Research Based On Event-driven Vision Sensor

With the development of integrated circuits and semiconductor technology,CMOS image sensors are widely used in mobile phones,automotive electronics,medical,aerospace and other fields.But with the advent of the artificial intelligence era,higher requirements have been placed on image sensors.Traditional CMOS image sensors use a “frame scan” method to process images.When the pixel array is enlarged and the accuracy is increased,it will cause a large amount of data redundancy,putting pressure on data transmission and storage,and limiting its efficiency.In order to overcome the above drawbacks,this paper combines a traditional CMOS image sensor with a biological vision system and designs an event-driven vision sensor.Eliminate the "time-driven" working mode of the "frame scanning" working mode,and replace it with a more flexible and intelligent "event-driven" mechanism.The so-called "event-driven" mechanism is essentially an asynchronous system that works in real time.Its principle is similar to that of the human eye.When the light intensity received by a pixel changes,an electrical signal is generated.This electrical signal is regarded as an "event".In order to read this "event" information,this design introduces the use of Address-Event Representation(AER)when transmitting signals between neuromorphic systems.After the pixel that generated the "event" sends a request to the bus,the address of the pixel unit that generated the pulse and the attributes of the event can be read through the AER communication system.Different from the traditional image sensor,each pixel in this design can be self-reset after being read without the need for uniform reset.Because multiple pixels can have light intensity changes at the same time,in the absence of a unified clock,an arbitration mechanism needs to be introduced to sort the readout order of pixels that have light intensity changes.To this end,this design proposes an arbitration circuit that implements fair arbitration based on cyclic priority,which can ensure that each pixel is read with equal opportunity,thereby helping the entire asynchronous system to work in an orderly manner.This paper implements an 8?8 pixel array event-driven vision sensor on the Cadence(Virtuoso)platform using SMIC 0.18?m technology and simulates it.The simulation results are in line with the design expectations,the circuit function is correct,and the pixel unit where the light intensity changes occur can be asynchronously read out,and the system delay is 9.19 ?s.This research belongs to the field of vision chips.The design reduces the amount of information from the source,avoids data redundancy,and improves the sensor performance while reducing overall power consumption.With its further development,combined with artificial intelligence and machine vision,it has broad prospects in high real-time application scenarios such as driverless,smart medical,and security monitoring.