Development of Knowledge Acquisition and Automatic Layout System for Printed Wiring Board Design

我们开发了一个自动布局系统，它从熟练的设计者的图纸中获取布局知识，并利用这些知识来获得满意的布局结果。熟练的设计者的布局策略的特点是“簇”(电路划分)、“簇布局”和基于总线法。我们的系统从熟练的设计者的布局图中获取关于聚类和簇布局的知识，并检查与已经确定的规则的矛盾。在每个规则的条件部分，引入一个带预测变量的线性不等式来判断所选择的候选对象是否适合作为对象。由于线性不等式中的系数和常数是使用判别函数法自动确定的，因此执行来自不同板的规则之间的冲突消解。结果表明，系统布局结果与人工设计结果的相似度随着用于知识获取的板数的增加而增加，并提出了一种考虑包括未布线线网在内的所有线网影响的公交线网布线分层布线方法。为了提高路由算法的处理速度，我们对几种多处理体系结构进行了评估，其中包括树形网络的超立方体体系结构是最适合实现的。

We have developed an automatic layout system, which acquires layout knowledge from skilled designers' drawings and makes use of it to get satisfactory layout result.Characteristic points in skilled designers' layout strategies are in "clustering" (circuit partitioning), "cluster placement", and bus based methods. Our system acquires knowledge for clustering and cluster placement from skilled designers' layout drawings, and checks contradictions with already fixed rules. In the condition section of each rule, a linear inequality with predictor variables is introduced to judge whether the selected candidates are suitable as an object. Since the coefficients and constants in the linear inequality are determined automatically employing the discriminate function method, conflict resolution between rules from different boards are performed. As a result, the similarity between the system layout result and the manual design has increased according to the number of boards used for knowledge acquisition.Furthermore, we have developed a hierarchical routing method for bus net routing and for taking influences of all nets including unrouted nets into consideration. For high speed processing of the routing algorithm, we have evaluated several multiprocessing architectures, and the hypercube type architecture including tree network has turned out to be the best for implementation.