Multi-step ART1 algorithm for recognition of defect patterns on semiconductor wafers

Abstract

The integrated circuits (ICs) on wafers are highly vulnerable to defects generated during the semiconductor manufacturing process. The spatial patterns of locally clustered defects are likely to contain information related to the defect generating mechanism. For the purpose of yield management, we propose a multi-step adaptive resonance theory (ART1) algorithm in order to accurately recognise the defect patterns scattered over a wafer. The proposed algorithm consists of a new similarity measure, based on the p-norm ratio and run-length encoding technique and pre-processing procedure: the variable resolution array and zooming strategy. The performance of the algorithm is evaluated based on the statistical models for four types of simulated defect patterns, each of which typically occurs during fabrication of ICs: random patterns by a spatial homogeneous Poisson process, ellipsoid patterns by a multivariate normal, curvilinear patterns by a principal curve, and ring patterns by a spherical shell. Computational testing results show that the proposed algorithm provides high accuracy and robustness in detecting IC defects, regardless of the types of defect patterns residing on the wafer.

A Review and Analysis of Automatic Optical Inspection and Quality Monitoring Methods in Electronics Industry

“…For example, at the IC placing stage, defect cases of missing, wrong or doubled components may occur. In terms of possible soldering defects, most of them happen after the reflowing stage, such as the defects at the IC package components (pseudo joint, excess solder, insufficient solder, shifting Solder and bridge defects) and the defects at the non-IC components (side termination, [63] Ring, scratch, zone and repeating types [64] Ring, scratch, random and new patterns [65] Systematic and random patterns [66] Circle, cluster, scratch and spots [67] [68] Bull's Eye, Edge ring, scratch, random, multiple zones, multiple scratches, ring-zone mixed pattern and ring-scratch mixed pattern [69] [70] Multiple zones, multiple scratches, ring-zone mixed pattern and ring-scratch mixed pattern [71] [72] Cluster defects such as scratch, strains and localized failures [58] Checkerboard, ring, right-down edge, composite and random patterns [73] Spatially homogeneous Bernoulli process, cluster, circle, spot, repetitive and mixed pattern [74] Scratch, center and edge [75] Quarter ring, up and left, Quarter ring, up and right, Edge effects, Ring effects, Semi-ring, up, Semi-ring, up Edge effects, up and bottom Cluster [76] Annulus, half-annulus, band and half-ring [77]- [81] Curvilinear, amorphous, and ring [82] Linear and circular patterns [83] [84] Bull's eye, Bottom, Crescent moon, edge and random [85] Random, ring, curvilinear and ellipsoid [59] Line, edge, ring, blob and bull's eye [61] [53] Bull's eye, blob, line, edge, hat and ring [86] Multiple patterns including ring, checkerboard and five radial zones [87] Random, systematic and ,mixed patterns [88] [57] Circle, cluster, repetitive and spot [56], [60], [62], [89]-…”

“…However, the low number of WBM provided to them limited their ability to identify further patterns. Choi et al in [85] conducted a study similar to what have been done in [64], [65]. However, they proposed an advanced ART1 algorithm called "multi-step ART1", where it sequentially uses the modules to classify each pattern separately, instead of training all patterns at once.…”

Randomized General Regression Network for Identification of Defect Patterns in Semiconductor Wafer Maps

“…In addition, ART1 had better classification abilities in detecting similar defect patterns, yet different in size, location, etc. In 2011, Choi et al [16] introduced a further improvement to the ART1 algorithm by proposing a MultiStep-ART1 that includes similarity evaluation and data pre-processing scheme. Experiments showed that the proposed method was capable of identifying mixed patterns on wafers, and was able to classify patterns with higher accuracy and lower computational time as compared to ART1.…”

A Deep Convolutional Neural Network-Based Multi-Class Image Classification for Automatic Wafer Map Failure Recognition in Semiconductor Manufacturing

“…If the wafer map failure types are predefined in the training data, then supervised learning techniques such as k-Nearest Neighbor (KNN) [7], decision tree [4], ANN [8], and SVM [3,9], etc., can be applied to generate the classification models for the common wafer map failure types. On the other hand, unsupervised learning techniques-such as the clustering [5], adaptive resonance theory network 1 (ART1) [10], multi-step ART1 [11], etc., methods-can be used to recognize wafer map failure types where there is unknown prior information regarding failure type in the training data. In addition, deep learning (DL)-based approaches such as convolutional neural networks (CNN) have recently been used for image processing tasks in many domains [12][13][14][15][16].…”