Are we heading for a major retooling in custom IC design? EDA veteran Jim Hogan thinks so, and in a keynote speech at the Silicon Integration Initiative (Si2) Conference Oct. 9, 2012, he argued that the consumer electronics marketplace will drive a new era he calls "Custom 2.0."
The Si2 Conference is a yearly, one-day event that provides a review of the EDA standards activities that Si2 is facilitating. In an opening speech Steve Schulz, Si2 president, talked about the need for standards in an era of consolidation and collaboration (more details below). Individual sessions reviewed standards activities in design for manufacturability (DFM), process design kits (PDKs), low power, 3D-ICs, and the OpenAccess database. Cadence and Docea Power sponsored the 2012 Si2 Conference, which continues this year's 10th anniversary celebration of OpenAccess.
Hogan is a 35-year veteran of the EDA and semiconductor industries, a former Cadence executive, and a present-day partner with Vista Ventures LLC. He has lots of stories about EDA history, and this speech was no exception. Hogan related how he started his career building transistor SPICE models at National Semiconductor, and went on to work for SDA Systems, the company that merged with ECAD in 1988 to form Cadence.
An Unexpected Market
In the late 1980s, Hogan helped launch the Analog Division at Cadence, along with the Analog Artist product. It was tough, he noted, to sell to conventional analog designers, whose idea of automation was "getting a bigger slide rule." The surprise, he said, was that "the guys who actually used this stuff were the digital standard cell designers. We sold a lot of our products to digital guys doing cell characterization."
Today, Hogan said, custom design has a compound annual growth rate over 25%. "It's just nuts, it‘s such a growth engine," Hogan said. "I would never have predicted this."
Why this growth? A lot of it has to do with the smartphones in your pockets, Hogan told audience members. Quoting a Gartner statistic, Hogan said that smartphones are expected to grow by 117% in 2012, while PCs will decline by 36.5%. And smartphones are evolving in a way that will call for more custom design. Hogan noted that the Apple iPhone 5 has resulted in some key changes, including a much smaller PCB, lower IC count, higher level of SoC integration, and analog/mixed-signal content in SoCs.
SoCs in 2013 and beyond, Hogan said, will "use a ton of processors and they will require power management - all custom. Higher frequency radios - all custom. Memory access and retention of data - all custom design." There will also be "dark silicon" that's turned off when it's not serving a function, and SoCs will provide "true GHz" performance.
"28nm Breaks Everything"
SoCs in 2013 and beyond will also used advanced processing nodes, and that's another driver for custom design. "28nm just breaks everything," Hogan said. "The physics are different, the numbers are huge, everybody's methodology has to shift." Transistors are shrinking but atoms are not - resulting in atomic-level variation effects. One result: "It's going to be very hard to make an analog transistor look like a digital switch."
Hogan went on to talk about design variation, widening threshold voltage, and the inadequacy of traditional fast-fast and slow-slow corners at advanced nodes. Characterization is a huge problem, he noted, with fabless companies re-characterizing libraries every time new process rules come out - which may be every week for a new process. "That could be a problem when I've got to do a billion simulations," he said.
Indeed, faster simulation is a key Custom 2.0 requirement. Hogan noted how a post-layout, 12-bit ADC simulation with 7.4M elements could take 36 days. Much worse, a 5 billion 6-sigma variation analysis at one second/simulation would take 159 years. "Simulators have got to go up capacity wise," Hogan said. "You need a bunch of them and they've got to be fast."
Custom 2.0 verification will go well beyond traditional simulation, and look at effects such as flicker noise, shot noise, thermal noise, transients, harmonics, and EM integrity. Another Custom 2.0 requirement is IP reuse, and this calls for "dependency management," which goes beyond conventional design management by looking not only at data but the dependency between data.
Hogan also noted that Custom 2.0 will require standards activity. "Ten years ago we [Cadence] offered OpenAccess to Si2 and you've done a good job with that," he said. "Standards help promote innovation because you get rid of the barriers that plague startups. That gives me a platform as an investor to consider other options."
"This is a great horse race," Hogan concluded. "This is going to be fun."
Why Standards are Still Needed
Speaking just before Hogan, Schulz noted that the semiconductor and EDA industries are entering an era of consolidation and collaboration. But even with fewer players, standards will still be needed. There's still a need to integrate products and technologies developed in different places, Schulz said, and customers still want tools that will fit into internal methodologies.
What needs to be realized, he said, is that not all standards - and standards organizations - are the same. "There are multiple, valid approaches to standards creation - there is no single right way." Schulz showed how Si2 uses different kinds of structures for different standards efforts.
Schulz identified two current Si2 goals. One is to work with the IEEE toward a single power description format; despite work already underway on convergence, there are still two formats. Secondly, Si2 is launching a silicon photonics technical advisory board (TAB) that is looking at extending OpenAccess for silicon photonics design.
"I think we have a lot to build on, and OpenAccess is a very strong foundation for that," Schulz concluded. "I'd like to thank everyone for ten excellent years."
Related Blog Post
DAC 2012: EDA Industry Celebrates 10 Years of OpenAccess
InfiniScale claims they can do High-Sigma simulation orders of magnitude faster than standard Monte Carlo based methods.