The evening panel at ICCAD looked at issues around "CAD research, pay now or pay later…" as an area of interest and concern for the EDA industry. The panel was comprised of moderator; Juan-Antonio Carballo of IBM and panelists; William Joiner of IBM and the SRC, Rob Rutenbar of Carnegie Mellon Unversity, Andrew Kahng of University of California- La Jolla, Leon Stok of IBM, Jacques Benkoski of US Venture Partners, and Andreas Kuehlmann of Cadence Berkeley Labs.

In general, all of the panelists agreed that the need for innovation and new EDA tools is still important. The academics all noted that research grants are down but there are still many areas of research that are important and viable for further expansion. Rutenbar opined that about every 5-10 years, a new "virtual wall" to the next generation comes up. So far, every challenge has been overcome. The next set of problems will require hiding the unpleasant physics to enable the increasing non-digital content of the designs. Automatic modeling that will link the physics and designers will facilitate the sub-65 nm designs which will be much more difficult than anything else ever tried. The main aspect of new tools needs to be usability.

Kahng noted that the next generation of tools will need to have a "snap-on" architecture, to allow incorporation of new (open source) functions as a way to address different and unusual problems in design. EDA only puts efforts into those issues that have incentives. The key to success over time is for companies to get and keep good people. The directions for R&D need to have clarity of purpose and a sense of reality—the problems must be solvable. The culture for success must allow for exploration and failure, without repercussions. The industry must reduce inefficiencies and redundancy to get better results from the small number of EDA engineers in the industry. One measure to increase productivity is to develop a formal language to help industry translate issues into specifications.

The industrial panelists were more concerned about the costs and timeframes than the specific details. Stok claimed that timing is important. The nature of invention cycles is for change and improvements to migrate to users over a process generation. Companies need to have some idea of when an invention will pay off, especially since most developments take over 5 years to get from concept to product. The industry is at an inflection point as we near the end of CMOS. The roadmaps for technology indicate new materials and devices, but researchers will need to find people and to trade off designer needs versus time for development. EDA is a whole eco-system including academics, in-house captive, and EDA companies. The start-up companies fill holes in the systems, while the existing EDA companies spend their time in legacy issues and financial and legal issues. It's very important not to start too early on developments!

Kuehlmann agreed with the academics about the drop in important papers while advocating a set of tools that would create a fast path from system description to implementation. Manufacturing is getting increasingly complex and now is becoming context dependent. The industry needs to have compact models that can adjust resolution on the fly and incorporate statistical methods. CAD is not dead, but the current format of slices of transforms using unstable algorithms is getting worse. More importantly, the results are not repeatable. To address these challenges, EDA needs to raise the level of abstraction while enabling design for verification. The lower level models for manufacturing need to be more compact and flexible.

Joiner also advocated movement towards higher levels of abstraction to the system level incorporating both hardware and software. Design has to evolve to an assembly of functions that are then optimized. The design- manufacturing interface has to propagate the system-level concepts all the way through the design chain. As the industry consortium representative, Joiner called for more government involvement to research overall industry issues and more effort to minimize overlap and redundancy. Success will come from partnerships of industry, education, and users with collaboration of the consortia and government.

Benkowski looks for more integration of IP and reuse. Power issues need to address the issues at the system level, since the 5 to 10 percent improvements available from post design processes is insufficient. Next generation tools need to include software integration as an automatic process. The users need to develop a lower tolerance for bugs, since too much of the software is unstable and requires work-arounds. RF and mixed signal verification is going to become very important soon as will system in package tools. The EDA marketplace needs a structure that performs pre-commercial R&D to minimize overlap and redundancy.

To comment on this article send email to:gmoretti@gabeoneda.com