Over the past decade, the percentages of ASIC design dollars spent on tools and labor have shifted dramatically toward labor. As the design process has grown in the number of steps and complexity, tools and tool flows have also increased in complexity and runtime. Some of the complexity stems from the underlying technology changes and increases in design size, and some comes from overstretched methodologies. The overhead of finding, testing, learning, integrating, purchasing, and using a new tool can easily exceed the labor involved in solving a problem with VI, Perl, and thousands of lines of tool scripts.
Even though the demand for more productive design solutions is widely recognized, the design productivity gap, defined as the divide between available silicon gates and design tool capacity, keeps widening. To reduce this gap, two different paths must be taken simultaneously. EDA vendors must continue to innovate by improving the features and performance of tools supporting existing methodologies. At the same time, new design methodologies must be explored to take larger leaps toward significantly improving the design flow and productivity.
However, EDA vendors are rarely successful in defining and developing new methodologies all by themselves. System companies must help define methodologies for adoption by a large group of users. Otherwise, radical improvements to productivity won't come for a while due to the system companies' cautiousness in moving to an unproven methodology, as well as the serious financial risks for EDA companies to buy into solutions that may never be widely accepted.
EDA companies have a strong history of investing in the development of new tools. On average, the EDA industry actually invests 25% of its sales revenue in research and development. That's about twice the investment made by semiconductor companies. Through this ongoing development, EDA companies have continually demonstrated the ability to improve upon tools that support existing methodologies.
These developments have yielded faster simulation, higher capacity, and better runtime in equivalence checking, improved automation of synthesis, and physical synthesis, to name a few enhancements. Simplifying tools, broadening capabilities, and making them faster will immediately begin improving design productivity. EDA companies need to continue this process.
Yet improvements in existing tool flows won't deliver the order-of-magnitude improvements necessary for this next decade. We need new methods of problem solving.
Fundamental changes in methodology have been most effective when started or sponsored by system companies. For example, initial approaches for performing schematic capture and simulation were first developed by system designers. Early logic synthesis work was done at IBM and GE. The Vera testbench language traces its history to Sun Microsystems.
The system-on-a-chip (SoC) verification problem is a current example of a situation requiring collaboration between systems and EDA companies. Here, verification needs are outstripping simulation capacity because of increases in intellectual-property use and software content in such systems. Over time, as the pain level rises, more and more designers will explore new approaches. But tool development will stay behind until there's sufficient consensus on tool features and a large enough customer base to financially support the tool development and ecosystem. This inevitably leads to a lag of years between the recognition of a severe problem and its solution.
During this time, generations of EDA start-ups and projects at established EDA vendors attempt to enter the new market and die. Some find success, not necessarily because the successive generations of tools are better but because of lucky market timing. This is a very inefficient system.
If systems companies and EDA companies were to cooperate today, a very promising potential solution already exists—prototyping. Although intensively used in some companies with great success, there's not yet widespread consensus on the use of prototyping for the SoC verification problem. A concerted cooperative effort in this area would rapidly yield benefits.
System and EDA companies must act as partners, collaborating to develop new methodologies that will be widely adopted and brought to market in a timely manner. It can take years for EDA companies to develop tools for new processes and new design challenges. Without engaging with EDA partners early in the process, system vendors won't see the required tools at the right time.
From a financial perspective, it also is important that both system and EDA companies bear the responsibility to ensure that new methodologies come to fruition. An EDA company that attempted to develop a radical new approach on its own would require a risk that may not be worth the reward. For system companies crying for new innovation, the risk is fairly small while the potential reward will benefit them for years.