by Katherine Derbyshire, Contributing Editor, Solid State Technology
What happens when you put a design software vendor, a chip manufacturer, and a photomask maker all at one table and ask them to talk about yield? Not surprisingly, the software vendor finds himself on the hot seat.
Monday (July 16) at SEMICON West, Synopsys VP Anantha Sethuraman found himself wedged between Dan Armbrust of IBM’s Systems and Technology Group, and Chris Progler, CTO of Photronics. The occasion was a panel discussion on DFM — which the panelists agreed stands for “design for money” — and yield. As Sethuraman put it, “we aren’t chasing dirt anymore.” Process variability and the ability to actually manufacture a given design are now critical to success in the IC industry.
Except it’s not so easy to actually do that. Many approaches to yield improvement focus on identifying process sensitivities, and avoiding those features. Most companies, Progler said, are making more regular, array-like designs, even for logic. Array structures allow NAND flash companies to achieve very high yields, even using aggressive lithography that pushes the theoretical resolution limit. Armbrust reminded the audience, though, that constraining the designers makes a design take longer. A chip that doesn’t yield can’t make money, but neither can a design that doesn’t exist, he pointed out, so software tools must help manufacturers balance these tradeoffs.
And in fact, the software is beginning to do exactly that. Sethuraman claimed that Synopsys software incorporates a good understanding of where variability comes from. For IBM, predicting yield is critical to the success of the company’s foundry business, and also critical for planning launches of integrated systems. Good yield predictions are the foundation of good inventory control.
On the mask side, though, the picture is not so bright, according to Progler. On some designs, he said, only 10%-20% of photomask defects actually matter, while on others, 80%-90% of defects are critical — yet mask acceptance criteria do not differentiate between the two. Though tools can identify defects as noncritical, people clearly don’t trust the results enough to risk their production. A more trustworthy link between inspection results and manufacturing outcomes would be “revolutionary,” giving enormous cost and scheduling improvements, he said. Moreover, as the first physical embodiment of the design, the mask potentially offers an enormous amount of information. Even without a complete process flow, a photomask allows manufacturers to begin to measure actual process outcomes, rather than relying on simulations. Yet so far, this resource remains largely untapped.
Despite their differences, all the panelists agreed that the situation will get worse as the industry moves to 65nm and 45nm features. The shift to 65nm will bring fundamental process changes, including some use of immersion lithography. At 45nm and below, stress fields are more difficult to manage, and new gate stacks begin to appear. Much like this SEMICON West DFM panel, software vendors are likely to find themselves wedged between manufacturers and designers for years to come. –K.D.