Constructed at Carnegie-Mellon University and funded in part by Hewlett Packard, the ULYSSES design environment project was an attempt to integrate several sophisticated tools for electronic design into one intelligently interactive, usable, and efficient package. This search for the Holy Grail of CAD/AI has also produced, among other packages, the Bell Labs “Designer’s Workbench,” the Stanford/Xerox “Palladio,” and the MIT project “Schema.”
One way the ULYSSES project aimed to avoid some of the mistakes of the past was by making the human interface familiar to and usable by electronic designers. It also added some features that are needed in the context of everyday industrial practice, such as a design database. The ULYSSES project initially made use of the “Electric” database created by Rubin at Schlumberger Palo Alto, but later implementations will have an INGRES-based version. The commercial availability of design databases for ECAD environments will contribute greatly to the environments’ acceptance by users in industry.
Much of the difficulty in building such an environment involves creating a model with the various levels of abstraction that designers find useful to represent the design process. That process itself is iterative and not strictly hierarchically decomposable (although it would be much easier to automate if it were). Consequently, one must allow for simultaneous representations of the same design (at RTL level and gate level, for example) and also maintain a number of different versions of a design at various stages of completeness (so that it is possible to backtrack without a total redesign).
The ULYSSES project attempted to solve problems like this by combining several intelligent ECAD tools (for example, WEAVER, an autorouter; MASON, a floorplanner; and TALIB, a design synthesis tool). A knowledge-based blackboard-type controller is used to create scripts that describe design tasks and methodologies; these scripts then coordinate the various tools. The human designer can interact with ULYSSES, to obtain an explanation of the design process model and plans or to inspect and intervene, at any point. Of course, when ULYSSES runs out of good ideas, it stops and asks the designer for a helping hand. This degree of interactivity is difficult to achieve in a design synthesis system, yet it provides an essential channel for human input. The ‘wetware’ that people possess can solve problems that are difficult to describe and negotiate in hardware and software.
Bushnell provides a thorough background of the design methodologies used and the AI methods with which they are modeled. The example design tack included in chapter 7 takes a moderately complex (700-1100 transistors) CMOS FIFO memory chip design from SCS hardware description language form to a complete layout in one week. An unaided human designer would need four weeks to produce even a partial solution, but the modest size and complexity of this example will not impress the industrial ECAD user. The aim of the MCC VLSI CAD project (1-million-transistor chip designs by 1995) is more on the order of what semiconductor companies need to provide real competition for their Japanese rivals. However, we must congratulate Bushnell on what he has achieved with the slender resources an academic environment provides.
This volume is best suited for graduate coursework in electronic design and for those interested in practical AI applications. Written in a terse style that reveals its origin as a Ph.D. thesis, it is nonetheless clearly laid out (with the exception of a few crowded figures, e.g., 5-3) and will give ECAD users and purchasers an idea of current trends in academic research into the theory and practice of design automation.
