The Study on IC implementation of dynamic memory using Linear Switched Diffusion Analog Memory

我们研制了一种用于集成人工神经网络的新型模拟存储器。已经进行了几次尝试来开发具有反馈电路的浮栅模拟存储器的线性特性。学习芯片必须有大量的电路。因此，我们提出了一种新的模拟存储器具有设计简单，占地面积小，开关速度快，线性度准确。为了提高准确的线性度，我们提出了一种新的电荷转移过程。该器件具有隧道结（多晶硅/多晶硅氧化物/多晶硅三明治结构）、薄膜晶体管（TFT）。该操作使得转移的电荷量与存储电容中的电荷无关，并且研究了具有局部连接的神经网络的性质。每个神经元从具有局部邻域的其他神经元接收输入并向其发送输出。首先提出了一种基于神经网络运动方程的力代替系统能量函数的连续时间模型下的网络动力学分析方法。在连续时间模型中，该方法不需要求解所有初始条件下的微分方程，就能方便地分析网络的动态行为。利用这种方法，我们对具有循环连接和非对称连接的连续时间神经网络的平衡点和极限环的类型进行了实验研究，并对某些类型的平衡点和极限环进行了理论分析。讨论了极限环的性质与连通数的关系。

We have fabricated a new analog memory for integrated artificial neural networks. Several attempts have been made to develop a linear characteristics of floating-gate analog memories with feedback circuits. The learning chip has to have a large number of circuits. Therefore, we proposed a new analog memory has a simple design, a small area occupancy, a fast switching speed and an accurate linearity. To improve accurate linearity, we propose a new charge transfer process. The device has a tunnel junction (poly-Si/poly-Si oxide/poly-Si sandwich structure), a thin-film transistor (TFT). The proposed operation is possible that the amounts of transferred charges are constant independent of the charge in storage capacitor.And we have studied properties of neural networks with local connections. Each neurons receives input from and send output to other neurons with a local neighborhood. First a method to analyze the dynamics of networks in continuous-time model based on the force derived from the equation of motion for neural networks instead of energy function of the system has been proposed. In continuous-time model, the dynamic behavior of network can easily be analyzed by using this method without solving differential equations for all initial conditions. By using this method, we investigated experimentally the types of equilibrium points and limit cycles of continuous-time neural networks with cyclic and asymmetric connections and analyzed theoretically some types of equilibrium points and limit cycles. We also discussed the relation between the properties of limit cycles and the number of connections.