Detection and imaging of sub-surface defects using time-resolved wide-field surface thermometry and artificial intelligence

This technology uses synchronized pulsed lasers and AI to non-destructively image and map hidden defects inside solid materials by analyzing rapid surface temperature changes, enabling fast, high-resolution detection without scanning or expensive equipment.

Background

The field of non-destructive testing is critical for ensuring the structural integrity and reliability of solid materials in advanced manufacturing sectors, such as microelectronics, aerospace, and automotive industries. As components like three-dimensional integrated circuits become increasingly miniaturized and complex, there is a vital need for high-resolution, high-throughput inspection methods. Manufacturers require reliable techniques to identify hidden subsurface defects, such as microscopic voids and cracks, during the production process. Effective inline metrology ensures quality control, prevents component failure, and maintains high production yields without physically altering or damaging the delicate manufactured samples.

Despite this critical need, existing inspection methodologies suffer from significant limitations. Traditional X-ray microscopy requires expensive infrastructure and hazardous radiation shielding, while offering impractically low throughput for inline manufacturing. Ultrasonic imaging necessitates immersing components in liquid coupling media, which is highly unsuitable for sensitive electronics. Furthermore, conventional infrared thermography lacks the spatial resolution required to detect nanoscale flaws and relies on one-dimensional approximations that fail to account for lateral heat diffusion. Additionally, current mechanical scanning techniques are prohibitively slow, and existing data-driven thermal models struggle with ill-posed inverse problems, often requiring massive labeled datasets to function.

Technology

The non-destructive subsurface defect imaging system combines wide-field surface thermometry with an artificial intelligence reconstruction framework. It utilizes a focused pulsed pump laser to inject localized heat and a synchronized wide-field pulsed probe laser to measure transient surface temperature changes. An electronic controller manages the delay, enabling a camera to capture thermal images without mechanical scanning. To identify buried defects, an unsupervised algorithm parameterizes the 3D thermal diffusivity field using a neural network. This network is optimized by minimizing the discrepancy between measured temperatures and those simulated by a differentiable physics solver, enforcing the heat diffusion equation as a hard constraint.

This solution is differentiated by achieving sub-micrometer spatial resolution and rapid, non-scanning imaging of the entire field-of-view simultaneously. Unlike traditional X-ray or ultrasonic methods, it requires no radiation shielding or immersion media, making it cost-effective and easily integrable into industrial assembly lines. Furthermore, the physics-informed AI framework is entirely unsupervised, eliminating the need for labeled defect datasets. By incorporating adjoint methods for memory-efficient gradient computation, frequency annealing, and total variation regularization, the system overcomes the ill-posedness of traditional inverse heat conduction to ensure the accurate, physically plausible reconstruction of distinct defect boundaries.

Benefits

  • High Throughput and Non-Scanning Operation: Captures the entire field-of-view simultaneously using wide-field imaging, eliminating the need for slow mechanical scanning and significantly increasing inspection speeds.
  • Exceptional Resolution: Achieves sub-micrometer spatial resolution (down to ~100 nm) and ultrafast temporal resolution (~100 ps) for highly precise 3D profiling of buried defects like voids and cracks.
  • Cost-Effective and Easily Integrable: Utilizes inexpensive pulsed diode lasers and standard optical components in a compact, modular design that seamlessly integrates into existing industrial assembly lines.
  • Non-Destructive and Safe: Operates safely at low power without requiring hazardous radiation shielding (unlike X-ray microscopy) or liquid immersion media (unlike ultrasonic imaging).
  • Unsupervised AI Reconstruction: Employs a physics-informed neural network to accurately reconstruct 3D defect shapes and locations without the need for large, labeled training datasets.
  • Robust and Generalizable: Enforces physical heat diffusion equations as hard constraints during AI optimization, making the system highly robust to noise and easily adaptable to novel materials and sample geometries.

Applications

  • 3D integrated circuit inspection
  • Semiconductor wafer defect detection
  • Aerospace composite subsurface imaging
  • Automotive parts quality control
  • Inline advanced manufacturing metrology

Additional info

This non-destructive imaging system uses a synchronized pulsed pump laser for localized heating and a wide-field probe laser to capture transient surface temperatures. An electronic controller manages delay without mechanical scanning. A processor reconstructs 3D subsurface defects using an unsupervised multilayer perceptron with sinusoidal positional encoding. This neural network is optimized via a differentiable physics solver enforcing heat diffusion, utilizing adjoint methods, frequency annealing, and total variation regularization.