UC Berkeley

Diffusion is a ubiquitous phenomenon encompassing a wide range of physical systems from atomic scale motion to cellular locomotion. Diffusion plays a central role in practical applications such as electrochemistry, batteries, and thin film growth. The external control of diffusion properties presents new possibilities for understanding fundamental physical phenomena as well as a basis for next generation technological advances. The contributions of this thesis include the fabrication, modeling, and characterization of diffusion on gate-tunable nanodevices. We first derive the fundamental results of diffusion which form the basis of our experimental analysis. Afterwards, we describe the procedures for fabrication and algorithms for characterization of diffusion in video scanning tunneling microscopy (STM). The physical platform upon which most of these results are based on is a single-layer graphene on hexagonal boron nitride (h-BN) field-effect transistor (FET) with fluorinated tetracyanoquinodimethane (F4TCNQ) adsorbates, which were chosen due to favorable gate-tunable electronic properties. This device architecture enables control of diffusion through external temperature control, application of lateral and vertical electric fields, and substrate engineering. The analysis presented in this work provides rich sandbox for locally probing diffusion behavior in different system regimes. In particular, we find that device level control allows us to tune the density of adsorbates on the device due to the proximity of molecular orbitals to the device Fermi level and the device classical capacitance. Charged molecules form uniform arrays in response to the gate voltage to screen the applied field. In addition, we see that gate voltage can directly control diffusion through the modification of transition state energies as a result of molecule charging. We show that substrate level control in the form of moiré superlattice engineering can suppress diffusion by inducing anomalous subdiffusion in short time scales. Such behavior is explained using thermal equilibrium statistics for diffusion which show that the introduction of the moiré superlattice increases the number of unique energy states available to the diffusing particle. As a result, the size of the state space increases which in turn increases the time required to visit all available states. For short times, the entirety of the state space is not visited with high likelihood, a violation of the thermodynamic hypothesis that all accessible states are equiprobable, a phenomenon known as ergodicity breaking. Despite the challenges inherent in uncovering this behavior under experimental constraints of data sparsity using canonical statistical methods, we demonstrate that it is possible to observe this phenomenon by using deep learning.