Fig. 1: Leaky integrate-and-fire dynamic of biological neuron vs AFeFET artificial neurons.

a The architecture and processing model of biological neurons. External information is perceived by dendrites and then transmitted to the soma for processing. When the membrane potential surpasses a threshold value, an action potential can be generated and transmitted to post-neurons through the axon. b The integrate-and-fire dynamics of a biological neuron. With the signal received by dendrites and transmitted to soma, few Na⁺ channels are activated, and the membrane potential goes up gradually (stage 1). As further signal is received, more Na⁺ channels are activated quickly, and a large amount of Na⁺ flows inside the membrane, inducing the membrane potential goes up rapidly (stage 2). Once the membrane potential exceeds a certain threshold, it decreases due to the inactivation of Na⁺ channels and the opening of K⁺ channels (stage 3). c The representative double hysteresis of AFE materials. The zero net macroscopic remanent polarization indicates the volatile characteristics of AFE materials. d The typical transfer curves of AFeFET, exhibiting counterclockwise hysteresis and volatility. The arrows indicate the switching sequences. e The continuous firing events of AFeFET neuron. After several milliseconds of free time, the AFeFET neuron can restart the LIF process and fire again.