Get email delivery of the Cadence blog featured here
Representatives of Cadence and Mentor Graphics don't often appear in the same webinar, but there's a new development in the industry that was important enough to bring these two EDA vendors together Dec. 7. That development is OpenPDK, a process design kit standardization effort launched by the Silicon Integration Initiative (Si2) and ten founding members, including Cadence, Mentor and Synopsys, in May 2010.
The webinar was part of the ongoing Cadence Silicon Realization webinar series. Titled "OpenPDK: What is the Goal of this Effort, and What is the Status?" it was presented by John Stabenow, group director for analog product management at Cadence, and Linda Fosler, director of marketing for Mentor's Deep Submicron Division.
Instead of imposing a "universal" PDK, the OpenPDK effort seeks to define a single data source that enables all PDK creation. Foundries publish the source once, and propagate it to all EDA vendors equally. Vendors can then "compile" this information into different PDK formats that are optimized to support each vendor's tools. Formats could include Cadence's SKILL language, Mentor's Ample language, or the Interoperable PDK Libraries (IPL) alliance format. OpenPDK works with OpenAccess, but does not require that EDA tools run natively on it.
Under OpenPDK, a single source can be translated into a variety of formats. Source: Si2
Here are two key takeaways from this webinar:
Who cares about PDKs? You do, if you have anything to do with semiconductor design or manufacturing. PDKs not only enable analog/custom design, but also make digital standard cell design possible. Thus anything that impacts PDKs has ripples throughout the entire IC development chain.
Swappability versus Interoperability
Swappability, Stabenow said, occurs when users can take two tools with identical functionality and interchange them at will. It doesn't provide much value. With interoperability, on the other hand, users can choose differentiated point tools that best solve their problems. "An interoperable EDA flow built on top of a universal data model is a way for our customers to drive profitability," he said.
There is some "buzz" around a universal PDK, Stabenow said, but that concept doesn't really solve the customer's problem. For a single PDK to work universally, EDA vendors would have to have functionally identical tools. "If your goal is swappable tools, that's the methodology you'd like," he said. "But if your goal is best-in-class tools that solve your design challenges in the way you uniquely want to solve them, then this [universal PDK] methodology won't work."
With the OpenPDK single-source approach, he said, foundries can publish process data once and propagate it equally to all EDA vendors. Large and small EDA vendors can then build PDKs in their own unique formats, gaining an "equal playing field." Benefits include increased design choices, faster process ramp-ups, lower customer support costs for foundries, and fewer requirements for retooling. Stabenow noted that the OpenPDK Coalition is now working on a detailed specification.
A "Storybook Myth"
Linda Fosler spoke next. "I agree with everything John said about what the market really wants," she said. "The market wants the freedom to choose best in class tools. And the myth of the universal PDK is only slightly more believable than your average storybook." She went on to say that "I encourage the industry to give up on that particular myth."
Fosler said that Mentor is "excited about the OpenPDK Coalition" and is actively participating in technical working groups. Meanwhile, she said, Mentor formed its own design kit initiative two years ago with the intention of supporting all significant design kit formats.
"We're very pleased to be part of the coalition and pleased to be working closely with Cadence in that coalition," she said. "We believe strongly in collaboration. That's not to say we don't want to compete, but we intend to compete on tool functionality that solves unique problems."
The archived webinar is available here for Cadence Community members (quick and free registration if you're not yet a member).
Interoperable PDK's are a myth? Somebody should tell the world's largest foundry which is fully (and successfully) committed to them.
As a founding member of both the IPL and the OpenPDK Coalition, I believe I can clarify something:
1. The IPL offers a standard for interoperable PCell libraries that can work today in virtually every custom IC design tool in the world. They have been proven in extensive qualifications in multiple companies to provide identical results in every tool because the API works with OpenAccess directly.
2. What the OpenPDK Coalition intends to provide is the ability to CREATE most or all PDKs from a single source, including IPL-compliant PCells for those who want interoperable PCells. Remember, there is still the issue of qualification of each resulting PDK element if interoperable standards are not used.
Let's not confuse "open" with "interoperable." I have not seen any work being done at the OpenPDK meetings that touches on interoperability. What I see being developed is a way for foundries and IDMs to create a single PDK information source that can be converted into the myriad of proprietary PDK elements. While this reduces the workload of companies with multiple consumers of PDK information, it does not preclude the use of truly interoperable standards where they exist. We support the OpenPDK coalition because we believe that an OpenPDK will make it that much easier for users to adopt interoperable PCells and other interoperable standards as they emerge.
The diagram in this article associates IPL with Synopsys in the same way that SKILL is associated with Cadence. This is not correct. The IPL format and standard is supported in tools from Magma, Springsoft, Pulsic, Synopsys, Mentor, Ciranova, AWR, and more.
Synopsys is the largest EDA company in the IPL coalition but it is by far not the largest company in the coalition.
The technology behind IPL is helping semiconductor companies produce ICs on the most advanced nodes in production. No myth.