A 0.15-million-parameter attention-equipped neural network classifies nine wafer-map defect types with 99.88 percent accuracy ...