edfas.org ELECTRONIC DEVICE FAILURE ANALYSIS | VOLUME 28 NO. 1 18 First, a smoothing filter such as a Gaussian blur is used to remove high frequency noise and decrease the chance of only partially masking a potential site. The algorithm receives this filtered image and calculates the average intensity value of all pixels. Next, it splits the image into subregions in an n x n grid, each of which is compared against the average pixel value of the total image. If the value breaches a pre-selected percent difference (the threshold referenced as the “sensitivity”), then the pixels in the subregion are marked as a region of interest (ROI). Equation 1 below describes the numerical thresholding used for ROI determination. (Eq 1) When the laser raster position is not over the determined ROI, SAILS sends pulses with the fastest singlepixel dwell-time of the tool; a “fast dwell” value of around 4 μs per pixel. When the laser position reaches the ROI, the laser dwell time is increased to the length of the test loop time. ARRAY-BASED IMAGE RECONSTRUCTION The cumulative averaging of standard SDL method is a lossy process since it discards individual pass/fail (binary) frames and only presents a single continuous averaging frame. As such, further algorithmic site detection improvements are limited because only a single reference image is available. To address this issue, an array-based image reconstruction methodology is introduced that aims to ensure lossless data capture by preserving all the individual binary frames for further analysis. In array-based image reconstruction methodology for site detection, each individual frame is captured and subjected to a reconstruction process, resulting in the production of numerous reconstructed frames. Subsequently, the processed reconstructed frames are processed and merged to form a single ROI mask that emphasizes the pri- mary components and borders of the site. This ROI mask is depicted as a red enclosure, as illustrated in Fig. 5. LASER SCANNING RESULTS For the test case, a marginally failing test case was chosen and both passing and failing sites were observed. SAILS was tested and compared against an unmodulated 200 frame scan of a 160 x 60 frame with a test loop frequency of 240 Hz. For a reference unmodulated frame scan, an external waveform generator was used to send a trigger signal at a constant frequency of 240 Hz, completing an average of 200 frames in 135 minutes. With SAILS, an average of 5 frames was captured as the rough scan data and passed to the site detection algorithm. The rough scan was subdivided into 100 boxes (each 10 x 16 pixels), and the sensitivity value in this case was set to 7%. With these parameters, two clusters of ROIs were identified by the site detection algorithm to be potential sites shown in Fig. 6. The scan continued for 195 more frames where the step frequency was modulated to dwell at a frequency Fig. 5 Methodology for continuous average and array-based site detections. Fig. 4 Smoothing filter applied to above comparison images, showing matching diameters (8 pixels) and centers at coordinates (37,40) of the site between images.
RkJQdWJsaXNoZXIy MTYyMzk3NQ==