Synaptic transistor with multiple biological functions based on metal-organic frameworks combined with the LIF model of a spiking neural network to recognize temporal information

Abstract

Spiking neural networks (SNNs) have immense potential due to their utilization of synaptic plasticity and ability to take advantage of temporal correlation and low power consumption. The leaky integration and firing (LIF) model and spike-timing-dependent plasticity (STDP) are the fundamental components of SNNs. Here, a neural device is first demonstrated by zeolitic imidazolate frameworks (ZIFs) as an essential part of the synaptic transistor to simulate SNNs. Significantly, three kinds of typical functions between neurons, the memory function achieved through the hippocampus, synaptic weight regulation and membrane potential triggered by ion migration, are effectively described through short-term memory/long-term memory (STM/LTM), long-term depression/long-term potentiation (LTD/LTP) and LIF, respectively. Furthermore, the update rule of iteration weight in the backpropagation based on the time interval between presynaptic and postsynaptic pulses is extracted and fitted from the STDP. In addition, the postsynaptic currents of the channel directly connect to the very large scale integration (VLSI) implementation of the LIF mode that can convert high-frequency information into spare pulses based on the threshold of membrane potential. The leaky integrator block, firing/detector block and frequency adaptation block instantaneously release the accumulated voltage to form pulses. Finally, we recode the steady-state visual evoked potentials (SSVEPs) belonging to the electroencephalogram (EEG) with filter characteristics of LIF. SNNs deeply fused by synaptic transistors are designed to recognize the 40 different frequencies of EEG and improve accuracy to 95.1%. This work represents an advanced contribution to brain-like chips and promotes the systematization and diversification of artificial intelligence.

Keywords: Electronic devices; Electronic properties and materials.

Fig. 1. Human brain processing complex information, the relationship between biological and electrical synapses, and the basic characterization of synaptic transistors.

a Schematic diagram depicting the complex information encompassed by the analog signals. b The human brain processes high-frequency information through sparse representation to obtain filtered low-frequency information. c Low-energy operating modes in the human brain are facilitated by neurons connected through different weighted nodes. d Structure diagram of the synaptic transistor and basic crystal of ZIF-67. e A biological chemical synapse is illustrated schematically, comprising a presynaptic terminal, a receptor, and a postsynaptic terminal. f Various functional areas in the human brain are composed of billions of neural networks. g SEM images of ZIF-67. h XRD patterns of ZIF-67. i Channel current dependence of the gate voltage analyzed at a V_post of 0.5 V

Fig. 2. Three biological functions and corresponding synaptic characteristics.

a The device simulates three typical functions of synapses: forming memory, synaptic plasticity, and stimulating membrane potential. b The PPF index is a measure of synaptic facilitation defined as the ratio of the amplitudes of the first (A₁) and second (A₂) EPSCs plotted against the pulse interval ($\mathrm{\Delta }\mathrm{t}$). c EPSC behaviors activated and modified by electric pulses with 9 different widths (50 ms, 100 ms, 150 ms, 200 ms, 250 ms, 300 ms, 350 ms, 400 ms, and 450 ms) at V_DS = 0.5 V. d EPSC triggered by 5 single electric pulses with different amplitudes (3 V, 4 V, 5 V, 6 V, and 7 V). e Low-pass filtering characteristics are shown by 10 continuous pulses of different frequencies (10 Hz, 12.5 Hz, 20 Hz, 30 Hz, and 40 Hz) applied to the presynaptic terminal. f The LTP/LTD characteristics demonstrate the controllable range and level of conductance. g Threshold effect of the synaptic device as biological neurons. h Trend of the STDP curve as the weight update rule transforms the temporal information

Fig. 3. SNN framework and LIF model combined with EPSC.

a Block diagram of feedforward and back propagation for the spiking neural network based on the LIF neuron model and STDP weight update rule. b Fluctuation area of membrane voltage in biological neurons. c The leaky integrator block, fire and detector block, buffer block, and frequency adaptation block are combined with the output of ZIT-67 synaptic devices to form the complete LIF system. d Operation of the VLSI circuit with input pulses of the synaptic device (t_on = 100 ns, t_rise = t_fall = 1 ns, period = 2 μs, I_in = 2.0 mA and 3.3 mA)

Fig. 4. The SSVEP identification task is based on the modified SNN.

a Steady-state visual evoked potentials (SSVEPs) are a favored signal in brain-computer interface (BCI) systems due to their high information transfer rate (ITR). b Eight types of electroencephalographic (EEG) waveforms from 40 different frequencies in the same channel. c SSVEP is processed into a visualized two-dimensional matrix through temporary coding. d Synaptic efficacy (conductance of channel), defined as the strength of the communication between neurons, undergoes temporal modulation characterized by both facilitatory and inhibitory changes. e Inner state of neurons based on STDP in SNNs. f Output spike of neurons in SNNs. g Test accuracy of the recognition task for SSVEPs in the improved network