Test for low cost CMOS image sensors

Summary form only given. Low cost CMOS image sensors are used in various applications, but the most prominent are camera phones, the fastest growing consumer electronics product in history, going mainstream in less than 5 years from initial introduction. The use of CMOS rather than CCD is discussed, and circuit details given of common pixel designs. The more recent four transistor cell is compared with three transistor cells with respect to image quality and noise. Necessary enhancements to a traditional CMOS process are discussed, needed to produce color filters over individual pixels and a microlens array to capture more light. An integral part of typical systems is an image processor, which takes raw sensor data and converts it into a color image. Brief details of a typical image pipeline are presented, which includes descriptions of demosaic, white balance, color correction and gamma correction. Test considerations deal primarily with the sensor array. The image pipeline is digital logic and tested using traditional approaches. Although these are primarily structural, the dedicated nature of the logic allows some functional tests to be used as effective screeners. Wafer test of digital logic must have high coverage as scan based tests are typically not able to be applied at module level. Array defects can give rise to either random or fixed pattern noise. The eye is significantly more sensitive to fixed pattern noise so special effort is needed to detect it. Causes of defects are discussed, breaking them down into silicon defects and fall-on particles. It is shown how manifestation of these defects, as image blemishes, varies considerably according to test conditions. These conditions include illumination level, exposure, temperature, and whether raw sensor images or demosaiced color images are analyzed. Defective pixel cluster size and amount of deviance are also parameters which need to be considered. Finally, pixel correction is discussed. Since the sensor is a large array, spatial redundancy is utilized to correct isolated defective pixels based on values of neighbor pixels. The challenge is to avoid classifying good pixels as bad, which results in replacing their values, thereby corrupting an otherwise perfectly good image.