Roots Of Distrust Spread

For most of the history of semiconductors there has been a persistent fear that someone would steal intellectual property from one company and sell it to another. There have been innumerable lawsuits involving corporate secrets that cross from one company to the next, and from one country to the next.

The biggest concerns always were at the leading edges of technology, where those secrets were considered the hardest to develop and where the stakes could be measured in billions rather than millions of dollars. But in one of the stranger byproducts of industry consolidation and the relentless march of Moore’s Law, those worries are becoming less of a concern at the bleeding edge than in the middle market.

The reason is a blend of economics and specialization. At 40nm and above, it was still possible to move a chip design from one foundry’s process to another without much of a penalty, and to develop IP that worked across multiple foundries’ processes. At 28nm and below, processes are so unique that everything is customized. Customers no longer can shift designs developed at those nodes from one foundry to the next without completely redoing it. There also are fewer players to point fingers at, and it’s obvious who’s doing what and contributing what. On top of that, there are only so many people in the world with the kinds of skills necessary to develop process technology at 7nm and 5nm, and many of them know each other. In many cases, they collaborate through groups such as Imec or the College for Nanoscale Science & Engineering.

As a point of reference, when TSMC pioneered the commercial foundry model in 1987, many industry players laughed at the possibility of sharing their chip data with an independent fab. Likewise, TSMC was criticized for not sharing data back with its customers to allow them to optimize their designs for TSMC’s processes. That complaint was echoed as other companies began entering the commercial foundry market after TSMC. But despite former AMD CEO Jerry Sanders’ statement that “real men own fabs,” the majority of chipmakers that have survived do not.

In fact, it is now so pricey to develop chips that it no longer can be handled just by the fab. Current estimates for a new process technology at the leading edge are in the $1 billion range, according to industry insiders. Companies that want to play at the leading edge are basically committing to developing those processes with the foundries. Rather than beginning their collaboration at version 0.9 or 1.0 of the process, they are now engaging at version 0.01 and working through each iteration of the process to refine it, improve yield, and rework their own piece of the puzzle. It has become such a close working relationship that sharing data isn’t an issue. It’s a necessity.

That doesn’t mean everything is perfect. Even at the leading edge of the mobility world, some updating of this data sharing process is in order.

“The fab gives you a design deck, you have a library, you give the fab a tape and you get back silicon,” said Michael Campbell, senior vice president of engineering at Qualcomm during a Semicon panel this month. “That has to change. This simplistic relationship does not work with 2 billion transistors, and in the server world it certainly won’t work for 25 billion transistors.”

But a few nodes back from there, the concerns are much different. There is even less data being shared, and much more concern about data leakage or outright theft. And the worries are growing as established companies meet startups vying for the different or the same pieces of IoT and other developing markets. Companies in this space often don’t know each other very well, and the churn in the market is so great that customers one day will be competitors the next.

The best way to deal with this is to bake in security at every level, from strict business methodologies for handing off designs and well-defined rules for managing data, to figuring out what data really does need to be transferred and why. At this point, the rules of engagement are limited or outdated, and the growing level of suspicion does little to foster access to data that can benefit everyone and grow the overall market.

Related Stories
Securing Chips During Manufacturing
Can directed electron writing change the security equation?
Making IP Secure
The semiconductor ecosystem is beginning to identify security holes, what tools can be used to plug them, and what else is needed.
Counterfeiting In The Automotive Supply Chain
Safety and security are at risk.