It looks like it’s time to throw in the towel on scaling. According to Bradley McCredie, Ph.D., IBM fellow and VP of IBM systems and technology center, classical scaling is dead, because atoms don’t scale. “Aggressively scaled oxides, 5-6 atoms thick, leak current like a sieve,” noted McCredie. The bad news about this; continuing along this path to further scale oxides will diminish reliability to the point that “leakage is the least of your problems.” The really bad news, according to McCredie, is that “clever” CMOS engineers compensate for these lost benefits by cranking up power (Vdd).
McCredie illustrated his point with a table that showed as scaling continues, how much of the advancements have to do with scaling and how much have to do with innovation, or “tricks” such as using strained silicon to achieve 65nm scaling. “The last time the industry really scaled was at 130nm,” explained McCredie. “Once silicon stopped scaling, packaging got really important.”
McCredie’s most important message was about how our industry can continue to grow beyond its exponential. “Processor companies will act like scaling can go on forever, but we’re going to hit the wall at some point.” He said. “No exponential can go on forever.” He referred to a presentation he listened to at DAC 2009, given by Frank Schellenberg of Mentor Graphics that talked about “life after the exponential”. Using the automotive industry as an example, he explained how once they had achieved 0-60mph in under 6 seconds, they turned to adding value through feature enhancements such as safety, fuel efficiency, production systems, entertainment navigation systems, etc. He also cited the Concorde, as an example of advancements that didn’t have the infrastructure to support its existence, therefore all the new jets that have come out since then are have not focused on matching the Concorde’s speed capability, instead recognizing the value of added features, comfort, maximized capacity, etc.
In the absence of innovation, McCredie says cost reduction is always next. And certainly this approach has been imposed on systems vendors. But he urged the audience to break from the past, saying that that what’s not next is to go back and innovate in the same space (chip design, circuit tuning, architecture, micro-architecture) and that we need to look at more process functions on our chips. Looking forward, from the packaging space this means innovations that address ease of implementation, highly differentiated packaging, leveraging 3D technologies, multichip heterogeneous solutions, and other things that we perhaps haven’t thought of yet.
This calls for what McCredie called “innovation across boundaries.” There needs to me more collaboration and cross-pollination of engineering disciplines. “We need people who can cut across the stack faster; people who understand everything from chip design to package, to software. “ For his part, he encourages his engineers to spend 30% of their time looking outside their area of discipline to learn more about what needs to be integrated. The more people know about what happens before and after their part of the supply chain, the more we can build an infrastructure to support the outcome. One thing’s for certain, as McCredie notes, “now we really have to earn our paycheck.”
