输入缓存的交换结构具有实现复杂度小的优点,但存在排头阻塞现象。已经提出了解决方法均未达到理想的性能。本文提出用神经元网络实现输入缓存中的信元优化选择,它可在避免网络排头阻塞出现的同时,达到理想的网络性能。 Input caching switch fabric has the advantages of small complexity, but there is the first row of obstruction. Neither solution has been proposed to achieve the desired performance. In this paper, we propose a neuron network to optimize the cell selection in the input buffer, which can achieve the desired network performance while avoiding the emergence of the first block in the network.