Automated knowledge acquisition for a computer hardware synthesis system

The MICON Synthesizer Version 1 (M1) is a rule-based system which produces a complete small computer design from a set of abstract specifications. The ability of M1 to produce designs depends on the encoding of large amounts of domain knowledge. An automated knowledge acquisition tool, CGEN, works symbiotically with M1 by gathering the knowledge required by M1. CGEN acquires knowledge about how to build and when to use various computer structures. This paper overviews the operation of CGEN by providing an example of the types of knowledge acquired and the mechanisms employed. A novel knowledge-intensive generalization scheme is presented. Generalization is a pragmatic necessity for knowledge acquisition in this domain. A series of experiments to test CGEN's capabilities are explained. A description of the architecture and knowledge-base of M1 is also provided.