Object-Oriented Software Architecture (control-lab-ly) to Control Automated Workflows—A Case Study of Quantum Dot Ligand Cocktails

Given the multitude of ways to produce new materials, such as altering composition and processing parameters, the state space in materials research is vast, and AI algorithms are vital in exploring the space and reducing discovery time. To accelerate the collection of copious amounts of experimental data required for these algorithms, many research groups are increasingly turning to automated and autonomous workflows.<br/><br/>Within the community, significant strides have been made in creating flexible and reconfigurable setups by using modular hardware. However, less thought is put into the software that controls these workflows, which are usually custom setups. Bespoke software is typically written to orchestrate actions and communicate between components in the setup. Once working, code is commonly left untouched, which does not leverage on the flexibility and reconfigurability afforded by having modular hardware. This has two complications. First, code is not easily reusable. For a group setting up a new workflow, it is difficult to adapt code from a previous project or from another group. Second, changing a component in the workflow requires substantial rework of the code, since abstractions of how different components interact have not been established. These serve as barriers to teams that intend to set up a new automated workflow or update its components during the course of a campaign.<br/><br/>In this work, we propose a modular software architecture that controls automated experimentation setups, and we use it in our study to improve the charge mobility of quantum dot (QD) thin films by investigating the combinatorial space of multi-ligand passivation of QDs. Effective passivation of QDs can be achieved by matching the appropriate ligand type with each QD facet and we examine combinations of up to five different ligands. Fabricating spin-coated thin films was the major bottleneck in our workflow, as making just one sample already takes multiple spin steps, each having considerable wait times. Evolution of our thin-film fabrication tools was enabled by the proposed software architecture, implemented as a Python package. The package (<i>control-lab-ly</i>) exposes available actions for each tool in a standard manner for its particular category. For example, tools with the same function of moving items around a workspace are given identical methods, regardless of form (e.g. gantry vs articulated robots). The modular nature of the package also allows for extension via user plugins, so drivers for new equipment can easily be written and integrated.