Configuration and Reasoning Functions for Knowledge-based VLSI Pattern Design System

我们从基于知识的系统结构和基于该结构的智能功能的角度对先进的VLSI-CAD系统进行了研究。为基于知识的CAD系统。除了目前与现有知识系统相关的分析或验证功能之外，还需要自动或半自动的综合功能。虽然有几种方法可以实现基于知识的大规模集成电路模式设计系统，但我们已经构建了一个具有功能单元库作为内核的原型系统。功能单元库基于知识结构，存储一组智能单元数据（组件），以响应请求进行组合。我们的LSI功能单元库系统名为OCLS，目前可以支持基于Mead-Conway设计方法的NMOS艺术品图案设计。开发了一种适合于分层模式描述、使用变量表示几何模式以及嵌入相关知识的模式描述语言。该语言在基于Prolog的面向对象编程风格框架下支持必要的功能。系统中还集成了一个设计规则检查器。对于不太熟悉OCLS语法的用户，提供了带有图形的语法导向编辑器。在我们的LSI设计系统中，设计基本上是选择必要的单元并将它们组合以产生更大的图案。选择目前由用户在单元格目录显示窗口的帮助下完成。由于我们认为基于假设的推理可以有利地用于自动化这一选择过程，因此我们开发了一种增强的知识表示和一种用于执行基于假设推理的知识库的学习机制。

We have conducted a research from a viewpoint that a knowledge-based system structure and intelligent functions based on that structure are required toward advanced VLSI-CAD systems. For the knowledge-based CAD system. automatic or semi-automatic synthesis functions are needed in addition to the analysis or verification functions explored so far in connection with existing knowledge systems. Although several approaches may be possible toward a knowledge-based LSI pattern design system, we have constructed a prototype system having a functional cell library as a kernel. The functional cell library is based on a knowledge structure and stores a set of intelligent cell data (components) to be combined in response to the request. Our LSI functional cell library system named OCLS can presently support NMOS artwork pattern design based on the Mead-Conway design methodology. A pattern description language has been developed, which is suitable for a hierarchical pattern description, expressing geometical patterns with using variables, and embedding related knowledge. The language supports necessary functions in the framework of object-oriented programming style based on Prolog. A design rule checker is also integrated in the system. For the users not enough familiar with the syntax of OCLS, a syntax-directed editor with graphics is provided. In our LSI design system, the design proceeds basically as sellecting necessary cells and combining them to produce larger patterns. The selection is done by the user at present with the help of a cell catalog desplay window. Since we consider that a hypothesis-based reasoning can be advantageously used for automating this selection process, we have exploited an enhanced knowledge representation and a learning mechanism for the knowledge-base performing hypothesis-based reasoning.