Machine Learning‐Enabled Correlation and Modeling of Multimodal Response of Thin Film to Environment on Macro and Nanoscale Using “Lab‐on‐a‐Crystal”

Abstract

To close the feedback loop between artificial intellegence‐controlled materials synthesis and characterization, material functionality must be rapidly tested. A platform for high‐throughput multifunctional materials characterization is developed using a quartz crystal microbalance with auxiliary in‐plane electrodes and a custom gas/vapor flow cell, enabling simultaneous scanning probe microscopy and electrical, optical, gravimetric, and viscoelastic characterization on the same film under controlled environment. The lab‐on‐a‐crystal in situ multifunctional output allows direct correlations between the gravimetric/viscoelastic, electrical, and optical responses of polymer film in response to environment. When multiple film properties are used to augment the training set for machine learning regression, prediction of material response to the environment improves by a factor of 13 when <5% of the total dataset is used for model training.

SMART transfer method to directly compare the mechanical response of water-supported and free-standing ultrathin polymeric films

“…To ascertain the presence of water in these hydrophobic films (PS and P3HT), we utilized QCM and NR. QCM is a technique that can provide a qualitative understanding of both mass uptake and energy dissipation (change in stiffness) throughout the depth of a film with particular sensitivity towards the interface 53 , 54 . This is accomplished through tracking the response of multiple crystal harmonics ( n ), whereby low n harmonics correspond to regions closer to the film–water interface of interest and higher n harmonics probe the film–substrate interface.…”

“…Gravimetric/viscoelastic response of films was measured, while the film was transferred from air and submerged in Milli-Q water, and then transferred back to air. Values of Δ f and Δ D were calculated by fitting QCM conductance peaks to a Lorentz distribution and extracting position and width 54 , 58 .…”

Fabrication and Nanoscale Properties of PEDOT:PSS Conducting Polymer Nanospheres

“…These results confirmed the previous observations by PF-QNM using Young's modulus maps. Considering that PEDOT has a lower Y compared to PSS [54,60], the quantified softening of the nanospheres could be related to a different distribution of PEDOT and PSS in the resulting nanospheres after reprecipitation. This idea is in line with a possible re-arrangement of the PEDOT and PSS phases in confined geometries, as introduced in the last paragraph.…”

Integrating multiple materials science projects in a single neural network

“…To achieve higher conductivity, a number of new annealing methods, including repeated chemical treatments with strong acids, bases, and solvents, have been reported 23 . Informatic approaches have been partially introduced to predict the properties of PEDOT-PSS 24,25 . However, the post-treatment methods have become too long and complex to be analyzed by conventional machine learning approaches or to be understood except by a few specialists (example scheme is shown in Fig.…”