Automated Defective Pin Detection for Recycled Microelectronics Identification

Abstract

Counterfeit integrated circuits (ICs) constitute a major threat to system reliability, as well as security from personal to national scale. Most current techniques for counterfeit electronics detection are expensive, time consuming, and prone to human error. We have proposed a convolutional neural network (CNN) architecture–based supervised technique along with two unsupervised techniques based on depth map and texture of pins to identify bent and corroded pins respectively with high accuracy, thus helping to identify recycled ICs. The supervised technique requires one-time training, and is amenable to be integrated in a fast and automated counterfeit IC detection methodology. We have also compared the proposed CNN–based classification technique accuracy with support vector machine (SVM) and K-nearest neighbor (KNN)–based classification techniques. Using these methods, both corroded and bent pins are differentiated with high accuracy. The unsupervised bent pin detection technique uses depth map images of ICs to construct 3D images of ICs and the corroded pin detection methodology uses Laws’ texture energy method and K-means clustering to differentiate between defective and non-defective pins.