Optical Inspection of the Silicon Micro-strip Sensors for the CBM Experiment employing Artificial Intelligence
This work addresses quality assurance in sensor manufacturing for high-energy physics experiments, representing an incremental application of existing AI methods to a specific domain.
The researchers tackled the problem of detecting surface defects in silicon micro-strip sensors for the CBM experiment by applying convolutional deep neural networks to optical inspection data, achieving identification of various defect types and proposing a method for quality grading based on defect counts.
Optical inspection of 1191 silicon micro-strip sensors was performed using a custom made optical inspection setup, employing a machine-learning based approach for the defect analysis and subsequent quality assurance. Furthermore, metrological control of the sensor's surface was performed. In this manuscript, we present the analysis of various sensor surface defects. Among these are implant breaks, p-stop breaks, aluminium strip opens, aluminium strip shorts, surface scratches, double metallization layer defects, passivation layer defects, bias resistor defects as well as dust particle identification. The defect detection was done using the application of Convolutional Deep Neural Networks (CDNNs). From this, defective strips and defect clusters were identified, as well as a 2D map of the defects using their geometrical positions on the sensor was performed. Based on the total number of defects found on the sensor's surface, a method for the estimation of sensor's overall quality grade and quality score was proposed.