چكيده به لاتين
Neuromorphic engineering means designing electronic devices which are inspired by human brains. The neuromorphic based architectures are implemented to accelerate machine learning computations, artificial intelligence, and cognitive tasks. Neuromorphic systems are more efficient compare to the Von-Neumann architectures from the aspect of performance. These architectures consist of distributed elements such as neurons and distributed memories (synapses). By combining these processing elements, spike neural networks are implemented. One of the most important cognitive operations in the human brain is associative memory. This memory is able to store a large amount of information and recover it with minimum error. Various architectures and designs for neural associative memory have been introduced, such as Hopfield neural network (HNN). However, these architectures suffer from low storage capacity. In this research, a new spiking associative memory architecture proposed that is called COM. COM has a large capacity and robust message retrieval compared with HNN. In this thesis, a low-cost Leaky Integrate and Fire (LIF) neuron is proposed that is used to implement a digital spiking WTA module. The results demonstrate that the proposed neuron has been improved by almost 39% in response speed, and 50% in resource-consumption compared to the recent works. These neurons are used to implement a spiking WTA neural network. A WTA has high noise robustness characteristics. A set of random is applied to the proposed dataset to validate the operation and accuracy of spiking WTA. This architecture improves the accuracy of the spiking neural networks by 27 percent. The maximum operation frequency of the implemented neuron and WTA neural network are 576.319 MHz and 314.095 MHz, respectively. Finally, a brain-inspired associative memory is implemented on FPGA, which is called COM. It consists of three WTA modules. It's highly robustness against the noises, and it provides a large capacity to retrieve the erased patterns.
