A study of brain-like integrated systems using nonlinear dynamics with pulse phase

1.脉冲定时非线性动力学图像处理大规模集成电路的研制采用模数合并/混合结构设计了一种振荡器网络大规模集成电路，采用振荡相位同步进行图像区域分割和提取。在该大规模集成电路中，非线性动力学的电路模块由脉冲调制电路组成，权重存储和控制部分由数字电路组成。脉冲定时类脑信息处理电路算法的开发通过将尖峰神经元模型(集成与触发类型)应用于反馈网络，我们可以更快地运行网络。基本概念是，较早输入的信号被认为更重要。为了区分反馈信号和晚输入信号，引入了一种具有斜坡抑制的全局抑制神经元和神经元放电时的重置机制。对具有20个神经元的Hopfield网络进行联想记忆运算的电路仿真结果表明，该电路的收敛速度是传统脉宽调制电路的20倍。针对后CMOS时代的一种电路技术，提出了一种基于多纳米点MOSFET的尖峰神经元电路。该电路可以更有效地产生突触后电位，并且具有超低的功耗。纳米点结构将通过使用纳米技术的自组装过程来构建。

1. Development of an image processing LSI using nonlinear dynamics with pulse timingAn oscillator network LSI that performs image region segmentation and extraction using oscillation phase synchronization was designed using analog-digital merged/mixed architecture. In this LSI, the circuit block for nonlinear dynamics consists of pulse modulation circuits, and the weight memory and control part consists of digital circuits.2. Development of an algorithm for brain-like information processing circuits using pulse timingBy applying a spiking neuron model (integrate-and-fire type) to feedback networks, we can operate the networks faster. The basic concept is that signals inputted earlier are considered more important. To discriminate between feedback signals and late-inputted signals, a global inhibitor neuron with ramped inhibition and a reset mechanism at firing in a neuron are introduced. From the circuit simulation results of associative memory operation in Hopfield networks with 20 neurons, it was confirmed that the proposed circuit converges 20 times faster than the conventional pulse-width modulation circuit.3. Design of a spiking neuron circuit with multi-nanodot MOSFETsFor a circuit technique in the post-CMOS era, a spiking neuron circuit with multi-nanodot MOSFETs is proposed. The circuit can generate a post-synaptic potential much more efficiently and with ultra-low power dissipation. The nanodot structure will be constructed by using self-assembly processes using nanotechnology.