Dynamic Vision Sensor Design Based On SMIC 0.13um Process

Traditional CMOS solid-state image sensors have limited the algorithm efficiency of moving objects recognition and tracking with their large redundancy and low temporal contrast.It was caused by the fixed timing and frame rate in conventional CMOS image sensors.On the contrary,the retinal nerve system in higher species could quickly response to the illumination change.Their bipolar cells and ganglion cells then would output a pulse signal to represent these changes.These signals would be transferred to the brain.Finally the brain would efficiently have a sense of dynamic information in the scene.The retinal nerve system has a high temporal contrast and can ignore the redundancy when tracking moving targets.This paper aimed at the design of a bionic dynamic vision sensors(DVS)based on the working mechanism of retinal cells and modern silicon process.This paper firstly talked about the DVS pixel's bionic model and took an introduction of processing circuit and photodiode in pixel.The device simulation tool would obtain the basic parameters of photodiode.Then the equivalent pixel's schematic and layout were verified by SMIC 0.13 um process design kit in circuit simulation tool.Secondly,a DVS 8×8 array frame was designed with the principle of multiplexing in Address Event Representation(AER)protocol.There had a verification of each module and the array.Thirdly,as the SMIC 0.13 um traditional image sensor process is confidential,a new DVS process was created with the circuits design.The new process combined the SMIC 0.13 um low leakage logic process with the created photodiode process.There had a detail presentation about this new process.Finally,there was a test and analysis of the DVS pixel tape-out results.The DVS pixel structure which based on the SMIC 0.13 um process could output ON or OFF event signal when the light current intensity had changed stronger or weaker.The pixel would have no response until the intensity had changed more than 20% relative to itself.The temporal contrast and power consumption was about 0.5us and 42 u W.The 8×8 array could fairly output the address of each simultaneous event with one channel and had a latency of 11 us.The created DVS process could be successfully used in the foundry and finished the first tape-out work in research group.And this process could provide a technology basis for the next tape-out of bionic sensors.The tape-out pixel initially realized the function of photoreception and photoelectric conversion.And the first test edition of DVS pixel could provide a good experience for the next stage of design and test.The DVS pixel also offered a pixel process model for industry research in bionic vision sensors.The design of this paper had an order from the photosensitive device,pixel circuits,array structures and manufacture process to the final test.It was a relatively complete work.