Why do neurons generally behave as nonlinear devices?

In the system composed of traditional devices, with the increase of synapses or neuron units, the number of transistors required increases exponentially, which seriously restricts the rapid development of brain-like morphological computing chips. Therefore, from the perspective of the realization of human brain neuron function, it is expected to build a real "electronic brain" by using a single transistor to obtain nonlinear computing power and realize intelligent computing with high performance and low power consumption.

In view of the huge demand for brain-like morphology technology, Zhou Peng and Hu introduced the unique non-doped polarity characteristics of two-dimensional materials, and the new neuron-like logic transistor proposed showed great application advantages at the device and system level: single transistors with different polarities (bipolar tungsten selenide, N-type molybdenum sulfide and P-type black phosphorus) can simulate neuron cells to realize complete Boolean logic operation; Based on the combination of different new devices, logic circuits with high area efficiency can be further constructed, and the maximum physical area can be saved by 78%. At the same time, the new neuron-like logic device can construct a three-dimensional XOR logic array and apply it to the binary convolutional neural network (BCNN). The simulation results show that the computational efficiency of the network has surpassed that of BCNN based on memristor storage and computing integration technology.