This review analyzes the evolution of automatic PCB inspection techniques, highlighting the shift from classical image-processing methods to Deep Learning–based solutions. The study shows that lightweight architectures equipped with attention mechanisms significantly improve micro-defect detection while preserving the performance required for industrial environments. The hardware analysis indicates that reconfigurable platforms such as FPGAs outperform CPUs and embedded GPUs by delivering deterministic, real-time inference with superior efficiency. Finally, the work points to promising future directions, such as self-supervised learning, explainable AI, and synthetic data generation, that are expected to drive the next generation of inspection systems toward accurate “Zero-Defect” manufacturing.