Scientists from the School of Energy and Power Engineering, Chongqing University, China, have discovered a highly efficient, time saving as well as a reliable machine learning (ML) method for the research and development of novel organic photovoltaic (OPV) materials. During the development of high performing OPV materials, if one can pre-establish the correlation between the structure of the designed material and its photovoltaic property, it becomes highly meaningful and time saving. The research is reported in the journal Science Advances.
Image Credit: BeeBright / Shutterstock
OPV cells are an easy and highly economical method for transforming the solar energy into electrical energy. Until now, the typical OPV materials-based research has focused on building a relationship between the newly developed OPV molecular material and its organic photovoltaic material properties. Such an experimental method involves the design, synthesis and the material evaluation of newly developed OPV material. Further realizations involve the characterization of optoelectronic behavior of new material, and then the assembling and optimization of cells from these OPV materials.
This is a traditional method or approach and it needs laborious and highly controlled synthetic chemistry as well as the fabrication of OPV devices from the developed materials. Further, it demands very laborious and lengthy laboratory purifications of materials and multi-step experimental procedures, sufficient input of resources, and a much longer time period to complete the research projects.
Luckily, the scientists have now established a database composed of more than 1,700 donor functional materials from the literature and developed the ML approach. This machine learning method can help realize the structure-function relation among the materials, and hence promotes the faster materials screening for OPVs. The method explored many expressions related to structure of a molecule. Examples includes, ASCII strings, images, descriptors, and most importantly, fingerprints, which acts as inputs for ML algorithms. An important observation here was that the fingerprints with more than 1,000 bits of length could help establish great accuracy in prediction of material properties.
The reliability of this scientific approach had further been verified through screening of new designed donors. A very interesting experiment was performed for proving the efficiency and reliability of this ML approach. 10 materials were designed and used as donors. ML model-based predictions were compared with the experimental findings from these 10 designed materials. It was observed that the prediction of models was consistent with the experimentally obtained outcomes, only with very insignificant variations. Based on these results, it becomes highly obvious that ML acts as a very powerful method to screen various designed OPV materials, before their actual synthesis, thereby appreciably quickening the growth and development of OPV research area.
From a database composed of real donor molecules or polymers gathered from literature, multiple programming language dissemination of donor materials such as images, descriptors, ASCII strings and fingerprints from molecules were utilized to develop ML models for realization of high performing OPV materials, having high power conversion efficiencies (PCE). Molecular materials having fingerprints of length higher than 1000 bits yielded best programming language expressions for donor materials because they are easily accessible and because of their uniqueness. "The RF algorithm is found to be able to handle complex and long inputs, even in the presence of noise. This is because an RF chooses multiple features rather than the complete content of the input to establish the relationship", the authors added.
Thus, a method had been developed for donor material design for OPV by combined utilization of not only experimental but also the use of ML methods beforehand, thus saving both time and chemicals. A wide range of synthetic donor target materials, could easily be scrutinized through pre-determination and pre-classifications of their most expected material properties, utilizing this ML approach. The top-notch target materials, after ML identification, could then be synthesized in lab and additionally explored by experimentation.
‘’Our study on the relationship between the chemical structure of molecule and PCE of the molecule-based OPV could speed up new donor material design and hence accelerate the development of high PCE OPVs’’ authors added. Saving of resources and time, is one of the most important outcome of the ML method. ML acts as aid for guiding the experiments. A large number of novel molecular materials could be very much rapidly evaluated, which otherwise is not possible with laboratory based physical experiments or by ab-initio models. This can significantly cut off the needful material candidates to be synthesized for further laboratory studies. ML together in combination with experimental procedures, could initiate the ground-breaking molecular material discovery for semiconducting OPV materials, at a very fast pace.
Source: Machine learning–assisted molecular design and efficiency prediction for high-performance organic photovoltaic materials, Wenbo Sun, Yujie Zheng1, Ke Yang, Qi Zhang, Akee A. Shah, Zhou Wu, Yuyang Sun, Liang Feng, Dongyang Chen, Zeyun Xiao, Shirong Lu,Yong Li and Kuan Sun, Science Advances 08 Nov 2019: Vol. 5, no. 11, eaay4275, DOI: 10.1126/sciadv.aay4275, https://advances.sciencemag.org/content/5/11/eaay4275
Disclaimer: The views expressed here are those of the author expressed in their private capacity and do not necessarily represent the views of AZoM.com Limited T/A AZoNetwork the owner and operator of this website. This disclaimer forms part of the Terms and conditions of use of this website.