July/August_AMP_Digital

A D V A N C E D M A T E R I A L S & P R O C E S S E S | J U L Y / A U G U S T 2 0 2 0 7 MACHINE LEARNING | AI SELF-DRIVING ROBOTIC LAB Ada is a self-driving robotic labora- tory that discovers new thin-film mate- rials. In 30 hours, it discovered a mate- rial with a high charge mobility partic- ularly well suited for organic solar cells. Ada can deposit, fabricate, and charac- terize a film in 20 minutes. After seven rounds, pipettes, glass substrates, and chemical supplies are replenished be- fore Ada proceeds with more experi- ments. The whole operation is driven by a powerful robotic arm with the as- sistance of a furnace, inbuilt spectrom- eters, a dark field microscope, and an instrument to measure electrical con- ductivity. Results of each experiment are fed into a machine learning model that suggests the experimental iteration needed to optimize a target property. Ada was developed by Alán Aspu- ru-Guzik, Curtis Berlinguette, and Jason Hein. “We launched Project Ada in July 2018. Within three months we had built the first version of the robotic platform. We continued to iterate and it took one year to go from project launch to man- uscript submission,” the authors said in a communication. The results, pub- lished in Science Advances, describe the optimization of hole mobility in spiro- OMeTAD thin films—a conducting layer used in or- ganic solar cells. Through 35 itera- tions, Ada demon- strated that dop- ing with 0.4 equiv- alents of cobalt and annealing for 75 seconds in the furnace produces the highest charge mobility in the se- lected parameter space. Only two parameters were chosen for optimi- zation in this ex- periment, though in principle Ada can handle multiple features and target properties. www.projectada.ca. AI ALGORITHM ANALYZES MATERIALS Scientists at the Center for Na- noscale Materials (CNM) at the DOE’s Argonne National Laboratory, Lemont, Ill., invented a machine learning algo- rithm for characterizing—in three di- mensions—materials with features as small as nanometers. Researchers can apply this discovery to the analy- sis of most structural ma- terials of interest to indus- try. “For example, with data analyzed by our 3D tool,” explains postdoctoral re- searcher Henry Chan, “users can detect faults and cracks and potentially predict the lifetimes under different stresses and strains for all kindsof structuralmaterials.” In the past, scientists visualized 3D microstructur- al features within a materi- al by taking snapshots at the microscale of many 2D slic- es, processing the individ- ual slices, and then pasting them together to form a 3D picture. However, that process is inefficient and leads to loss of information. “The beau- ty of our machine learning algorithm is that it uses an unsupervised algorithm to handle the boundary problem and produce highly accurate results with high efficiency,” says Chan. “Coupled with down-sampling techniques, it only takes seconds to process large 3D sam- ples and obtain precise microstructural information that is robust and resilient to noise.” The team successfully tested the algorithm by comparison with data ob- tained from analyses of several differ- ent metals (aluminum, iron, silicon, and titanium) and soft materials (polymers and micelles). The new algorithm is not restricted to solids. The team has ex- tended it to include characterization of the distribution of molecular clusters in fluids with important energy, chemical, and biological applications. CNM re- searchers believe this machine learning tool will prove especially useful for fu- ture real-time analysis of data obtained from large materials characterization facilities such as the DOE’s Advanced Photon Source and other synchrotrons around the world. anl.gov/cnm. Ada, a self-driving robotic lab. Courtesy of Fraser Parlane, University of British Columbia. Characterization of 3Dmicrostructure enabled by machine learning. Courtesy of Argonne National Laboratory.