August_EDFA_Digital

edfas.org 15 ELECTRONIC DEV ICE FA I LURE ANALYSIS | VOLUME 24 NO . 3 described in previous work are followed. [11] The raw SEM image shared by the SiP integrator is denoised, binarized, and followed by calculating connected components to identify all the cell components in the images as shown in Fig. 3. A set of distinctively labelled vertically sorted connected cell components, C = {c 0 , c 1 , ... ..., c T-1 } (Fig. 3d) is obtained where each c i represents a component with top-left (x 1 ,y 1 ) and bottom-right (x 2 ,y 2 ) points (Fig. 3f). These components have been used to separate cell rows from images as described in the next section. CELL SEPARATION Connected components acquired in the previous section are used todrawboundaries between consecutive cell rows. Later cells are extracted one-by-one from each row R k (Fig. 4a) and passed through the classifier. Scanning through the vertically sorted component list (C), component c i marks the beginning of a row if and only if it situates lower than the previous component c i-1 . This has been measured by comparing the y-coordinates of the corresponding components. The above inequality provides the row-decider com- ponents c i , (marked bywhite boxes in Fig. 4a). Then based on the top-left coordinate of each c i , rows are separated. After separation, each row is scanned left to right to identify single and composite cells. Each row R k is com- posed of two symmetric parts, P 0 and P 1 (see Fig. 4b(i)). For simplicity, one of the Ps is processed. While scanning from left to right on a row R k , distances between con- secutive components are measured. Two components C i and C i-1 have been part of a single cell if the pixel- level distance between them falls below a threshold (see Fig. 4b (ii) and (iii)). Fig. 3 Different stages of preprocessing. Cell margins are determined according to the newly merged components. Then cells are extracted following the margins as shown in Fig. 4b (iv) and (v).