The implementation of 5G increases the demand for data acquisition, thus increasing the pressure of data processing. Although artificial neural network shows great potential in processing big data, efficient neuromorphic visual system is desired due to the waste of computation resources when processing non-structural visual data. Although reservoir computing (RC) has advantages in temporal information processing, the separation of sensors and RC results in addition cost. Here, an optoelectronic RC system is proposed for temporal information processing in sensors. The reservoir is built on photodetectors based on a non-uniform MoS2 film. The persistent photoconductivity effect of the photodetectors enables mapping different temporal inputs into corresponding reservoir states. The readout layer could be simply trained to identify different reservoir states. As a proof of concept, different classification tasks of numbers are demonstrated. The proposed optoelectronic RC system provides a low training cost strategy for intelligent edge machine visual system to process temporal information.
- edge computing
- low training cost
- neuromorphic visual systems
- Optoelectronic reservoir computing
- temporal information processing