EDA lost in the clouds?

Cloud computing is all the rage in the world of information technology. IDC estimates that cloud computing will grow from $16 billion worldwide today to $55 billion in 2014. Indeed, the market research firm predicts that 80% of new commercial enterprise applications this year will be deployed on cloud platforms.

And yet, EDA vendors and users have not moved very quickly to the cloud. Why not? The quick answer: because the economic equation that is driving IT departments across Corporate America to the cloud doesn’t compute for EDA. At least not yet.

Of course, it’s a bit more complicated than that. And that has vendors and customers alike scratching their heads to figure out how, if ever, electronic design will be done in the cloud. There have been forays here and there. For example, Cadence Design Systems Inc offers EDA SaaS (software-as-a-ser vice) for small- to mid-size companies and startup Nimbic Inc does electromagnetic simulation in the cloud. But most of EDA is still done in the traditional way, in house on individual users’ machines or in the company’s own data center.

There are several factors holding EDA back.

First, most parts of the design flow aren’t amenable to cloud computing, says Gary Smith, chief analyst of Gary Smith EDA. Whereas typical IT business processes do quite well in a cloud bank of Intel-based servers, EDA requires different kinds of processing, he says.  “The industry is moving to heterogenous computing because of parallel processing,” he says. “The high-end apps are the ones that need that performance the most, and EDA is one of those apps.”

Indeed, in-house EDA servers have a much higher utilization rate than traditional data center servers, which undercuts part of the reason for virtualizing and moving to the cloud. When machine utilization is low, it pays to virtualize and move to a cloud model, says Raul Camposano, CEO of Nimbic. But if you’re running regression tests in a digital simulator, “these are 24x7 jobs that are run for months in a complex digital design so the computers are very highly utilized,” he says. “Then it’s not clear that the cloud is cheaper or more cost effective.”


In addition, large semiconductor companies already own large server farms to accommodate their EDA computation needs. In some cases, they are virtualizing the servers and adding provisioning, developing the farms into something akin to private hybrid clouds. But they have no pressing need to turn to a public or third-party cloud. A large integrated device manufacturer “already sits on 30,000 to 70,000 servers,” says Colgan. “Integrated device manufacturers already have made that investment in hardware, so no there’s no real need for IDMs to use the public cloud.”


Nevertheless, the company -- which provides a cloud where EDA vendors can serve up pre-sales demonstrations of their software -- is talking with customers about new ways to use cloud in the EDA business. And Cadence is hearing more cloud rumblings from customers. Large multi-national companies have started approaching Cadence over the last two years to discuss possible cloud scenarios, says Desharnais.

They are adopting cloud for general business applications and processes, and now C-level executives are asking why they shouldn’t move EDA to the cloud as well. Some are starting to expand from an in-house private cloud to a private cloud that is hosted by a third party, says Desharnais. “They’ve gone to the max on their own private cloud,” and may not be ready to expand internally or upgrade their hardware, “so they are looking to a third party.”

Similarly, Colgan thinks large companies may turn to the cloud when they temporarily need additional computing power. When they are close to finishing the design “and want to be sure they meet their tape-out date, they may need 5,000 more servers to run regression tests.” In that scenario, it may be easier to go to a cloud rather than try to get all those new servers installed and booted up in time, he says.  

Another possibility is for semiconductor companies to open up their private clouds for use by others. “Some of our customers have become so sophisticated in running their own private design cloud that their hardware is actually underutilized,” says Desharnais. He says that a couple of large Japanese customers are talking to Cadence about such a scenario.

Lastly, customers remain concerned about the security of their intellectual property. Engineers and lawyers do not want the brains of their chips -- the register transfer logic -- leaving their facility. “There is a huge sensitivity to having that out anywhere other than on premise,” says Desharnias. “Our customers say there’s no way they’re putting their RTL out there, particularly in a multitenant environment.”

Despite these obstacles, there are a few areas where EDA has already found a home in the cloud. Cadence’s SaaS has been operating for five years and serves about 50 clients, says Desharnais. Small and mid-sized companies don’t have the large data centers, and it makes economic sense for them to turn to cloud and pay for computing by the drink rather than make large capital expenditures.

And Nimbic, which sells software that checks designs for signal integrity and electromagnetic interference, started offering a cloud-based solution last year. It’s a good fit for the cloud because the data files for this part of the design are small, in the tens of megabytes, so bandwidth isn’t a problem, says Camposano. Nevertheless, many large Nimbic customers still prefer running the calculations in house. “The transition is going much slower than we thought,” he says.

But most think that EDA will eventually move to the cloud. Colgan recalls that most semiconductor companies used to own their own fabs and couldn’t imagine sending their designs off to be manufactured by someone else. He predicts that chip companies will go through a similar transition with EDA as the cost of computing in the cloud continues to drop.

"Today, it costs about 8 cents an hour to get an instance [virtual machine] on Amazon, and the price keeps getting lower and lower,” he says. “Eventually, the economic imperative will be clear."