| ANAHEIM, Calif. Industry executives debated IP interoperability at the Design Automation Conference. |
Joachim Kunkel, Synopsys vice president of engineering, ARM vice president of engineering Keith Clarke and Drew Wingard, chief technical officer of Sonics, exchanged views on the problem of IP interoperability. The discussion initially focused on making blocks work together, but ended in what may be the next step in the creation of standards.
Initially defining interoperability as the ability of IP blocks to exchange information with each other, the panelists zeroed in on standards whether bus, fabric or simply standardized wrappers as the underlying issue. Standards are good, they agreed, but they are costly to propagate and to maintain.
"Things that work at the board level, such as plugfests and third-party evaluations, can be a problem for semiconductor IP," said Wingard. "We have to work with extensive models and properties instead."
The process takes time, according to Kunkel. "We started building transaction-level models of the AMBA bus architecture three years ago, and today we have useful models," he said. "In the beginning, we had to figure out what a transaction-level model was."
University of California Professor Alberto Sangiovanni Vincentelli suggested that long, complex standards were frequently laden with ambiguities and outright contradictions.
Kunkel agreed, saying that Synopsys spent a great deal of time working with standards committees to determine their intent regarding implemetation of a standard interface. Wingard went a step further, saying that Sonics was engaged in producing a collection of executable properties that would unambiguously describe their standard interfaces, and including those in the OCP-IP documents.
ARM's Clarke said that while this was a welcome step, it was not a panacea. "Assertions can tell you when something goes wrong, but all the assertions you can dream up won't assure you that everything is right," he warned.
Sangiovanni Vincentelli adde, "Assertions can be a better way to explore the verification space for a block," he said. "But a proof of completeness for a set of assertions is impossible."
Wingard said Sonics recognized this. "To be as complete as possible, we are soliciting from the OCPIP membership all the things that they are checking in their designs," he said. "Then we will encode all of them in PSL."