Jim Hogan and Paul McLellan opened an interactive, thought-provoking discussion at ICCAD this week on how EDA needs to change to be successful in the year 2020.

The conclusion of this well-known duo–Hogan, the VC, and McLellan, the blogger, in their current incarnations–pointed toward optimization and software signoff, given the amount of software that is now moving into designs and the need for eking better optimization out of the individual components. They’re right on a couple of important points, which are listed below, but that still doesn’t answer the overriding question.

First of all, it’s clear that EDA isn’t commanding the value it should, given the complexity of the problems it deals with and the incredible engineering feats it provides for customers. There simply aren’t enough large customers left doing enough design starts using big enough budgets.

There are fewer design starts of bigger, more complex chips, to be sure. At 32nm and beyond, designing chips is like the application of what was learned in a theoretical physics class several years ago. When you get to designs where the insulation is measured in single digits of atoms it’s clear this stuff has gotten beyond the capabilities of many companies without development budgets that are larger than the GDP of entire countries.

It gets worse, too, when you think about validating and verifying everything from multiple states and power islands to multiple cores. This is complex stuff, and could well pass beyond the capabilities of any single company—even the largest and most prominent names—within several process nodes.

Second, software is now a key part of every design. It’s not that the software is necessarily doing anything different. It’s that there is more of it, and it all has to be integrated so it works together, as planned. You don’t want to be getting a phone call while reading your mail and have the device go dead because of an unexpected glitch in scheduling caused by yet another open window.

Because of the complications of developing software, and the sheer magnitude of the task of getting the hardware to work with the software in proper order with all the right rules, software and hardware engineering now have to come together in the same place. This is an unnatural fit, because neither of these groups speaks the same language or things the same way. But economics often forces odd combinations, and the old way of doing things in isolation no longer works.

So what becomes of an EDA company in all of this? Does it really focus on optimization? Or does it become something other than an EDA company? Our guess is both—and neither. Looking at the big guys in this market, Mentor’s approach has been to expand well beyond the core EDA flow into areas like yield and test (read yield & test, rather than two separate areas), as well as other pieces of DFM. Synopsys has done the same in areas such as standard IP and high-level synthesis. Cadence has focused on mixed signal and system in package, in addition to its core EDA market.

We’re already seeing these companies branching out into new areas, and the exploration is just beginning. They’re also continuing to invest in their cash cows. But where they’re reaching is far beyond just optimization. It’s also far beyond how we describe EDA today. What’s not clear is where exactly all of this will go, which markets will provide the best opportunities, and where the real money will be—whether it’s at the leading edge of design or enhancements at existing nodes.

The bottom line is this is still all about Moore’s Law. While it’s technically feasible, and for some companies it’s economically feasible, the number at the front end is shrinking while the number at the back end is exploding. Adding real value may be less about pushing the envelope than building an ecosystem for the mainstream developers in mass markets. It may be less about the iPhone than about the Droid, and in developing markets it may be less about either of them than communication that works all the time.

This is another kind of economic reality, and value changes from one market to the next. In these markets, one size doesn’t fit all and one solution may not solve everything. What’s good enough in China may not be the same as what’s good enough in India, which may be entirely different from what’s good enough in the United States, Europe or Japan. How do you craft a business model around that?

–Ed Sperling