SANTA CLARA, Calif. — With classical scaling dead and Moore's Law no longer relevant to device performance, the time is now for design-for-manufacturing (DFM) to "come of age," according to Bernie Meyerson, vice president and chief technology officer of IBM's Technology group.
Speaking to a large audience attending the EDA Tech Forum 2005 event here Thursday (Nov. 3), Meyerson preached the need for "highly advanced risk mitigation," whereby DFM capabilities provide a unified yield model generating a "net yield statement," giving manufacturers knowledge about how a device will yield before it is built.
"It's critically important to have foreknowledge," Meyerson said. "You cannot be behind in this business."
Meyerson said chip makers currently look at risk mitigation as a "unit process based affair," utilizing individual capabilities to increase yield such as critical area analysis, redundant via insertion, routing optimization and yield checks. But much more sophisticated DFM capability is required, he said.
"Unit process checks won't help in the future, because you have to take them as the minimum price of entry," Meyerson said.
DFM automation by computational technology is required to give chip makers the capabilities to measure, model and mitigate yields in the face of "increasing and inevitable variability" created by random dopant effects, oxide thickness and lithography.
The sophistication of device modeling technology must increase dramatically, Meyerson said, a significant challenge that is compounded by the fact that the number of devices that need to be modeled — and the complexity of those devices — is growing dramatically.
"This is a hugely complex and incredibility important road we've taken," Meyerson said of DFM.
Meyerson told the audience that adherence to Moore's Law — the famous proclamation by Intel Corp. co-founder Gordon Moore that the number of transistors on a chip would double every 18 months — continues to drive down the cost of devices, but the industry has come up against the physical limits of scaling and could no longer increase performance in this way.
"Moore's Law is utterly irrelevant to the performance issues that you are dealing with," Meyerson said. "People recognize this, but it's very hard for people to stand up and say that the world is different. In the past, classical scaling drove performance. In the future, it will be innovation."
Meyerson outlined some of the innovation challenges facing the industry and pointed to strained silicon as an example of innovation that would yield dramatic changes in the structure of devices and materials.