A spatiotemporal attention operator for monitoring thermo-mechanical stress in wafer-scale integrated circuits using an infrared camera

An attentive vision method for thermal stress detection and monitoring using a multi-scale spatiotemporal attention operator is proposed for monitoring overheating in wafer-scale integrated circuits. This method represents a multi-scale and multi-temporal analysis of infrared image sequences of the inspected surfaces by an attention operator to detect feature points. Such points may indicate possible sources of abnormal thermo-mechanical stress and overheating. This operator implements linear aggregation of a temporal change filter with a corresponding spatial saliency filter using an optimized value of the temporal change coefficient to extract the multi-scale feature points. The monitoring is mostly carried out through the detection and tracking of stress-relevant feature points based on the multi-scale area spatial and temporal descriptors extracted at the feature points. The experiments conducted with infrared image sequences have confirmed the reliability of the proposed operator and showed its high potential in surface image analysis for monitoring of wafer-scale integrated circuits.