Date: Monday , July 02, 2007
Who says we live in a digital world?” questions Ravi Subramanian, President and CEO of Berkeley Design Automation, a company that thinks and breathes analog.
He explains, “As the world gets digital the paradox is that there is more of analog.” For instance, be it the audio or the display or the memory peripherals of an iPod—analog is an essential component; digital music and movies means so much more analog stuff.
While the design engineers today are able to get the digital substance right, analog is where the big challenge is. Quite often managers scram, “Where’s the analog stuff. It is not working yet.” Subramanian questions the thesis: Why analog teams couldn’t get across the hurdle?
Ten to 15 years ago, analog designers were creating circuits using about a few dozen transistors. Analog designs did not need to use the most advanced silicon technology; there was no dearth of existing device models and design rules, and they benefited from the experience gained with previous designs implemented on the same. Basic circuit simulators were adequate to help the designer produce acceptable designs.
Today, analog design involves implementing 2.4-GHz RF transceivers using thousands of transistors in a 65-nm silicon technology, and the architecture is prone to wide performance variations across the process, voltage, and temperature ranges. This transceiver shares silicon with millions of gates of digital logic that produce substrate and power-supply noise that the transceiver must tolerate. And the cost of a silicon re-spin has escalated to more than a million dollars.
The old method of relying solely on experience from previous designs and using silicon for verification doesn’t work in this case. It is difficult or impossible for the designer to intuitively determine which of the thousands of devices in the transceiver is most vulnerable to the power-supply noise generated by the switching of the digital logic.
Today’s designs require complex verification and typically have many more variables than abstraction can handle. For example, radio-transceiver circuits have a broad range of requirements, including noise figure, linearity, gain, phase noise, and power dissipation. Advanced simulation and verification tools are necessary to verify these complex designs.
In the last few years, a number of EDA companies have tried to come out with analog circuit synthesis tools that help in reducing the effort required to build circuits. Most of these techniques have achieved only partial success at best, because many issues need significant improvements in both analysis tools’ functions and engineering methods.
Despite much work, SPICE, developed more than 30 years ago, is still the circuit simulator workhorse in the analog world. In the last few years, a handful of EDA companies have introduced versions of fastSPICE. These simulators trade off some accuracy to achieve simulation speed-up. They allow designers to verify a circuit for 5 to 10 percent margin of error, eliminating the most obvious design problems in less time. Analog circuits require greater precision that what traditional digital fastSPICE offers—they require the highest levels of accuracy.
“What we need today are tools to employ advanced circuit-analysis techniques that accurately capture the complex physical behavior of analog circuits like RF transceivers to ensure working designs and minimize or eliminate the need for the silicon simulator,” says Amit Mehrotra, co-founder and CTO of the company.
Subramanian and his team are keenly watching the markets such as consumer (flat-panel TV, DVD players, digital camera, multimedia), wireless (WiFi, cellular transceivers), computing (PCI, XAUI, SerDes), and networking (10G, SONET). “All these markets have large and growing analog/RF problems, with aggressive time-to-market pressures,” says Subramanian, and adds that the demand for company’s products comes from the increasing gap that analog, RF, and mixed-signal chip designers are facing between what today’s analog verification tools can deliver and what is required for today’s chip designs in consumer, communications, computing, and networking applications.
According to Dataquest, the market for analog/RF EDA tools will be worth over $700 million in 2008. It is one of the few sectors in EDA that is growing quickly, with double-digit growth in the last two years. Subramanian is sure to latch on to this growth.
Founded in 2003, Berkeley Design Automation is funded by Woodside Fund, Bessemer Venture Partners, Matsushita Electric Industrial, and NTT Corporation. The company was founded by Mehrotra and Amit Narayan, both of who are a doctorate from the University of Berkeley, California—the very place where SPICE originated in 1972. Subramanian too is a PhD from UC Berkeley.
To succeed in a market dominated by incumbents like Cadence and Synopsys, a startup had to offer something radical. Therefore, the founders came up with a mathematical theory that helped solve circuit equations by a factor of 5x to 10x with no loss of accuracy. “Many disbelieve this. But we can demonstrate it everyday without any compromise on silicon accuracy,” claims Subramanian.
The trio knew that despite their aforementioned theory, they would need to synergize experts from various disciplines to achieve significant breakthroughs. Early on, they were able to attract talent from places like IIT, Stanford, and Berkeley. They recruited engineers who relished challenges and had a hunger to win. The formula worked. In the next 12 months, the team focused on building a product that addressed noise—a very critical problem in the analog world. The Noise Analyzer, as it came to be called, is based on the company’s Stochasitc Nonlinear Engine and is used for the noise analysis of analog or RF integrated circuits that incorporate a phase-locked loop (PLL).
“People have cannons lined up at the door so you don’t take up the analog piece they are already solving for someone else,” says Subramanian. So his strategy was not to dash head on with the incumbents. He could demonstrate value to clients with a niche product that had an underlying engine to solve the noise problem with speed and accuracy.
With the initial success as a confidence booster, the team went ahead with building what constitutes the company’s main simulation products: Analog FastSPICE and RF FastSPICE which are ideal for difficult verification challenges such as multi-day or multi-week long simulations that otherwise cannot converge. Clients, who had a taste of the company’s noise product, were ready to try the new products. These products require no block-level tuning and are fully compatible with today’s popular SPICE simulators, RF simulators, and their environments. “It is essential to launch a product and have the customers standing behind your claims,” says Subramanian.
Today many leading players in the consumer electronics space like Sony, Panasonic, NEC, Fujitsu, Silicon Labs and Qualcomm have tested the simulators that Berkeley Design Automation offers and used the tools on commercial production designs.
Toshiba Corporation, the world’s largest supplier of semiconductors for consumer electronics, has recently adopted the company’s Analog FastSPICE circuit simulator. With Analog FastSPICE, Toshiba’s design teams can now complete analog circuit verification tasks that were previously impractical or impossible. “Berkeley Design Automation offers a breakthrough verification technology that is critical for Toshiba’s consumer electronics business,” says Tamotsu Hiwatashi, senior manager of planning department, System LSI Design Department, System LSI Division of Toshiba. “Our verification requirements exceeded the capabilities of traditional SPICE and digital fastSPICE simulators. Analog FastSPICE delivers accuracy that is as good as or better than traditional SPICE at higher performance than digital fastSPICE simulators on analog and RF circuits. This enables us to perform verification tasks that were previously impossible.”
The New SPICE
Every circuit simulator tool has to solve the same set of circuit equations. The trick is in the way mathematical techniques are applied to solve the equations. With the core knowledge of applied mathematics for electronic circuit simulation, the founders wrote fundamentally new algorithms.
To get ahead of the competition, BDA had to get the simulations executed at much faster speed than the rest. The trick was in how it defined the software architecture to do the numerical analysis. “Today we deliver 5x to 10x speed-up factors on a single CPU. We are now demonstrating to customers how fast it could be on a multi-CPU. People are astounded by the speed,” notes Subramanian.
The Road Ahead
The company’s base technology is like a core engine on top of which other silicon analysis such as optimization of circuit layout can be performed. Subramanian wants to leverage the core technology and provide solutions in other areas of analog world where the solution earlier was not practicable. “For further growth, expansion in the existing line of business using the base technology developed by us is paramount,” he says. “As we achieve profitability, we would focus on developing tools such as addressing transistor-level silicon problems.”
According to iSuppli, a market intelligence provider, at the beginning of the present decade less than 10 percent of the chips shipped had significant analog content as defined by the silicon area. By the end of this decade at least 90 percent of them will have significant analog content. This opens up exciting possibilities for Berkeley Design Automation. Subramanian is quick to latch on. “We are here to build the next analog RF company,” he says.
Having got his doctorate from University of California, Berkeley, Amit Mehrotra went to University of Illinois, Urbana Champaign as a faculty member to pursue research in RF design. He quickly realized that with the current set of tools that were available in the market for doing RF and Analog design no meaningful research contribution could be made. Mehrotra then thought why not build a new simulation tool from scratch to help design the analog circuits the better way. He shared this idea with his good friend Amit Narayan whom he knew from his days at IIT Kanpur and later in UC Berkeley. Narayan explained to Mehrotra that if a research faculty like him saw the constraints of the existing tools to design, then there was a much bigger need in the enterprise world for a new tool that addressed the issue.
Narayan went out and spoke to experts in the industry. While many agreed that a new tool would be a great idea, they also said that it was impossible to build such a tool. Not disheartened, the duo went ahead and founded Berkeley Design Automation in 2003.
Mehrotra spent the first three years steering the product development. Early this year, he temporarily relocated to Bangalore to kick-start the India Development Center. As the founder of the company, Mehrotra is best placed to attract right talent, and nurture and retain them. While his presence motivates the team here, he has made sure that the India team is engaged in equally challenging projects as in the U.S.
Work is being done in India on several fundamental components of circuit simulators. Device models, and specific type of compatibility between different kinds of simulators are the main areas being worked on.
The basic core simulation engine is constantly evolving. There is maintenance and upgrades happening all the time. For instance, new device models come up virtually everyday. “For a startup company like us, it is impossible to implement all of the device models into the simulator. We look at the customer demand and decide what needs to be implemented. We have seen that while some of the models may not be mainstream ones, these are still being used by specific customers and we cannot proceed in our engagements unless we support such implementation,” says Mehrotra. All such actions happen in the Bangalore Center.