Semiconductor　defect　inspection　is　crucial　for　yield　improvement　but　is　hindered　by　manual　inspection＇s　subjectivity　and　error.　This　paper　employs　Convolutional　Neural　Networks　(CNNs)　for　automated　wafer　defect　classification,　addressing　the　challenges　of　time-intensive　training　and　complex　hyperparameter　tuning.　We　propose　the　Arithmetic　Optimization　Algorithm　(AOA)　to　efficiently　optimize　CNN　hyperparameters　like　momentum,　initial　learning　rate,　maximum　epochs,　and　L2　regularization.　Our　method　reduces　the　trial-and-error　in　hyperparameter　tuning.　Using　the　AOA-optimized　ResNet-18　model,　our　simulations　show　superior　performance　in　defect　classification　compared　to　the　unoptimized　model,　demonstrating　its　effectiveness　and　practical　potential.　©　The　2024　International　Conference　on　Artificial　Life　and　Robotics　(ICAROB2024).