Predictor BIST: An 'All-in-One' Optical Test Solution for CMOS Image Sensors

The optical test of CMOS Image Sensors (CIS) represents a major part of the overall test time. It consists in applying two-dimensional image processing algorithms on the output images of the CIS under test. In [1], we demonstrated that it is possible to embed 50% of these algorithms in the CIS by using a dedicated Built-In Self-Test (BIST) engine without impacting the defect coverage and at a negligible area cost (approximately 0.25% of the total sensor area). The use of a BIST solution reduces the optical test time by roughly 30% when compared to the optical algorithms traditionally applied by using an Automatic Test Equipment (ATE). However, the effectiveness of this approach is limited by the fact that some global subtle defects cannot be detected. This is a real problem as these defects can seriously impact the output image quality. To overcome this drawback, this paper presents a new BIST solution, called Predictor BIST (PRED BIST) based on a global overview of the image under test. The novelty of this approach lies in the use of an arithmetic method to build an ideal (i.e., predicted) image from wisely selected samples extracted from each output image of the CIS under test. The process of building an ideal image is repeated for all images to be checked for the CIS under test. By comparing each output image to the ideal image, it is then possible to classify good and bad images. Moreover, PRED BIST is an all-in-one test solution that allows to fully replace the optical test done today with an ATE by an embedded test solution. This is an important and unique feature of PRED BIST. Two databases were used to validate this new BIST solution. The first one is an extension of the monocolor database of 27,600 images used in [2] and the second one contains 81,840 images from a RGB sensor. Validation shows that compared to conventional ATE-based test, a correlation of 99.90% and 99.40% respectively is achieved regarding the classification between good and bad images of two considered databases. Moreover, the global ...