The Institute of Microelectronics of the Chinese Academy of Sciences has made important progress in the field of memristor-based sensory computing

Recently, the team led by Academician Liu Ming from the Key Laboratory of Microelectronics, Institute of Microelectronics, Chinese Academy of Sciences, proposed a solution to build an artificial sensory neural system with habituation characteristics based on memristors, and used habituation, a biological learning rule, to build a habituated pulse neural network that can be used for autonomous robot cruising and obstacle avoidance.                              

The research team built a sensory neuron based on Mott memristors and sensors. The neuron can sense external analog signals and convert them into real-time dynamic pulse signals, realizing the basic functions of sensing and transmitting external signals. The sensory neuron is further connected to the relay neuron through synaptic devices to build a habituated perception system.

The synaptic device has a habituated evolution trend of weights under continuous stimulation, which in turn affects the efficiency of the transmission of sensory neuron signals to relay neurons, making the output of relay neurons show a frequency reduction characteristic (i.e., habituation characteristic, as shown in Figure a). Based on this habituation characteristic, the team further constructed a habituated pulse neural network to realize the robot's obstacle avoidance function. The test results show that the memristor-based artificial sensory neural system constructed based on habituation learning rules can effectively improve the robot's obstacle avoidance efficiency. The habituated sensory neural system can also be applied to different perception systems such as smell, taste, vision, hearing, etc. through different sensors. By realizing a biologically realistic perception system, it is expected to achieve a more biologically intelligent terminal system.