A Generic Fast and Low Cost BIST Solution for CMOS Image Sensors

Abstract

This paper demonstrates the generalization of a novel test solution embedded inside CMOS Image Sensors (CIS) to classify PASS/FAIL sensors during the test production phase. In [1], a Built-In Self-Test (BIST) solution was proposed to reduce the test time of a CIS, which can represent up to 30% of the final product cost. The major part of the test is dedicated to optical (i.e. image processsing) algorithms performed on the output images from the sensor under test with an Automatic Test Equipment (ATE). The BIST solution reuses these optical algorithms by simplifying and embedding them inside the sensor, to avoid a large amount of data storage and to limit the optical test time. First results on 4,800 output images from a package of sensors have shown a 99.95% correlation between results gathered from an ATE and those achieved with the proposed BIST, with a saving of approximately 30% in optical test time and a negligible area footprint. In this paper, to verify the effectiveness of the BIST solution on a wider set of different CIS (i.e., architecture, size and technology), we experimented the solution on a new database of 28,000 output images from a package of different sensors compared to the first package used in [1]. The BIST parameters have been configured to fit with the new type of sensors and results show a 99.64% correlation, which demonstrates the possible systematic implementation of the proposed BIST solution inside all CIS irrespective of their architecture and technology.

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

“…With the correct constants settings of the BIST (cf. [2]), the correlations between ATE-based optical tests results and PETS BIST results are 99.95% and 99.64% respectively. The correlation depends on the comparison between the label from the STDF file and the label from PETS BIST for each image in the DB.…”

“…The motivation of the work presented in [1] and [2] lies in the fact that the optical test cost, when performed by an ATE, is high in terms of memory usage and test application time due to the storage of the output images and the 2D image processing algorithms.…”

“…This genericity relies on the settings of a percentage of the local average (thanks to amin and amax) and the addition of offsets (thanks to bmin and bmax) to create the most precise range depending on the conditions. The calibration of the constants has been done by comparison between the number of defective pixels in the STDF file and the number of defective pixels found by PETS BIST [2].…”

“…Section II deals with the background of the study by providing information on what is a CIS and how CISs are tested today. Section III summarizes the previous work done in [1] and [2]. Section IV details our proposed innovative PRED BIST solution.…”

“…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.…”