Looking through samples of material through a microscope for hours and hours, trying to locate monolayers, is one of the most laborious and intimidating tasks for undergraduate assistants in university research laboratories.
Researchers can process 100 images covering 1cm × 1cm samples like this in about nine minutes using a new system that greatly simplifies the often tedious search for monolayers in the lab. Image credit: Photo of the University of Rochester / J. Adam Fenster.
As a result of their unique properties, these two-dimensional materials, which are less than 1 / 100,000 the width of a human hair, are in high demand for use in photonic, electronic and optoelectronic devices.
Research labs are hiring armies of undergraduate students to do nothing but look for monolayer. It’s very tedious, and if you get tired, you may miss some of the monolayers or you may start making misidentifications.
Jaime Cardenas, Adjunct Professor, Optics, University of Rochester
Even after all this, labs have to review materials using expensive atomic force microscopy or Raman spectroscopy.
Jesús Sánchez Juárez, Ph.D. Cardenas Lab student, has simplified the work for undergraduate students, their research facilities, and companies that have difficulty identifying monolayers.
The unique technology is an automated scanning device that can detect monolayers with 99.9% accuracy, surpassing any other technique to date. The study was published in the journal Optical Materials Express.
At a fraction of the price. In much less time. With easily accessible materials.
One of the main goals was to develop a system with a very small budget so that students and labs could replicate these methodologies without having to invest thousands and thousands of dollars just to buy the necessary equipment.
Jesús Sánchez Juárez, PhD student and lead author of the study, University of Rochester
For example, the device you developed can be replicated using a low-cost OEM (OEM) camera and a low-cost microscope with a 5 × lens.
A creative adaptation of an AI neural network
We are very excited. Jesus did several new and different things here, applying artificial intelligence in a new way to solve a big problem in the use of 2D materials.
Jaime Cardenas, Adjunct Professor, Optics, University of Rochester
By training an artificial intelligence (AI) neural network to scan monolayers, numerous labs have attempted to eliminate the need for expensive backup characterization tests that require human scanning. According to Cárdenas, most laboratories that have tried this approach plan to build a network from scratch, which takes a long time.
Rather, Sánchez Juárez started with AlexNet, a publicly available neural network that has already been trained to identify objects.
He then devised a new method for inverting material images so that what was bright in the original image would look black and vice versa. Additional processing steps are applied to the inverted images.
At the moment, the images “do not look good to the human eye, but it is easier for a computer to separate the monolayers from the substrates on which they are deposited,” says Cárdenas.
Conclusion: Sánchez Juárez’s system can process 100 images covering 1 cm × 1 cm size samples in 9 minutes with almost perfect accuracy, compared to the long and laborious scanning hours of undergraduate students.
Sánchez Juárez adds: “Our demonstration paves the way for the automated production of monolayer materials for use in research and industrial environments, greatly reducing processing time.”
Applications include 2D materials suitable for photodetectors, lasers, excitonic light emitting devices (LEDs), single photon emission, optical generation of spin-valley currents and modulators.
Magazine reference:
Sánchez Juárez, J., et al. (2022) Automated system for the detection of 2D materials through digital image processing and deep learning. Express optical materials. doi.org/10.1364/OME.454314.
Source: