Machine learning applied to molecular electronic devices discovery

Alexandre Giguère, Rodrigo Bogossian Wang

Molecular electronic devices are extensively studied to explore their potential use in electronics and to gain an understanding of electron transport mechanisms at the molecular scale. The source-sink potential (SSP) method [1,2] provides a simple tool for the qualitative analysis of the conductance of molecular electronic devices. The COMPAS 1D database is extended to include the molecular conductance of every planar hydrocarbon and every contacts combination possible [3]. This extended database is then used to train a generative machine learning model inspired by the GaUDI model [4].

This model will then be used to predict molecular structures with the desired conductive properties and can contribute to the development of innovative materials in molecular electronics.

1. Goyer, F., Ernzerhof, M., J. Chem. Phys. 2011, 134, 174101.
2. Ernzerhof, M., J. Chem. Phys. 2007, 127, 204709.
3. Wahab, A.; Pfuderer, L.; Paenurk, E.; Gershoni-Poranne, R., J. Chem. Inf. Model. 2022, 62 (16), 3704.
4. Weiss, T.; Mayo Yanes, E.; Chakraborty, S.; Cosmo, L.; Bronstein, A. M.; Gershoni-Poranne, R., Nat Comput Sci  2023, 3, 873–882.

 Alexandre Giguère