Bright chips beyond the cloud
A few weeks back, I wrote a first article about the good discussions held at the GSA European Executive Forum. I promised at the time to come back with more... It took me a couple of weeks, as I was waiting to see what was going to be announced at the DAC conference last week before writing this article about EDA on the cloud.
One of the panels at the GSA Forum was about "Changing Business Models", led by Wally Rhines, CEO of Mentor, a Siemens company (and subject of a past article I wrote). Wally also gave an additional interesting presentation that was well reported on this EETimes article.
In his panel, Wally asked his fellow panelist on how they see the cloud changing their business.
Naveed Sherwani, CEO of SiFive was the first to say that he and his company fully embrace the cloud as a means of democratization of chip design. In fact, SiFive claim-to-fame is to bring to hardware design what open-source SW packages made to the Software industry, and the cloud is a vehicle that could facilitate that.
As far as I remember, Helmut Gassel, CMO of Infineon, took the cloud discussion to where his interests are - marketing and customer experience - by saying that his company are still looking for better ways to leverage the data it collects from customers visiting their website.
And the discussion continued to other topics...
When they opened the floor for questions from the audience, I asked for the mic to present Wally himself exactly the question he had raised before: As a major EDA player, how do you see the cloud changing your business?
Wally's said that he saw very little demand from his customer-base to offer EDA tools on the cloud, (maybe because of security concerns) other than an eventual Veloce emulation box on a private cloud to be shared among different teams within an organization. In fact, 2 days after this talk, Mentor announced that Veloce can now be available on an AWS cloud. (The credit goes to John Cooley's deepchip who leaked the news a few days earlier.). But Wally's overall approach is this is not something that EDA vendors nor their customers are really interested in.
After the session, I was approached by a manager at one of the big EDA companies about the the future of EDA on the cloud. His prevailing view was still that cloud is a pain-in-the-neck for EDA vendors, especially in respect to how to monetize it. As if providing the exact software licenses when people actually need them is bad business, and the EDA industry should continue to push for customers to hold software they don't need just for the sake of it - because it's good for the EDA business.
In my view - I told the guys - this is expected to change: the cloud, or more generally, the share economy for that sake, is changing so many industries! Today people and companies tend not to hold a computer system, car, room, bicycle and even umbrellas un-utilized if one doesn't need it - you let other people use it and get paid for it. Or there are companies that their business model is based on sharing such resources among different users. Bottom line, you use what you need, when you need it, for as long as you need it - and pay (only) accordingly.
(Umbrella sharing system in Singapore - Photo: Esther Leong/TODAY)
'But why would we - EDA vendors - offer such services?', they asked. The answer is twofold: First because if there is customer demand, one of the vendors will start offering it, and if the others will not follow - they will be left out. That's what happened when Cadence introduced "on-demand" EDA cards many years ago - giving customers a better use of their investments - which was quickly adopted by other vendors. Likewise, we once paid loads of money for a cellular subscription and once some of the vendors started offering better/cheaper/more flexible packages, vendors who didn't catch up, were left out.
Secondly, played right, EDA on cloud can actually be good for EDA business! In fact, with the increased flexibility of EDA cards, customer ended up buying more software because they now could use more or less licenses at different stages of the project. A truly flexible EDA on cloud business model can have a similar effect of increased business and usage, just like people like me increased usage of taxi rides due the flexibility and ease-of-use of an UBER-like application and service.
The fact is that during the DAC conference that took place late June in San Francisco, Cadence was the first major EDA vendor to announce a broad cloud offering as the "future of electronic design". This is really a great step, as finally the EDA industry is acknowledging the change and making initial steps to adapt.
Picture source: Cadence newsletter
However, in my view, two major benefits of the cloud setup are still not properly addressed:
First of all, it seems the whole approach is "whatever we were doing until now with on-premises IT systems, let's move it to the cloud". It's like the story when Graham Bell invented the phone, people said - "great, now we can send telegrams more effectively!" without fully appreciating the potential of the new innovation.
The beauty of using the cloud is its unlimited elasticity. For example, one of the interesting business models offered by public cloud vendors allows bidding for "Spot Instances", i.e. extra CPU capacity available on the cloud. For example, one may pay X for 100 CPUs/month, but due to a momentary extra availability of additional CPUs, the customer may be able to tap into 1,000s of cores for short periods of time for a fraction of the price. This could do wonders to accelerate the verification regression suite of a project and shorten timelines. But do EDA licenses support this model? Usually, the shorter the license period is, the highest the cost per license is, so it might be counter-productive. (There is an interesting pioneering offering by Metrics where one can access System Verilog simulation and pay by the minute - but this is not the big 3 and it's only functional verification).
But beyond that: The leading cloud vendors don't offer just CPUs and storage space (often called IaaS - Infrastructure as a Service). On top of that, they offer and enable a range of "services" and "applications" in different areas, such as database management and queries, big data, and machine learning to name a few (picture source: AWS site).
Leveraging such features of public clouds can take chip design to a whole new level. The Cadence announcement is a great start, but we're just starting to scratch the surface of the future of chip design beyond the cloud.