Design And Application Of Neuron Network Based On TTA

Artificial neural network is the mathematical model of human brain.It is also an information processing system based on the structure and function of brain.It is actually a complex structure consisted of a large number of neurons connected to each other.Today,artificial neural networks have been widely used in signal processing,medical,control system,business,pattern recognition,speech recognition and other fields.In general,the implementation of artificial neural network is based on the software,but it can't do real-time operation in many cases.This paper designed an artificial neural network hardware system,which can improve the running speed of neural network.Author also innovatively gave a programmable neural network architecture based on transmission triggering architecture(TTA),which can reflect the inherent parallel feature in artificial neural network.The basic idea of the design is to see the neurons as functional units in TTA,and to transmit the different data to the neuron unit to program the neural network.This implementation will make the processor structure simple,parallel,scalable,and will make its instruction set less.The neural network processor is configurable machine,because it can build a feedforward neural network with any number of layers which can contains any number of neurons.The processor must get the configuration constants off-line firstly.Then we must import these constants to the processor.After those,the system can run.For the learning algorithm,it uses a combination of supervised and unsupervised learning algorithms.The paper introduced the development and background of the artificial neural network firstly.Then we analyzed the theoretical basis of it,including the neuron model,the network structure and the learning algorithm.Secondly,a single neuron was designed on FPGA based on IBM neuron.Then,the hardware system of the artificial neural network was described in three parts,including instruction,connection and functional unit.Furthermore,a digital recognition application ran on the machine to test its performance.Finally,the paper summarized the work,and gave some problems that had to be solved in the future.