Artificial Neural Network-driven Diffraction Imaging for Nanoscale Optical Critical Dimension Metrology in Semiconductor Manufacturing

This study introduces a new optical measurement technique that integrates artificial neural network (ANN) technology with coherent optical scatterometry and back focal plane image to address the limitations of conventional methods, such as scanning electron microscopy and atomic force microscopy (AFM). Designed for three-dimensional advanced semiconductor packaging applications, the proposed approach combines diffraction imaging with energy distribution in periodic structures to train an ANN model. The technique enables precise measurement of line width, line spacing, and the height of redistribution layer structures or through-silicon vias. Compared with AFM measurements, the method exhibits a less than 2% measured bias for structure top width and spacing, and under 1.2% for height measurements. This nondestructive optical method completes measurements in less than 3 ms, a 104-fold speed increase compared to the traditional library search. It also offers considerable improvements in speed and cost-efficiency over traditional techniques, which make it ideal for integration into production environments. The development has been proven an effective tool in semiconductor manufacturing and provides an inline process control solution for accurate critical dimension measurements.