Perovskite-based optoelectronic systems for neuromorphic computing

Abstract

Brain-like neuromorphic computing system offers the ability to be used in neural networks on demand and are rapidly gaining interest by researchers due to their outstanding photoelectric stimulation response-ability and multi-segment storage function. In this work, we show the processing capabilities of the neuromorphic computing units for optoelectronic signals and high label noise ratio signals. Demonstrated the functions of potassium ion doped perovskite-based unit resembles properties found in neuronal synapses. Potassium ion passivates perovskite and promotes the formation of the electric double layer and ion doping in the unit so that the unit can simulate the characteristics of biological synapses generated by various electrical simulations according to the demand. Perovskite significantly enhances the unit's responsiveness to light pulses in the visible light range. By implementing the 'resolving training biases via influence-base data relabeling' strategy, the unit demonstrates remarkable performance in high noise recognition tasks. These results open a new direction for the construction of brain-like computers that deal with a complex and high proportion of harmful samples.