The Pixel Design Of Silicon Image Sensor By Imitation Of Retinal Cells

The artificial vision processing system is one kind of systems, which has a processing unit with separated photodiode. It's difficult to improve the current artificial vision system perfomence because of the increasing amount of image algorithm computation with high frame rate and resolution. Many researchers begin to emulate the retina cells with the in-pixeled processing unit in image sensor. The Silicon retina divided into two modules. One realizes the main function of several retina cells with the traditional APS image sensor, another is focused on modeling one vision imformation channel of all channels in retina. Dynamic vision sensor(DVS) image sensor, for example, responds asynchronously to relative temporal changes in intensity.Recently, the vision chip has been a hot topic due to the high speed and small data redundancy of DVS based image sensors. However, the frame-free image sensors are usually suffered from sporadic noise events. For reducing the noise, a structure is proposed to remove background activities based on the noise analysis spatially. The proposed pixel, connected with four neighor pixels, emulate the mechanism of horizontal cells for filtering background activies. A novel 5×5 shared connection type is used to reduce the size of pixel array. Besides,analog mixed signal, isolation protection for steady signal, antenna effect and matching problems are taken into account to improve stability in layout. The proposed pixel is simulated in 0.13?m SMIC CIS technology and the thesis gives the performance comparison of pre-simulation and post-simulation. The layout size is 23.76×24.71?m~2. The proposed pixel adds less than 3 ns latency to the DVS circuit and the behavioral test model of the pixel array shows that it can filter out 93% of the background noise events.The pixel is designed to comunicate with four neighor pixels and the idea comes from the work mechanism of horizontal cells in retina. The results of simulation and test model show that the proposed pixel is able to reduce a lot of background noise events in DVS image sensor. Whats more, this achievenment provides a promising method for tracking and recognition algorithm in DVS based image sensors.