All-small-molecule organic photovoltaic (OPV) cells based upon the small-molecule donor, DRCN5T, and nonfullerene acceptors, ITIC, IT-M, and IT-4F, were optimized using Design of Experiments (DOE) and machine learning (ML) approaches. This combination enables rational sampling of large parameter spaces in a sparse but mathematically deliberate fashion and promises economies of precious resources and time. This work focused upon the optimization of the core layer of the OPV device, the bulk heterojunction (BHJ). Many experimental processing parameters play critical roles in the overall efficiency of a given device and are often correlated and thus are difficult to parse individually. DOE was applied to the (i) solution concentration of the donor and acceptor ink used for spin-coating, (ii) the donor fraction, (iii) the temperature, and (iv) duration of the annealing of these films. The ML-based approach was then used to derive maps of the power conversion efficiencies (PCE) landscape for the first and second rounds of optimization to be used as guides to determine the optimal values of experimental processing parameters with respect to PCE. This work shows that with little knowledge of a potential combination of components for a given BHJ, a large parameter space can be effectively screened and investigated to rapidly determine its potential for high-efficiency OPVs.
Keywords: bulk heterojunction; design of experiments; machine learning, optimization; organic photovoltaics; small molecules.