As we look towards the future, we can see that we are on the cusp of a new paradigm that has the potential to revolutionize the way materials are discovered. This new paradigm is centered around the idea of a self-driving laboratory (SLD) that combines robotic platforms and machine learning to achieve autonomous experimentation for energy materials. The potential implications of this technology are vast, with applications in fields such as agriculture, synthetic biology, and climate.
We envision that the future of material discovery platforms will be built upon five key components: AI models, robotic platforms, orchestration software, storage databases, and human intuition. These components can work together in a closed-loop approach, replacing the traditional paradigm of design, synthesis, characterization, and testing.
In the past, there have been many attempts to create autonomous experimentation systems, including robotic platforms, machine learning models, and integrated material discovery platforms. However, these systems have been limited by reaction type and experimental conditions. We see the integration of AI as a key driver for overcoming these limitations.
One promising approach for the automatic generation of small molecules is the method developed in Burke's laboratory. This approach uses a combination of flow and batch setups to perform C-C couplings using a three-step reaction, with an in-line catch-and-release purification protocol for the molecular building block. Another approach uses plug-and-play modular design and miniaturized microfluidic synthetic devices to generate small molecules through diverse chemical transformations. While these systems are efficient, they can be expensive in terms of time and resources. By integrating AI, we can enhance these platforms to accelerate the pace of materials discovery and reduce its cost.
AI technologies will play a crucial role in material discovery thanks to their ability to improve automatically through data. Two areas of particular interest are inverse design and active learning. Robotic platforms can provide high volume, control, and precision over experimental processes, generating high-quality data for use in these AI-driven approaches. For example, active learning can be used to optimize high-dimensional parameter spaces, such as reaction conditions and material properties. Inverse design, on the other hand, can generate candidate materials with desired properties, complementing the "idea generation" component of material discovery platforms. Generative models, such as variational autoencoders, generative adversarial networks, and hybrid methods, offer an efficient way to explore chemical space, estimated at around 10^60 molecules.
Finally, a material discovery engine requires the integration and management of robotic platforms, machine learning models, and human users. Orchestration software, such as ChemOS, can be used to provide accessible communication between components of a material discovery platform, enabling efficient experiment planning.
An example of the SLD in action is the optimization of the active layers in organic solar cells. Using an SLD driven by ChemOS, new opportunities for performance and photo-stability can be discovered based on all the measurements collected before. With the help of a machine learning decision maker, the SLD can infer the best material properties in real-time, resulting in more efficient and cost-effective experimentation. This is just the beginning of what is possible with the integration of AI and automation in materials discovery. We’re excited to fund new technologies in which materials can be discovered at an unprecedented rate and truly reimagine the world of matters.