In his latest Industry Insights blog entry, Cadence blogger Richard Goering published a quick summary of the recent OpenPDK (process design kit) Webinar that was broadcast on December 7. A PDK allows an ASIC foundry to publish process data needed by EDA tools to make good decisions about design options. Each new process node and each variant within a process node (low-power, high-performance, etc.) needs a PDK. There are two current approaches to standardizing on PDKs. The approach sponsored by industry standards-setting Si2 is called OpenPDK, which supports multiple delivery formats for process data all based on Si2’s OpenAccess database. The purpose of course is to enable the creation of a wide variety of PDKs for EDA tools that are rooted in a common understanding of a process node. Formats for the PDKs resulting from this process data could include Cadence’s SKILL and Mentor’s Ample languages or the format defined by the Interoperable PDK Libraries (IPL) alliance. In other words, the process data goes into a database and then EDA companies extract that data and use it to create PDKs for their EDA tools. OpenPDK uses the OpenAccess database, but does not require that EDA tools employ that database natively. The database is a container. Cadence and Mentor Graphics are both supporters of this approach.
The other approach is to create a universal PDK—one PDK used by all vendors. The position of the OpenPDK Consortium is that one universal PDK limits the options of EDA vendors and is not a particularly open approach. OpenPDK is very much an example of the cooperation foretold by the EDA360 vision. The universal PDK…not so much.
PDK evolution is not merely an interesting technical aspect to the world of ASIC and EDA tool development. The physics of building chips has become very complex with second-, third-, and fourth-order effects becoming important like never before. Foundries must provide EDA vendors with growing amounts of complex, detailed design data to allow the tools to create proper designs that yield well in the fabs without resorting to the stratospherically costly alternative: respins. Anyone who thinks that things are going to get any easier is mistaken.
For a look at Richard’s review of the Webinar and a link to the Webinar, click here.