Chelyabinsk Researchers Develop Neural Network Microscope for Non-Destructive Electronics Testing
Scientists at South Ural State University are building a “digital microscope” that can look inside electronic components without cutting or damaging them, using neural networks and standard electrical measurements.
The project is led by Associate Professor Vladimir Surin. The team is focusing on diagnosing varistors - semiconductor resistors that protect electronics from voltage surges. Today, such components are typically examined using costly electron microscopy that requires destroying the sample. The new method takes a different approach: the system reads the component’s electrical characteristics, and an ensemble of neural networks reconstructs its internal microstructure within seconds.
The system combines three types of neural networks: LSTM filters signal noise, PINN incorporates physical laws, and GAN generates an image of the internal structure. The result is a microstructure map that is statistically indistinguishable from images obtained through electron microscopy.
The technology is implemented as a software module that can be integrated into existing testing equipment. In the future, the method could be applied to diagnostics of ceramics, sensors and composite materials.
