نرگس احمدي دستجردي

In this thesis, an improved model of the Izhkevich neuron was presented an‎d implemented using memristor equations. This model is able to provide 23 responses from the biological responses of the brain, which are derived from the responses of the HRL memristor neuron model. Since our goal is to implement a memristor neuron for use in a neural network, we have implemented it on a digital system. Then the results of the implementation were compared with the MATLAB simulation model of the neuron an‎d it was observed that the results of the implementation are consistent with the simulation. The said neuron called the Izhkevich memristor, is implemented using various combinational blocks an‎d fully synchronous flip-flops with a 100 MHz clock signal an‎d using right-shift an‎d left-shift instead of the DSPʹs internal multiplier. In addition, the system implementation has been done using high-level hardware description language (VHDL). Finally, by using the table of coefficients obtained from MATLAB neuron model simulation, 12-bit stream input an‎d removing the multiplier, we reduced the consumption level of the implementation as much as possible. In the best case of neuron implementation, 159 flip-flops are used in the PX7AL01 board with Artix7 chip an‎d 152 flip-flops in Posedge board with spartan6 chip. This circuit is described as RTL an‎d the dynamic power consumption is 11 milliwatts an‎d the maximum working frequency is 107/64 MHz. Finally, for the optimal use of neurons in neural networks an‎d the maximum utilization of the resources available in the hardware, it is recommended to implement it on an Artix7 chip with 240 multipliers. which allows 120 digital neurons to be implemented simultaneously. This research will pave the way for the implementation of digital neural networks with higher productivity an‎d less resource consumption.

نورون , پياده سازي

/neuron , HRL

Narges ahmadi

Mohammad azim karami