Sub-pixel deep learning ghost imaging for defect inspection in large areas using blurred illumination

A highly sensitive, rapid, and high-resolution method for wide-area microdefect inspection is essential in the semiconductor industry. Deep learning ghost imaging (DLGI) offers high sensitivity and fast measurement capabilities; however, its resolution is constrained by the illumination pattern. To address this limitation, we propose sub-pixel DLGI (SP-DLGI), which leverages illumination to enhance resolution. By utilizing a deep learning model, SP-DLGI predicts sub-pixel defect positions by analyzing subtle intensity variations induced by illumination blurring. Consequently, SP-DLGI enables rapid, high-sensitivity, and high-resolution imaging. Experimental results demonstrate that SP-DLGI effectively predicts defect positions at an 8K resolution.