Guest Post: Rain or Shine for the EDA Cloud?

Recent announcements from IBM and others about supporting EDA tools in the cloud have spurred renewed discussion on this topic, including here at The Breker Trekker. As expected, the recent posts have been very popular with our readers. Those of you who have been following this topic for a while may recall that, almost exactly two years ago, EDA vendor OneSpin announced cloud support for their formal tools. We invited their VP of Marketing, Dave Kelf, to fill us in their experiences since then:

Two years ago OneSpin introduced the cloud version of it’s Design Verification (DV) formal-based products. Some commentators pointed at other failed EDA attempts to make the same move, suggesting more of the same. Others hailed the announcement as a bold move whose time had come. So… did it work out and what have we learned? The results are surprising, and suggest trends that make some EDA solutions a natural fit for the cloud, whereas others are questionable.

Of the various benefits projected to make a cloud-based solution succeed, the most lauded is the pay-per-use business model. The idea of an on-demand tool, which allows access only when required and does away with expensive, redundant EDA licenses, would appear a winner. To a certain extent it is, particularly as a top-up license mechanism on an existing installation. When a few more verification cycles are required for coverage closure, the cloud is simple and accessible. More importantly, usage can be funded from an operating rather than a capital budget without calling up the EDA salesperson.

However, as an alternative to a more traditional licensing model, pay-per-use is still in its infancy. The purchasing process turns on a decision whether to just negotiate the overall price down, versus having to explain to management and the legal team how the cloud solution is secure, and bet the company IP on it.

It has been suggested that IP security is an insurmountable entry barrier for cloud EDA. At OneSpin we took great pains to ensure that only formal proof problems were sent to the cloud, not the design IP. Engineers understand this process and accept the explanation. However, IP security decisions are not in their hands. Cloud usage must often be signed-off by the legal team, folks not known for taking chances, especially with the company IP. In an interesting twist of irony some of these legal teams use the cloud for their own work. Much of the resistance is inertia rather than a real concern, and this can change.

There is a crack opening in the security barrier stranglehold, and that is companies who both use formal verification and have their own cloud solution. Some of these organizations are open to cloud-based usage. A snowball affect is possible when these corporations leverage the cloud, tipping the security barrier to potential benefit trade-off for the rest of the market, and we are starting to see signs of this trend.

There is another benefit that has the potential to smash the security barrier. In IC verification, the ability to run more cycles is extremely compelling. Parallel processing provides this capability. In formal, the assertions tend to operate independently of each other, opening up a highly effective parallel execution opportunity. Couple this with the appearance of infinite computing in the cloud and a dramatic difference in verification performance is suggested. Furthermore, with a time-based payment model, the license/machine cost is the same for a run on 1 machine in 10 hours, or on 10 machines in 1 hour. That’s a free performance improvement!

Could this not be done with simulation? In large block simulation, much of the run is required to simply get the design into a specific state to start verifying scenarios, a serial operation. In formal, this is eliminated, as all design states are available straight away. Simulation tools must execute the SystemVerilog code in sequence, handling code branches and decisions as they come up, again limiting parallel operation. This issue does not limit formal tools.

It turns out that parallel processing is a big advantage of interest to many users. Of course it only applies to EDA tools that efficiently run in parallel, limiting the cloud opportunity. It should also be pointed out that parallel operation could be executed locally in a regression environment. However, a warehouse full of compute servers available on demand with no overhead is extremely compelling.

We have discovered, as with many EDA developments, that the pull of a significant capability, such as verification performance, will drive the cloud where a business model change will not. The security concern is set to dissipate over time, especially when traded-off for time to market at the management level.

So, what happened at OneSpin? There are a number of companies using the OneSpin cloud solution today and more expressing interest. Some EDA trends take a while to mature, and the EDA cloud is going to be one of them. It’s worth sticking with it, and we are!

Dave Kelf
OneSpin Solutions

For more information, visit http://www.onespin-solutions.com/

Tags: Accellera, Breker, cache coherency, cloud, cloud computing, EDA, formal, functional verification, graph, graph-based, OneSpin, portable stimulus, scenario model, simulation, SoC verification, use cases, uvm, VIP

This entry was posted on Wednesday, July 15th, 2015 at 3:02 pm. You can follow any responses to this entry through the RSS 2.0 feed. You can leave a response, or trackback from your own site.