MSGID: 1:317/3 6279eae8   
   PID: hpt/lnx 1.9.0-cur 2019-01-08   
   TID: hpt/lnx 1.9.0-cur 2019-01-08   
    'Self-driving' microscopes discover shortcuts to new materials    
      
     Date:   
         May 9, 2022   
     Source:   
         DOE/Oak Ridge National Laboratory   
     Summary:   
         Researchers are teaching microscopes to drive discoveries with an   
         intuitive algorithm that could guide breakthroughs in new materials   
         for energy technologies, sensing and computing.   
      
      
      
   FULL STORY   
   ==========================================================================   
   Researchers at the Department of Energy's Oak Ridge National Laboratory   
   are teaching microscopes to drive discoveries with an intuitive algorithm,   
   developed at the lab's Center for Nanophase Materials Sciences, that   
   could guide breakthroughs in new materials for energy technologies,   
   sensing and computing.   
      
      
   ==========================================================================   
   "There are so many potential materials, some of which we cannot study   
   at all with conventional tools, that need more efficient and systematic   
   approaches to design and synthesize," said Maxim Ziatdinov of ORNL's   
   Computational Sciences and Engineering Division and the CNMS. "We can   
   use smart automation to access unexplored materials as well as create   
   a shareable, reproducible path to discoveries that have not previously   
   been possible."  The approach, published in Nature Machine Intelligence,   
   combines physics and machine learning to automate microscopy experiments   
   designed to study materials' functional properties at the nanoscale.   
      
   Functional materials are responsive to stimuli such as heat or electricity   
   and are engineered to support both everyday and emerging technologies,   
   ranging from computers and solar cells to artificial muscles and   
   shape-memory materials.   
      
   Their unique properties are tied to atomic structures and microstructures   
   that can be observed with advanced microscopy. However, the challenge   
   has been to develop efficient ways to locate regions of interest where   
   these properties emerge and can be investigated.   
      
   Scanning probe microscopy is an essential tool for exploring the   
   structure- property relationships in functional materials. Instruments   
   scan the surface of materials with an atomically sharp probe to   
   map out the structure at the nanometer scale -- the length of one   
   billionth of a meter. They can also detect responses to a range of   
   stimuli, providing insights into fundamental mechanisms of polarization   
   switching, electrochemical reactivity, plastic deformation or quantum   
   phenomena. Today's microscopes can perform a point-by-point scan of a   
   nanometer square grid, but the process can be painstakingly slow, with   
   measurements collected over days for a single material.   
      
   "The interesting physical phenomena are often only manifested in a small   
   number of spatial locations and tied to specific but unknown structural   
   elements.   
      
   While we typically have an idea of what will be the characteristic   
   features of physical phenomena we aim to discover, pinpointing these   
   regions of interest efficiently is a major bottleneck," said former ORNL   
   CNMS scientist and lead author Sergei Kalinin, now at the University   
   of Tennessee, Knoxville. "Our goal is to teach microscopes to seek   
   regions with interesting physics actively and in a manner much more   
   efficient than performing a grid search."  Scientists have turned to   
   machine learning and artificial intelligence to overcome this challenge,   
   but conventional algorithms require large, human-coded datasets and may   
   not save time in the end.   
      
      
      
   ==========================================================================   
   For a smarter approach to automation, the ORNL workflow incorporates   
   human- based physical reasoning into machine learning methods and uses   
   very small datasets -- images acquired from less than 1% of the sample   
   -- as a starting point. The algorithm selects points of interest based   
   on what it learns within the experiment and on knowledge from outside   
   the experiment.   
      
   As a proof of concept, a workflow was demonstrated using scanning   
   probe microscopy and applied to well-studied ferroelectric   
   materials. Ferroelectrics are functional materials with a reorientable   
   surface charge that can be leveraged for computing, actuation and sensing   
   applications. Scientists are interested in understanding the link between   
   the amount of energy or information these materials can store and the   
   local domain structure governing this property. The automated experiment   
   discovered the specific topological defects for which these parameters   
   are optimized.   
      
   "The takeaway is that the workflow was applied to material systems   
   familiar to the research community and made a fundamental finding,   
   something not previously known, very quickly -- in this case, within a   
   few hours," Ziatdinov said.   
      
   Results were faster -- by orders of magnitude -- than conventional   
   workflows and represent a new direction in smart automation.   
      
   "We wanted to move away from training computers exclusively on data   
   from previous experiments and instead teach computers how to think   
   like researchers and learn on the fly," said Ziatdinov. "Our approach   
   is inspired by human intuition and recognizes that many material   
   discoveries have been made through the trial and error of researchers   
   who rely on their expertise and experience to guess where to look."   
   ORNL's Yongtao Liu was responsible for the technical challenge of getting   
   the algorithm to run on an operational microscope at the CNMS. "This is   
   not an off- the-shelf capability, and a lot of work goes into connecting   
   the hardware and software," said Liu. "We focused on scanning probe   
   microscopy, but the setup can be applied to other experimental imaging   
   and spectroscopy approaches accessible to the broader user community."   
   The journal article is published as "Experimental discovery of structure-   
   property relationships in ferroelectric materials via active learning."   
   The work was supported by the CNMS, which is a DOE Office of Science   
   user facility, and the Center for 3D Ferroelectric Microelectronics,   
   which is an Energy Frontier Research Center led by Pennsylvania State   
   University and supported by the DOE Office of Science.   
      
      
   ==========================================================================   
   Story Source: Materials provided by   
   DOE/Oak_Ridge_National_Laboratory. Note: Content may be edited for style   
   and length.   
      
      
   ==========================================================================   
   Journal Reference:   
      1. Yongtao Liu, Kyle P. Kelley, Rama K. Vasudevan, Hiroshi Funakubo,   
      Maxim   
         A. Ziatdinov, Sergei V. Kalinin. Experimental discovery of   
         structure- property relationships in ferroelectric materials via   
         active learning.   
      
         Nature Machine Intelligence, 2022; 4 (4): 341 DOI:   
         10.1038/s42256-022- 00460-0   
   ==========================================================================   
      
   Link to news story:   
   https://www.sciencedaily.com/releases/2022/05/220509150750.htm   
      
   --- up 10 weeks, 10 hours, 51 minutes   
    * Origin: -=> Castle Rock BBS <=- Now Husky HPT Powered! (1:317/3)   
   SEEN-BY: 15/0 106/201 114/705 123/120 129/330 331 153/7715 218/700   
   SEEN-BY: 229/110 111 112 317 400 426 428 470 664 700 292/854 298/25   
   SEEN-BY: 305/3 317/3 320/219 396/45   
   PATH: 317/3 229/426