Defect Detection at HP Labs:
In the summer of 2007 I worked as an intern at HP Labs under Carl Taussig, Warren Jackson, and Alison Chaiken. In order to assist their team with the creation of new display technologies I developed a system to optically inspect fabricated circuits for defects. The system works roughly as follows:
The circuit has a large array which contain copies of a basic structure. The user takes a representative image of this structure. Using an interface I created they associate classifiers with various regions of this template image. The types of classifiers which can be are either one of several fixed algorithms, or the user can train Support Vector Machines by providing labeled examples of the defect.
At a later time, a display which contains an MxN array of the basic object is scanned with an industrial-grade digital camera with a high-zoom lens. The image is classified according to the template created by the user and statistics on the locations and counts of defects found in the images is presented to a user. In this way the user can quickly determine whether a change in the manufacturing process increased or decreased the yield, and where the defects tend to be clustering. This system is intended to be used in the months to come in order to fine-tune their manufacturing process.