AI finds 2D materials in the blink of an eye
A team of researchers from the University of Tokyo have succeeded in discovering the extremely hard to find two-dimensional (2D) crystals of graphene under a microscope by making an Artificial Intelligence (AI) algorithm. “By using machine learning instead of conventional rule-based detection algorithms, our system was robust to changing conditions,” says first author Satoru Masubuchi. They trained the algorithm to detect the outline and thickness of flakes without having to fine-tune the microscope by using several examples under various lighting.
The two-dimensional (2D) material offer an exciting new avenue for the creation of electronic devices such as transistors and light-emitting diodes. The family of crystals that can be made just one atom thick include metals, semiconductors, and insulators. Several of these are stable under normal conditions and have very distinct properties from their three-dimensional counterparts. Stacking a few layers of these together can significantly alter the device – making it suitable for next-gem batteries, smartphone screens, detectors and solar cells; all from an abundantly available material – graphene. The 2010 Nobel Prize in Physics was awarded for the realisation that atomically thin “graphene” can be obtained by an exfoliating piece of pencil lead, graphite, with a piece of sticky scotch tape.
If graphene is so ‘easily’ and widely available, why then do we not make more electronic devices with this material? Because the 2D crystals of graphene are extremely thin, have low fabrication yield and their optical contrasts comprise a very broad range, and finding them under a microscope is a tedious job. However, the AI algorithm has managed to automate this job, opening up avenues for the development of new electronic devices and further experimentation using 2D materials. “The automated searching and cataloguing of 2D materials will allow researchers to test a large number of samples simply by exfoliating and running the automated algorithm,” lead author Tomoki Machida says.
The algorithm isn’t only limited to detecting graphene crystals. It has been able to detect tungsten diselenide and molybdenum diselenide flakes just by being trained with tungsten ditelluride examples. With the ability to determine, in less than 200 milliseconds, the location and thickness of the exfoliated samples, the system can be integrated with a motorized optical microscope. “This will greatly speed the development cycle of new electronic devices based on 2D materials, as well as advance the study of superconductivity and ferromagnetism in 2D, where there is no long-range order,” mentioned Machida.