Basic Research on Optoelectronic Devices for Neural Networks

神经网络光电器件基础研究

基本信息

批准号：
01460082
负责人：
YONEZU Hiroo
金额：
$ 3.65万
依托单位：
Toyohashi Institute of Technology
依托单位国家：
日本
项目类别：
Grant-in-Aid for General Scientific Research (B)
财政年份：
1989
资助国家：
日本
起止时间：
1989 至 1990
项目状态：
已结题

项目摘要

A basic study on neuro-devices and optical interconnection has been conducted for large-scale neural networks in the viewpoint of an optoelectronic integrated circuit (OEIC). The fundamental feasibiity and technological barriers were investigated.Optical outputs from other neurons were multiplied by synaptic weights and summed in the artificial neuron. Then the optical output was emitted with the sigmoid function, which is easily generated by an LED connected to a differential amplifier. The function of synaptic connection is most important in the neuron. Firstly, an optoelectronic circuit was developed, which acted as a synaptic connection with variable synaptic weights. As a result, the basic function as well as the preferable performances of the optical interconnection were clarified. Then a new device with a self-organizing function was developed, whose synaptic weight is adapted to a teacher's signal. These optoelectronic fundamental device and circuits were fabricated using Si IC technology. Preferable characteristics for large-scale integration were clarified such as a very low current operation of the order of nA and a simple structure.Initial growth mechanisms were also studied in epitaxial growth of GaAs on Si, which could be needed for one-chip OEIC. The initial three dimensional growth must be avoided essentially for a growth of high quality GaAs on Si. A quasi-two dimensional growth was realized by introducing an AlAs initial layer having a stronger chemical bond or a thin GaAs layer grown by solid phase epitaxy. In addition, a mirror surface was obtained in homoepitaxy of GaAs(111)B and (110) by enhancing a Ga-migration.

在光电集成电路（OEIC）的角度，已经针对大规模神经网络进行了有关神经设备和光学互连的基本研究。研究了基本的流行性和技术障碍。其他神经元的光输出乘以突触权重，并在人工神经元中求和。然后使用Sigmoid函数发出光学输出，该功能很容易由连接到差分放大器的LED生成。突触连接的功能在神经元中最重要。首先，开发了光电电路，该电路充当带有可变突触权重的突触连接。结果，阐明了光学互连的基本函数以及优选的性能。然后开发了具有自组织功能的新设备，其突触重量适合教师的信号。这些光电基本设备和电路是使用SI IC技术制造的。阐明了大规模整合的优选特征，例如NA阶的电流非常低和简单的结构。还研究了SI上GAA的外在生长的生长机制，这对于单芯片OEIC可能需要。对于SI上的高质量GAA的增长，必须避免最初的三维增长。通过引入具有更强的化学键或通过固相外延生长的薄的GAAS层的Alas初始层来实现准二维生长。此外，通过增强GA迁移，在GAAS（111）B和（110）的同性恋中获得了镜面。