Simulation And Study Of Power Consumption In LTP Process Based On PCNE Electronic Synaptic

The rapid development of artificial intelligence poses a severe challenge to the processing speed of computers based on the Von Neumann architecture,and has also stimulated research enthusiasm for artificial neural networks of humanoids.Among them,The key to constructing a neural network is to enable synaptic devices to operate at low power to achieve synaptic functions.Phase Change nanometer element(PCNE)have both energy accumulation effects and nano-integration characteristics,and is one of the most powerful competitors in electronic synaptic devices;however,the non-ideal transmission characteristics of synaptic devices,namely The non-linearity and asymmetry between long-term enhancement(LTP)and long-term inhibition(LTD)is the direct reason that hinders the improvement of the accuracy of neural network learning,and power consumption will restrict the large-scale integration of artificial neural networks.Therefore,combined with the non-linear transmission characteristics of electronic synapses,the simulation of the power consumption of the LTP process is of great significance to improve the performance of neural networks.In this paper,based on the working principle of PCNE,a physical model of PCNE three-dimensional structure and thermal,electrical,and phase simulation models are established on the MATLAB simulation platform.Secondly,the PCNE phase distribution and conductance change are adjusted by controlling the number of applied pulses.thus,the feasibility of PCNE as an electronic synapse was verified by simulation.Finally,the effects of unit structure parameters and pulse parameters on the non-linear transmission characteristics and power consumption of the PCNE electronic synapse LTP process were simulated.The parameters were optimized and pulse schemes were designed based on the simulation results.Simulation results show that the PCNE electronic synaptic weights are updated and stored by controlling the number of pulses,and the basic non-linear transmission characteristics of synapses are simulated,and the power consumption of synaptic weight updates is on the order of pico-watts.So the feasibility of PCNE as an electronic synapse was verified by simulation.;Secondly,when the effects of cell structure parameters and pulse parameters on the power consumption and non-linear transmission characteristics of electronic synaptic LTP processes were simulated,the smaller the GST-heater contact area,the larger the GST thickness.The lower the power consumption of the LTP process,and the nonlinearity of the G-N curve decreases first and then increases with the increase of the contact area and the thickness of the GST,and the thickness of the heater has almost no effect;the power consumption of the LTP process varies with the pulse amplitude and width Increased and increased,and the trapezoidal pulse will reduce the nonlinearity of the G-N curve,and the double pulse can effectively reduce the power consumption.