Jenny Leung

Quantitative analysis of in-field defects in image sensor arrays

Growth of pixel density and sensor array size increases the likelihood of developing in-field pixel defects. An ongoing study on defect development in imagers has now provided us sufficient data to be able to quantify characteristics of defect growth. Preliminary investigations have shown that defects are distributed randomly and the closest distance between two defective pixels is approximately 79-340 pixels apart. Furthermore, from an observation of 98 cluster-free defects, the diameter of the defect is estimated to be less than 2.3% of a pixel size at 99% confidence level. The fact that no defect clusters were found in the study of various digital cameras allows us to conclude that defects are not likely to be related to material degradation or imperfect fabrication but are due to environmental stress such as radiation. Furthermore, as verified by a statistical study, the absence of defect clustering provides information on the size of defects and insight into the nature of the defect development.

Statistical Identification and Analysis of Defect Development in Digital Imagers

The lifetime of solid-state image sensors is limited by the appearance of defects, particularly hot-pixels, which we have previously shown to develop continuously over the sensor lifetime. Analysis based on spatial distribution and temporal growth of defects displayed no evidence of the defects being caused by material degradation. Instead, high radiation appears to accelerate defect development in image sensors. It is important to detect these faulty pixels prior to the use of image enhancement algorithms to avoid spreading the error to neighboring pixels. The date on which a defect has first developed can be extracted from past images. Previously, an automatic defect detection algorithm using Bayesian probability accumulation was introduced and tested. We performed extensive testing of this Bayes-based algorithm by detecting defects in image datasets obtained from four cameras. Our results have indicated that the Bayes detection scheme was able to identify all defects in these cameras with less than 3% difference from visual inspected result. In this paper, we introduce an alternative technique, the Maximum Likelihood detection algorithm, and evaluate its performance using Monte Carlo simulations based on three criterias: image exposure, defect parameters and pixel estimation. Preliminary results show that the Maximum likelihood detection algorithm is able to achieve higher accuracy than the Bayes detection algorithm, with 90% perfect detection in images captured at long exposures (>0.125s).

Tradeoffs in Imager Design with Respect to Pixel Defect Rates

Previously we have shown that image sensors are continuously subject to the development of in-field permanent defects in the form of hot pixels. Based on laboratory measurements of defect rates in 21 DSLRs and 10 cell phone cameras, we show in this paper that the rate of these defects depends on the technology (APS or CCD) and on design parameters the like of imager area, pixel size, and gain (ISO). Comparing different sensor sizes has shown that the defect rate does not scale linearly. Comparing different pixel sizes has demonstrated that defect rates grow rapidly as pixel area shrinks. Finally, increasing the image sensitivity (ISO) causes the defects to be more noticeable, thus increasing the defect rate. These defect rate trends result in interesting tradeoffs in imager design, allowing the designer to determine the specific imager parameters based on the imager’s designated function and reliability requirements.

Tradeoffs in imager design parameters for sensor reliability

Image sensors are continuously subject to the development of in-field permanent defects in the form of hot pixels. Based on measurements of defect rates in 23 DSLRs, 4 point and shoot cameras, and 11 cell phone cameras, we show in this paper that the rate of these defects depends on the technology (APS or CCD) and on design parameters the like of imager area, pixel size, and gain (ISO). Increasing the image sensitivity (ISO) (from 400 up to 25,600 ISO range) causes the defects to be more noticeable, with some going into saturation, and at the same time increases the defect rate. Partially stuck hot pixels, which have an offset independent of exposure time, make up more than 40% of the defects and are particularly affected by ISO changes. Comparing different sensor sizes has shown that if the pixel size is nearly constant, the defect rate scales with sensor area. Plotting imager defect/year/sq mm with different pixel sizes (from 7.5 to 1.5 microns) and fitting the result shows that defect rates grow rapidly as pixel size shrinks, with an empirical power law of the pixel size to the -2.5. These defect rate trends result in interesting tradeoffs in imager design.

Analyzing the impact of ISO on digital imager defects with an automatic defect trace algorithm

Reliability of image sensors is limited by the continuous development of in-field defects. Laboratory calibration on 21 DSLRs has revealed hot pixels as the main defect type found in all tested cameras, with 78% of the identified defects having a time-independent offset. The expanded ISO range that exists in new cameras enables natural light photography. However, the gain applied to all pixels also enhances the appearance of defects. Analysis of defects at varying ISO levels shows that compared to the number of defects at ISO 400, the number of defects at ISO 1600 is 2-3 times higher. Amplifying the defect parameters helps differentiate faults from noise, thus detecting larger defect sets and causes some hot pixels to become saturated. The distribution of defect parameters at various ISO levels shows that the gain applied to faults with moderate defect magnitude caused 2-10% of the defects to saturate at short exposure times (0.03-0.5s). With our expanded defect collection, spatial analysis confirmed the uniform distribution of defects, indicating a random defect source. In our extended study, the temporal growth of defects is analyzed using our defecttracing algorithm. We introduce an improved defect model which incorporates the ISO gain, allowing the detection of defects even in short exposure images at high ISO and thus providing a wider selection of historical images and more accurate defect tracing. Larger area sensors show more hot pixels, while hot pixel rates strongly grow as the pixel size decreases to 2.2 microns.

Predicting Pixel Defect Rates Based on Image Sensor Parameters

Experimental measurements have shown that image sensors are continuously subject to the development of in-field permanent defects in the form of hot pixels. Based on measurements of defect rates in 23 DSLRs, 4 point and shoot cameras, and 11 cell phone cameras, we show that the rate of these defects depends on the technology (APS or CCD) and on design parameters the like of imager area, pixel size, and gain (ISO). Increasing the image sensitivity (ISO) (from 400 up to 25,600 ISO range) causes the defects to be more noticeable, with some going into saturation, and at the same time increases the number of defects. Partially stuck hot pixels, which have an offset independent of exposure time, make up more than 40% of the defects and are particularly affected by ISO changes. Comparing different sensor sizes has shown that if the pixel size is nearly constant, the defect rate scales linearly with sensor area, suggesting a measurement metric of defects/year/sq mm. Plotting this rate for different pixel and sensor sizes (from 7.5 down to 1.5 microns) shows that defect rates grow rapidly as the pixel size shrinks. Curve fitting shows an empirical power law with defect rate proportional to the pixel size to the power of-2.5. However, separating the pixel types shows that CCDs scale more slowly, with a power of-2, which translates into the pixel area. CMOS sensors, on the other hand, scale more rapidly with the pixel size to the power of-3.3. The result is that for 6-7 micron pixels the CCD defect rate is ~2.5 greater than the CMOS, but for 2 micron pixels the defect rates are both much higher and about equal. This paper presents a formula for predicting the expected rate of defect development for a given set of sensor parameters. This formula can be used by sensor designers when determining the imager parameters, taking into account the length of time the imager is expected to be in service.

Characterization of Gain Enhanced In-Field Defects in Digital Imagers

The quality of images produced by a digital imager is degraded by the presence of defects, mainly hot pixels, which develop continuously during the imager’s lifetime. We previously studied the spatial and temporal distributions of these defects (at ISO 400) and concluded that they most likely result from random radiation and are not material related. With the advancement in imaging technology, the noise level at high ISO had been overcome and new cameras have a wider ISO range (ISO 100-6400). ISO gain is applied to all pixels, good or defective; thus defect parameters get amplified, causing defects to become more visible at high ISO settings. Preliminary defect identification with high ISO has revealed 2 to 3 times more defects at ISO 1600 compared to the standard ISO 400 setting. Amplification of the defect parameters causes defects to become more distinguishable relative to the background noise level. In fact, by measuring the distribution of defect parameters, our experiment results suggest that 2-3% of the faulty pixels behave as stuck-high defects at ISO 1600. With more defects found at higher ISO, we gain a more complete map of defects from each sensor and thus improve our statistical analysis of the spatial and temporal defect distributions. Our current results show that although more defects were found in the tested sensors, the defects are very small and not clustered, pointing to a random defect source rather than a material related one.

A Fault-Tolerant Active Pixel Sensor to Correct In-Field Hot-Pixel Defects

Solid-state image sensors develop in-field defects in all common environments. Experiments have demonstrated the growth of significant quantities of hot-pixel defects that degrade the dynamic range of an image sensor and potentially limit low-light imaging. Existing software- only techniques for suppressing hot-pixels are inadequate because these defective pixels saturate at relatively low illumination levels. The redundant fault-tolerant active pixel sensor design is suggested to isolate point-like hot-pixel defects. Emulated hot-pixels have been induced in hardware implementations of this pixel architecture and measurements of pixel response indicate that it generates an accurate output signal throughout the sensors entire dynamic range, even when standard pixels would be otherwise saturated by the hot defect. A correction algorithm repairs the final image by building a simple look-up table of illumination- response of a working pixel. In emulated hot-pixels, the true illumination value can be recovered with an error of plusmn5% under typical conditions.

Characterization of pixel defect development during digital imager lifetime

The reliability of solid-state image sensors is limited by the development of defects, particularly hot-pixels, which we have previously shown develop continuously over the sensor lifetime. Our statistical analysis based on the distribution and development date of defects concluded that defects are not caused by single traumatic incident or material failure, but rather by an external process such as radiation. This paper describes an automated process for extracting defect temporal growth data, thereby enabling a very wide sample of cameras to be examined and studied. The algorithm utilizes Bayesian statistics to determine the presence and absence of defects by searching through sets of color photographs. Monte Carlo simulations on a set of images taken at 0.06 to 0.5sec exposures demontrating that our tracing algorithm is able to pinpoint the defect development date for all the identified hot pixels within ±2 images. Although a previous study has shown that in-field defects are isolated from each other, image processing functions applied by cameras such as the demosaicing algorithm were found to casue a single defective pixel to appear as a cluster in a color image, increasing the challenge pinpointing the exact location of hot defects.

Projecting the rate of in-field pixel defects based on pixel size, sensor area, and ISO

Image sensors continuously develop in-field permanent hot pixel defects over time. Experimental measurements of DSLR, point and shoot, and cell phone cameras, show that the rate of these defects depends on the technology (APS or CCD) and on design parameters like imager area, pixel size, and gain (ISO). Increased image sensitivity (ISO) enhances defects appearance and sometimes results in saturation. 40% of defects are partially stuck hot pixels, with an offset independent of exposure time, and are particularly affected by ISO changes. Comparing different sensor sizes with similar pixel sizes showed that defect rates scale linearly with sensor area, suggesting the metric of defects/year/sq mm. Plotting this rate for different pixel sizes (7.5 down to 1.5 microns) shows that defect rates grow rapidly as pixel size shrinks. Curve fitting shows an empirical power law with defect rates proportional to the pixel size to the power of -2.1 for CCD and to the power of -3.6 for CMOS. At 7um pixels, the CCD defect rate is ~2.5 greater than for CMOS, but for 2.4um pixels the rates are equal. Extending our empirical formula to include ISO allows us to predict the expected defect development rate for a wide set of sensor parameters.