A CMOS Vision Sensor Based On Address Event Representation

Based on the transmission mode of frame, traditional vision sensor has excelledin small pixel area, high resolution, high fill factor and low production cost. But withthe increase of ten million level pixels in demand, and development of CMOSprocesses, the development of traditional vision sensor is limited in large data transfer,small frame frequency and low DR. In order to achieve high frame frequency, highDR and low data quantity of visual information accurately, a vision sensor based onaddress event representation is proposed.Based on the basic principles of AER, the theory how AER vision sensor caneliminate redundant data from source, realize the "no-frame" output is presentedfirstly. Secondly, double sampling PWM circuit, the core of the system, is studied byworking principle, noise elimination mechanism. A time error mode is establishedwhich can help to choose appropriate parameters. Then, system architecture of visionsensor based on AER is established. By behavioral model of pixel architecture,arbitration module and system architecture, the feasibility of the system is proved. Atlast, circuit-level design, including the design of CDS PWM circuit, change detectorcircuit and arbitration circuit is implemented. The simulation results of the circuits arealso given.The AER vision sensor in the paper applies a128×128pixel array. The pixel candetect and quantify the light intensity changes by change detector circuits and doublesampling pulse width modulation circuits. The sensor works by event trigger,asynchronous, so that redundant information can be eliminated from source, DR andframe frequency increase. The output can be decreased by arbitration function andtime delay also decreases. The simulated results show that the minimum equivalentframe frequency is100fps to1000fps. DR can reach133dB. Maximum time delay ofarbitration circuit is160.65ns. It proves that the method proposed can performreal-time optical signal capturing, processing and readout; as well as be applied to theflied of high-speed and high DR image and vision.