AI-driven microscope boosts materials science

Chinese scientists have developed the world's first artificial intelligence-driven transmission electron microscope system, marking a milestone leap from traditional manual operation to fully autonomous, AI-driven research.

The system, named "Aeye-1", was developed by researchers from the Chinese Academy of Sciences' Dalian Institute of Chemical Physics, in collaboration with the CAS' Shenyang Institute of Automation.

On May 24, the technology officially passed a rigorous scientific and technological achievement evaluation conducted by the China Petroleum and Chemical Industry Federation in Beijing. An expert panel unanimously concluded that the system represents a global first, positioning China at an internationally leading level in advanced scientific instrumentation.

Advanced scientific instruments are crucial to national self-reliance and technological security. As scientific exploration reaches into the ultra-microscopic realm, the transmission electron microscope has become an indispensable tool for studying advanced materials, clean energy, and life sciences.

However, a century has passed since the invention of the transmission electron microscope, and TEM technology has relied heavily on manual operations. This reliance has long created major bottlenecks, including low efficiency, high subjectivity, and difficulties in generating accurate, large-scale quantitative data.

To resolve this critical "choke-point" challenge, the CAS research team spent years focusing on the integrated development of both software and hardware of the TEM system.

By overcoming five core technical hurdles — using embodied intelligence for high-vacuum sample transfer, autonomous electronoptics alignment, nanoscale precision localization, autonomous TEM imaging and intelligent analysis, and system state perception and intelligent scheduling — the team successfully built the "Aeye-1". The system acts as a "smart eye" for the microscopic world, achieving completely unmanned, intelligent operations across the entire workflow of sample supply, imaging and data analysis.

"The system enables fully unmanned, AI operation across the entire workflow. Its image analysis efficiency is more than 300 times higher than manual processing," said Deng Dehui, lead professor at DICP.

In practical testing involving the microstructural analysis of molecular sieves — porous crystalline materials widely used as industrial catalysts to accelerate chemical reactions in energy and chemical engineering — the AI-TEM system demonstrated unprecedented productivity.

On average, "Aeye-1" can analyze 168 samples and acquire more than 4,000 images daily. This automated workflow allows it to process data at a speed more than 300 times faster than manual processing. Remarkably, the volume of data obtained from just two weeks of operating "Aeye-1" is equivalent to an entire year's workload for a conventional, manually operated TEM.

Furthermore, the system automatically generates comprehensive, professional analytical reports detailing microstructural statistics.

The system is expected to accelerate breakthroughs in fields like materials genomics — a cutting-edge method that uses massive data libraries to fast-track the discovery of new materials — as well as green energy development and the life sciences.