Analysis and application of large chaotic neural networks

利用开关电容电路技术将混沌神经元模型实现为集成电路形式。该模型包含了生物神经元的重要特征——轴突丘锁的相对耐受性和模拟输出函数。集成混沌神经元电路具有电流输入节点，便于增加多个输入电流。为此专门设计了一种电流-电压转换电路。此外，还将突触电路集成为集成电路，通过连接多个神经元电路来构建大规模的混沌神经网络。在突触芯片内，集成了许多电流模式突触电路和突触权重的模拟存储器。模拟存储器由主电流复制电路周期性地刷新。利用上述神经元芯片和突触芯片，构建了一个小规模的全功能混沌神经网络。利用以下三个基本问题对网络特性进行了测试；动态联想记忆：从实验中，一种像动物一样的动态记忆检索已经被证实。优化问题：以一个旅行商问题为例，利用网络求解了该问题。结果表明，该网络具有很高的求解能力。混沌同步：实现了网络中某些混沌的同步运动。所实现的混沌神经网络系统的行为可以实时观察。因此，通过构建更大的网络，可以从动力学理论或信息论的角度来分析耦合混沌系统的信息处理过程。网络规模的扩大和多阶并行观测系统的构建是未来需要解决的问题。

A chaotic neuron model has been implemented as an integrated circuit (IC) form using switched-capacitor circuit technique. The model includes a relative refractoriness and an analog output function of the axon hill-lock which are important characteristics of biological neurons. The integrated chaotic neuron circuit has current input nodes in order to add many input currents easily. A current to voltage converter circuit was specially designed for this purpose.Furthermore, a synaptic circuit has been integrated as an IC form in order to construct a large scale chaotic neural network by connecting many neuron circuits. Within the synaptic chip, many current-mode synapse circuits and analog memories for synaptic weights are integrated. The analog memories are periodically refreshed by a master current copier circuit.By using the above neuron chips and synaptic chips, a small-scale but full-functional chaotic neural network has been constructed. The network characteristics have been tested using the following three basic problems ;1.Dynamical Associative Memory : From experiments, a dynamical retrieval of memories like animals do has been confirmed.2.Optimization Problems : As an example, a traveling salesman problem has been solved using the network. As a result, very high solving ability of the network has been confirmed.3.Synchronization of Chaos : synchronous motion of some chaos in the network has been achieved.The behavior of the implemented chaotic neural network system can be observed in real-time. Therefore, it is possible to analyze the process of information processing of a coupled chaotic system from the view point of dynamics theory or information theory by constructing a larger network. The expansion of the size of the network and construction of a multi-order parallel observation system are future problems.