使用相位型阵列分束器作为互连元件的可编程光学神经网络

本文提出了一种２－Ｄ可编程神经网络的新装置，在这种装置中，采用透镜和一块相位阵列分束器作为光学互连元件，由于用一块分束器代替了子透镜阵列，所以克服了透镜阵列加工和装校的困难，以及因子透镜像差和通光孔的衍射对提高空间信道容量密度的严重限制，从而使系统具有增大信道容量的潜力。

A PROGRAMMABLE OPTICAL NEURAL NETWORK USING A PHASE ARRAY BEAM SPLITTER AS THE INTERCONNECTION DEVICE

In this paper,a new architecture of two-dimensional programmable optical neural network is presented,in which a phase array beam splitter is introduced in the interconnection device. It has advantages over the device using a lenslet array that the difficulties of both fabricating and adjusting the lenslet array can be avoided.The capacity of spatial channels of the system can be increased by overcoming the restrictions in aberrations and diffraction of the lenses.