Manufacturing Bits: March 24

Autonomous microscopes; nanoindentation; AI tomography.


Autonomous microscopes
FLEET, also known as the ARC Centre of Excellence in Future Low-Energy Electronics Technologies, has developed an autonomous scanning probe microscopy (SPM) technology.

SPM is an instrument that makes use of an atomically sharp probe. The probe is placed in close proximity above the surface of a sample. With the probe, the SPM forms images of the surface of the sample at the atomic scale.

As with most metrology tools, SPM requires human supervision. But using machine learning, FLEET has demonstrated a framework for autonomous SPM operations. A subset of AI, machine learning utilizes a neural network to crunch data and identify patterns. It matches certain patterns and learns which of those attributes are important.

FLEET’s technology is called DeepSPM. “DeepSPM includes an algorithmic search of good sample regions, a convolutional neural network to assess the quality of acquired images, and a deep reinforcement learning agent to reliably condition the state of the probe,” according to researchers in Communications Physics, a technology journal. “DeepSPM is able to acquire and classify data continuously in multi-day scanning tunneling microscopy experiments, managing the probe quality in response to varying experimental conditions.”

All told, the autonomous SPM operations removes the need for constant human supervision. “Optimizing SPM data acquisition can be very tedious. This optimization process is usually performed by the human experimentalist, and is rarely reported,” said FLEET Chief Investigator Agustin Schiffrin. “Our new AI-driven system can operate and acquire optimal SPM data autonomously, for multiple straight days, and without any human supervision.”

“Crucial to the success of DeepSPM is the use of a self-learning agent, as the correct control inputs are not known beforehand,” added Cornelius Krull, project co-leader. “Learning from experience, our agent adapts to changing experimental conditions and finds a strategy to keep the system stable.”

Researchers at Monash University’s School of Physics and Astronomy worked on the SPM technology with the Max Planck Institute of Molecular Cell Biology and Genetics, Max Delbrück Center for Molecular Medicine and Heidelberg University.

Using machine learning, the Massachusetts Institute of Technology, Brown University and Nanyang Technological University have developed a way to make nanoindentation more accurate.

Nanoindentation is a technology that measures the mechanical properties for a multitude of materials. It measures the modulus and hardness of materials.

Several companies provide a system called a nanoindenter, which performs the art of nanoindentation. For example, Nanoscience Instruments has a system, in which a diamond probe with a 100nm tip is used to indent the surface of a sample. “The load applied to the tip and the depth of penetration is subsequently measured during the process,” according to Nanoscience Instruments. “The indentation depth is used to calculate the area of the tip that was in contact during the indentation. This area is used to measure the hardness of the material.”

Generally, the indentation force can be measured on the order of one-billionth of a Newton, according to MIT. The Newton is a unit of force. “One newton is the force needed to accelerate one kilogram of mass at the rate of one meter per second squared in the direction of the applied force,” according to one definition of the term.

Using machine learning, MIT, Brown, and Nanyang Technological University have developed a new analytical technique to improve the estimation of the mechanical properties of metallic materials. This technique is said to have a 20X greater accuracy than existing methods.

The new method doesn’t require any new equipment, but rather it makes use of machine learning using a neural network. Machine learning requires a large amount of data. Otherwise, the results may not be accurate.

Instead of the traditional methods, researchers from MIT and others took a different approach. “The team found that doing the neural network training with lots of low-cost synthetic data and then incorporating a relatively small number of real experimental data points — somewhere between three and 20, as compared with 1,000 or more accurate, albeit high-cost, datasets — can substantially improve the accuracy of the outcome,” according to MIT.

Faster APT
The U.S. Department of Energy’s (DOE) Argonne National Laboratory has advanced atom-probe tomography (APT) using deep learning.

APT is a metrology system, which enables 3D atomic reconstruction of materials. By using machine learning, researchers can accelerate the APT process without having to sacrifice accuracy.

“Our method is scalable, you can put it on high performance computing and fully automate it, rather than going through manually and looking at different concentrations,” Argonne materials scientist Olle Heinonen said. “Here you send your code and push a button.”

Leave a Reply

(Note: This name will be displayed publicly)