“I wandered lonely as a cloud,” wrote Wordsworth. Well, clouds are pretty lonely in EDA these days. Despite some of the advantages on paper that mean that companies from salesforce.com to Netflix make heavy use of cloud-computing, semiconductor design has barely touched the cloud. One exception was Nimbic (acquired by Mentor last month).
There are many reasons, and there have been several discussions about this already on semiwiki. The biggest ones to my mind are:
- security: big fabless (and IDM for that matter) companies have policies in place that their intellectual property does not leave the premises. Every time there are reports in the press that passwords have been stolen or other breaches, those policies only get stricter
- all or nothing: the data volumes involved in designs are so large that either the whole design is done in the cloud or none of it is. It is not really feasible to move designs back and forth. Plus trying toe maintain a design environment in the cloud that is identical to the one locally hosted is a herculean task
- commercial terms: to get everything into the cloud requires cooperation with the EDA companies on the terms of licenses, and this hasn’t been worked out yet. If you are going to have computing on demand you also need EDA licenses on demand.
- status quo. the companies who might be expected to dip a toe in the water already have internal CAD groups, internal farms, EDA contracts and adding a few percent of additional cloud resource just adds problems
Instead, the place to start experimenting with the cloud is somewhere that doesn’t have any of these problems:
- no massive IP issues so no massive security concerns
- no existing infrastructure so can get going with a cloud-only approach
- no commercial terms required
- no status quo so can start from scratch
Sounds too good to be true. But it turns out there is somewhere just like that. Universities in countries that don’t really have very much in the way of IC design infrastructure. But universities don’t really have any money so you can’t just go to the universities and sell them a cloud solution.
However, “Nation building” countries such as Singapore, Malaysia, Saudi Arabia, Morocco, Egypt, Brazil, Vietnam, Pakistan have all targeted IC design as a way to move to a knowledge-based economy, move up the value chain and increase per-capita income. They all have government institutions charged with making things happen in this area. With budgets. Today, although these countries all wonder how to get an Intel or equivalent to set up a group there by tweaking tax policy, the reality is that without designers you can’t pay them to go there. Their big weakenss is that their countries don’t graduate enough designers, dozens rather than hundreds or thousands.
Silicon Cloud International has been targeting these institutions with the proposal to set up a complete cloud-based design environment. They will be the CAD organization for the entire academia of these countries. Instead of being paid by the universities (and in these countries they have even less money than in the US or Europe) they are paid by the institutions charged with getting things going. If they graduate lots of designers then there is scope for locally-based entrepreneurs doing things, or for established semiconductor companies to set up design groups there.
The SCI environment includes all the tools from all the usual suspects (who supply them for free). IP and PDKs are all there too. To address security issues they do something that sounds insane at first. You can neither upload nor download anything. As long as the designs are completely created in the SCI environment then there is no need for it. They already have an agreement with MOSIS so you can tape out without requiring to download anything.
To further address security and IP issues, the cloud based design environment can only be accessed through a Chrome-based thin client. After all, very little local computing power is needed, this is cloud-based. There is not even local storage. The clients cost around $200.
Universities in places like the US are also interested but for different reasons. They are perfectly capable of maintaining their own design environment but it doesn’t really add any value to the university for either teaching or research to do so, it is pure overhead.
What stage are they? The prototype has been active since September last year and a pilot program will start late July.
“All at once I saw a crowd,” Wordsworth continued, “a host of golden daffodils.” Well SCI are hoping to host their own crowd…but of students not daffodils.
More information on SCI’s website here.
More articles by Paul McLellan…
Next Generation of Systems Design at Siemens