Process design kit (PDK) developers have been using the SKILL language to code PCells (parameterized cells) for over two decades. SKILL is a flexible, extensible, Lisp-based language that can do just about anything a programmer wants to do. But is there an easier way to create PCells? Yes, according to a presentation at the recent CDNLive! Silicon Valley.
The presentation was given by Romain Feuillette, PCell Team Leader at STMicroelectronics in Crolles, France. He discussed his experience benchmarking the PDK Automation System (PAS) tool and its Graphical Technical Editor (GTE). You can listen to the presentation and download the PDF from the CDNLive! On-Demand web site (see paper titled PCell Generation with the Cadence PAS Tool in the Silicon Realization IV session. Access requires a free Cadence.com account).
I attended the presentation and also spoke with Romain Feuillette separately. The packed room at CDNLive! showed keen interest. This is not surprising, given that PDKs are a foundation technology for all IC design. Easier PDK generation will provide better access to advanced process nodes, and make it easier to improve mature process nodes.
PCell Generation Challenges
In his presentation, Romain Feuillette noted that his group supports 20 different technologies with PDKs, ranging from advanced nodes down to 20nm as well as mature process nodes. There are a number of challenges. One that he spoke about is filling an octagonal ring with vias, a potential "nightmare" that can take a great deal of time. ST has a proprietary algorithm for this, but must then ensure that all possible PCell configurations are DRC clean. "We can spend 90 percent of our time fixing less than one percent of the problems," he said.
Romain Feuillette also spoke of a "disconnect" between the process developers and the PCell developers in Crolles. Today, he noted, the PCell developers manually import data from Excel files provided by process developers. In our separate conversation he explained further. "We get Excel files from the process people, and most of the time we don't have a routine or script to import them, so in most cases a [PCell] developer will just look at the file and write in the PCell. This is a manual process and can cause errors. We are in the process of implementing an improved way to import an electronic format design rule manual [e-DRM], most likely in XML format, straight into the PCell development tree structure to reduce the manual step."
Romain Feuillette's presentation briefly described PAS. The first step is to manually create a GTE file or "book," a document that includes text and graphics. This is the input to PAS, which automatically creates SKILL code for PCells. The main concern is that this manual process of creating a GTE file must be enhanced with an ability to digest an XML e-DRM.
Faster PCell Generation
Pointing to the octagonal ring example, Romain Feuillette noted that a few GTE frames can represent hundreds of lines of SKILL code. For example, filling vias in a 45 degree rectangle used 4 GTE frames instead of 300 SKILL code lines.
"GTE doesn't require a lot of coding," Romain Feuillette said in our separate conversation. "We can have a newcomer use this tool. It's really easy to use compared to coding in SKILL, which requires some experience." Another advantage of PAS, he noted, is that it can automatically import data from Excel CSV files.
But there's still some work to be done. Right now, he said, the display time for GTE frames takes too long. There are also issues related to display.drf file modifications, GTE frame size limitations, and the time it takes to fill polygons with vias. Romain Feuillette noted that Cadence support is working on these issues.
It was interesting to hear from a user who is still in the process of benchmarking a new tool and is considering its advantages and limitations. In this case, it's a tool that could someday have a broad impact on process development.