Semiautomatic Segmentation of Transistor Gates in Integrated Chips

Piali Das, Olga Veksler, Vyacheslav Zavadsky, Yuri Boykov

In Workshop on Applications of Computer Vision, (ACV) in conjunction with ECCV, Graz, Austria, May 2006.

Abstract

In recent years, interactive methods for segmentation are increasing in popularity due to their success in different domains such as medical image processing, photoeditng, etc. In this paper we discuss a challenging industrial application of transitor gate segmentation in the images of integrated chips, which is esential for reverse engineering tasks. Segmentation in the domain of integrated chips is very difficult due to large variations in contrast and noise type and also due to extreme variation in the size of the transitor gates, which can range from a few pixels to a few thousands of pixels in length, and from one to several hundred pixels in width. We present a semiautomatic segmentation algorithm that produces reliable and accurate segmentation of a transistor gate from its background with the minimum guidance from the user, who just has to click on one pixel inside the transitor gate of interest. The algorithm is based on the powerful graph-cut interactive segmentation technique of Boykov and Joly [1]. In order to obtain accurate and robust segmentation with such low user interaction, we make several assumptions based on our observations of the transitor gate images. The main assumption is that the transitor gates are aproximately "compact" in shape, or can be approximated by several roughly collinear "compact" parts. To achieve robustness in segmentation, we incorporate the "compact" shape prior in to the framework of [1]. The use of the "compact" shape prior alows us to introduce a parameter "bias" to bias the segmentation towards larger object boundaries, which counteracts the general tendency of the algorithm in [1] to produce smaller segments. In order to accommodate large variation in the quality of the images, most parameters in the algorithm are selected automatically to adapt to the current image. An application developed on the basis of our algorithm runs in real-time and is being used by Semiconductor Insight Inc.


WHOLE PAPER: pdf file (274 Kb)
OTHER VERSION: see CRV 2006 paper