Dr. Richard Newton, Dr. Andrew Mayer, Dr. Wray Buntine
(DARPA) DABT63-95-C-0074-NEWTON-06/96
The Electronic Design Automation (EDA) industry is struggling to adapt its methodologies to the ever-increasing complexity of modern electronic design problems. However, considerable evidence suggests that existing tools and techniques will not scale, and that continued industry growth will require fundamental methodological changes. UC Berkeley's WELD project is pioneering the attack on the EDA problems of tomorrow, with the goal of creating a new, scalable architecture for complex system design. Our efforts are concentrated on the two major sources of complexity: the managerial complexity of coordinating an ever-increasing number of humans involved in high-level design and the technical complexity of optimizing for orders-of-magnitude increases in number of deep-submicron physical components for low-level design.
At the high-level, WELD aims to construct the first operational prototype of a Internet-wide complex systems design environment. The goal is to enable collaboration between humans distributed over different areas of specialization, geography and time. Towards this end we have concentrated on the technologies of Java and the World Wide Web, and are experimenting with new visualization tools and user interface and groupware paradigms. In addition, we are developing the software infrastructure which will allow such applications to interoperate.
At the low-level, WELD is pioneering the development of the next generation combinatorial search, planning and hierarchical abstraction subsystems for behavioral synthesis and physical design. This effort draws heavily on recent and dramatic advances to computational problem-solving using the methods of statistical decision theory, Bayesian probabilistic inference, and learning, for instance to build adaptive stochastic search systems. These techniques couple naturally with advances in programming languages to support rapid prototyping of design systems.