Sorting out a Process-Structure-Property Relationship in Polymer Organic Electronics

The world is full of data, but automated extraction of knowledge from this data is often difficult. In this talk the focus is on processing of polymer organic electronic devices. Two challenges that will be discussed are: how to extract meaningful and quantitative structure-property relationships from high-dimensional AFM data and how to learn process-property relationships from a literature database. Both challenges will be discussed using poly-3-hexylthiophene (P3HT) as a case study.

A semi-automated approach to AFM structure quantification will be presented. This method enabled the assignment of process-property variability in spin-coated P3HT organic field effect transistors (OFETs) to the process-structure relationship. More importantly, the structure-property relationship was elucidated and had less variability. Meso-scale orientational order of P3HT nanofibers was shown to be a quantitative predictor of hole mobility in P3HT OFETs, arising from a combination of solution aging and the spin-coating flow field.

A tremendous amount of experimental literature exists on P3HT processing. However, many challenges remain in comparing “apples to apples.” The property of interest, hole mobility, is influenced by starting material properties, solution preparation, deposition method, and environmental factors during characterization. These experimental features are inconsistently reported by researchers, creating a sparse matrix of process-property data in the literature. A sortable database was created to peruse the literature and compare devices across studies.

Ultimately we determined that the best use of the literature database may be to generate new hypotheses for additional experimental studies. These hypotheses can be encoded in mathematical models, which are then used in an experimental design algorithm to calculate the “best” next experiments to perform. Once the data is obtained, it can be used to update the models, and repeat the cycle. Tailored methods for sequential statistical design will be presented here to formalize this linkage between models and experiments.