October 9, 2006—ICCAD, Santa Clara, CA—Leon Stok, director of EDA at IBM, taled about "the changing nature of innovation in EDA" at his keynote address at the International Conference on Computer-Aided Design. The definition of innovation in not invention, but a process that drives changes in society and industry. Unfortunately, no single company can supply the expertise to address all of issues and cannot afford to do the R&D. Innovation doesn't happen in isolation, but must align with the current market dynamics.

In the earliest period of EDA, the invention era, the programs were small and focused on PCB layout and discrete component design. The underlying algorithms were small and complex, but overall these tools changed the level of automation in design. Then ICs were invented and their growth in size made schematics too limited for design functions.

This limitation ushered in the next phase of EDA, the implementation era. The new algorithms were linear and most innovation came from the academic areas. At Moore's law took hold, the design flow integrated synthesis and created the RTL implementation flow. Scaling started to break as interconnect delays dominated leading to the demise of prediction, as secondary and tertiary effects became primary ones. This change drove some of the later steps into the front end of design as new cost functions and reduction of guard banding pushed limits.

The next stage was the integration era. The back-end tools are integrating and merging functions. The best example of this type of tool is the Magma tool set, even as the other EDA companies fall far behind. The latest tools must function in a world of atomic granularity, and in this environment, the sources of innovation are suspect. Corporate CAD has spawned start-up companies, the big three EDA companies are mired in legacy issues, academia is not providing practical implementation capabilities, and the start up companies are finding that agility is insufficient to get to market. The next set of problems requires "global" domain expertise, something not readily available.

Now, EDA is moving towards a level of platform innovation, like an iPod, where the monolithic structures are changing to multi-layer structure with a single interface. The platform integration is a tool architecture for design implementation and analysis that uses point-type tools as plug-in modules into the main engine. This type of tool is necessary to make the design jobs easier and reduce overall complexity. The plug-in functions include analysis, test (especially delay test) and links to the test equipment, silicon diagnostics, and design for manufacturing-type functions. This type of tool needs a simple set of APIs and links the system-level design to the rest of the flows over a distributed and heterogeneous compute platform.

The need for a major change in tools is being driven by the problems of variability and design complexity. Because most current tools are not designed for multiprocessor operations, the whole tool chain needs to change. Unfortunately, there are some significant hurdles to overcome before the new iPod era is here. One problem is the decrease in investment for R&D as well as the lower availability of funding for new companies. Another is the development of standards and APIs that are not ASCII based so various tools can be plugged in as needed.

With this new breed of toolset, engineers will be able to focus on differentiation. EDA needs to create the disruptive technologies that offer greater than 10 x improvements in throughput and to look into greater reuse of interfaces and standard functions. Companies need to develop a high tolerance for risk and find some path to early proof points. Other industries are going through deconstruction and rearrangement to include partners, suppliers, users, consortia, and even government support for common functions and interfaces.

The failure to address the needed changes increases the risk for the whole eco-system. EDA has to stop using proprietary formats and standards that increase switching costs for users and change to platforms with "open" APIs to enable their own and their partners' success. This openness will create new market segments and increase functionality for all. EDA has too many niches, so it must move towards a classical market structure with 2 major companies and some support companies.

This collaborative innovation is new and difficult, but eventually will enable industry growth and will change the fundamental concepts of intellectual property in tools. Open source type of developments for common areas like databases, operating systems, and some middleware is a viable and necessary transition. It is better to create APIs and data structures that allow users to access only the fab information that is relevant to their design, as not all of the data is useful at all stages of design. All of the CAD functions belong in a single eco-system, that provides multiple platforms for academics, and reduces the costs and resources to develop the plumbing and middleware that are common to all tools.