چكيده به لاتين
Nowadays, neural network due to its considerable characteristics such as learning ability, capability to approximate nonlinear function and independency to mathematical model attracts researchers attention in variety fields of science. These network are employed for different applications such as pattern recognition, system identification, control system and etc. Neural networks is usually implemented via software in computer. However, hardware implementation of them has been the considerable challenge in recent years. In hardware implementation, the weights of neural network is always constant and their regulation needs chip compiling. The design of a neural network using a system of memories or memristors can be a solution to this problem. Nonlinear and time invariant behavior of resistance in these systems can lead to the creation of a neural network that can be adapted to different operating conditions. In this thesis, a novel scheme for hardware implementing of the neural network based on the emergence of the memristor element is presented. In the proposed method, Memristor is used to implemented switch and synapse performance, which, in addition to being inactive, also occupies a very low level of consumption. Furthermore, hardware in loop structure is considered to train the neural network. In order to realize training process, two soft wares called HSPISE and MATLAB are connected. The proposed scheme is simulated to cluster the pendulum system pattern and the simulation results confirm the effectiveness of this method.
