Héctor Vázquez presents Towards structure-conductance relationships using machine learning

Towards structure-conductance relationships using machine learning in atomistic
simulations of single molecule circuits

Single molecule circuits, where a small electrical current flows through
individual molecules bonded to two electrodes, represent the ultimate
downscaling limit of microelectronic devices. These nanoscale junctions are
fascinating systems where fundamental concepts in physics and chemistry can be
studied and developed.

In my talk, I will briefly introduce the basics of single molecule conductance
and discuss the potential of machine learning techniques for this field. I will
first show how conductance is determined not only by the nature of the
molecular
backbone and linker groups, but also by the details at the atomic level of the
metal-molecule interface. At room temperature, thermal fluctuations cause the
junction geometry and conductance to change continuously and scanning-probe
experiments aggregate all results in conductance histograms. Atomistic
simulations including quantum effects are ideally suited to model single
molecule circuits but are traditionally computationally expensive.

Next, I will examine recent efforts to apply machine learning techniques to
this
field [1-3], mostly focused on conductance experiments. I will present our new
method to compute conductance for thousands of geometries [4], opening the way
for a detailed analysis of the vast configuration space of junction geometry.
The combination of machine learning and physical-chemical intuition opens the
way for quantitative structure-conductance relationships in single molecule
junctions.


[1] H.J. Kulik and M.S. Sigman, Acc. Chem. Res.54 2335 (2021)
[2] M. Rupp, Int. J. Quantum Chem. 115 1058 (2015)
[3] D. Cabosart et al., Appl. Phys. Lett. 114 143102 (2019)
[4] H. Vázquez, J. Phys. Chem. Lett. 13 9326 (2022)