The work aims to propose a lightweight printed circuit board (PCB) surface defect detection framework, EYOLOv8, based on YOLOv8, to address the issues of large model size and slow speed in surface defect detection of PCB. Based on the YOLOv8 network structure, a network backbone was reconstructed according to the RevCol network feature fusion concept, the neck structure was redesigned according to the Slim-neck design concept, and the detection head structure was reconstructed according to the mechanism of convolutional weight parameter sharing. While maintaining nearly the same accuracy, the overall network structure was optimized for lightweight design. Finally, the training process of the lightweight model was optimized with the WIoU loss function. Experimental results on public datasets showed that EYOLOv8 reduced the model size by 46% compared with YOLOv8, achieved a detection accuracy mAP50 of 97.7%, and operated at a detection speed of 256 frames per second with a model size of 3.3 MB. Compared with other algorithms, EYOLOv8 demonstrates greater competitiveness when deployed on PCB surface defect detection devices.