Particle physics laboratory

Researchers at Oak Ridge National Laboratory have developed a DKL approach via stand-alone microscopes that enable physical discovery in the automated experiment

This Article Is Based On The Research Paper 'Experimental discovery of structure-property relationships in ferroelectric materials
via active learning'. All Credit For This Research Goes To The Researchers 👏👏👏

Please Don't Forget To Join Our ML Subreddit

Microscopes are taught to drive discoveries using an intuitive algorithm that could lead to new materials for energy technologies, sensing and computing. Many possible materials cannot be examined with traditional methods. Intelligent automation can help acquire new materials and create a repeatable path to unreachable discoveries.

The method combines physics and machine learning to automate microscopy studies to determine the functional characteristics of materials at the nanoscale. From computers and solar cells to artificial muscles and shape-memory polymers, active materials respond to stimuli such as electricity or heat and serve both familiar and new technologies. Their distinct characteristics relate to atomic structures and microstructures visible with sophisticated microscopy. However, to develop effective strategies to identify regions of interest where these traits appear and can be examined.

Scanning probe microscopy is useful for studying structure-property relationships in functional materials. The instruments use an atomically precise probe to scan the surface of materials to map structure at the nanometer scale, or one billionth of a meter. They can also detect responses to various stimuli, allowing researchers to learn about polarization switching, electrochemical reactivity, plastic deformation, and quantum events. Today’s microscopes can scan a square grid at the nanoscale point-by-point, but it takes days for a single substance to work.


Although one generally knows the physical processes that one wishes to study, the localization of these regions of interest is an important bottleneck. The goal is to educate microscopes to actively search for intriguing physical locations which is much more efficient than a grid search. Scientists have turned to machine learning and artificial intelligence to solve this problem, but traditional algorithms require massive human-coded datasets and may not save time.

The procedure combines human-based physical reasoning with machine learning approaches. It starts with tiny datasets (images extracted from less than 1% of the sample) for a better approach to automation. The algorithm chooses places of interest based on what it learns throughout experience and external knowledge.

A method was presented using scanning probe microscopy and applied to well-studied ferroelectric materials as proof of concept. Ferroelectrics are reorientable surface charge functional materials used in computing, actuation and sensing applications. Scientists want to know how the amount of energy or information these materials store relates to the local domain structure that governs this attribute.

The automated experiment has determined for which topological defects these values ​​are optimal. The takeaway is that the methodology was applied to well-known hardware systems in the scientific community and resulted in the fundamental discovery, something previously unknown, very quickly – in this case, within hours.

The results were orders of magnitude faster than traditional procedures, indicating a new approach to intelligent automation. Rather than educating computers solely on data from past trials, this method educates them to think like researchers and learn. The process is based on human intuition. He acknowledges that many material discoveries have been achieved through trial and error by researchers, depending on their skill and experience.

While the focus was on scanning probe microscopy, the technique can be adapted to various additional experimental imaging and spectroscopy methodologies available to the general public. For more information, refer to the published article.