Heard at DAC: Management panel debates how to organize a global design team

The panel discussion at the conclusion of the Management Day at the Design Automation Conference this week began with a hard look at the way design and verification teams are organized. Asked for a solution to the growing resource crunch on teams, Ken Wagner, vice president of engineering in the Communications Products Division of PMC Sierra asked for people-but people with the right skills, located in the right places. He cited his experience in trying to build a 24/7 global design team.

“Transferring a block around the world with the sun doesn’t help,” he warned. “The information flow between segments of the design team is too complicated to just do a hand-off at the end of the work day. People have to talk to each other. Having the team split between the US and India is just about worst-case-they are never all at work at the same time.” Wagner suggested partitioning the design itself across sites, with each team self-contained, rather than trying to partition tasks across sites.

Intel strategic technology solutions manager Michael Jassowski agreed. He said that Intel was often forced to employ multiple sites by the sheer size of its designs. The company employed modularity-each site would design a self-contained module, defined so that the interfaces between modules were readily-verifiable standard interfaces. Any cost in silicon overhead was more than justified by improved control over the design process. This organization required each site to be relatively self-sufficient in skills, Jassowski said, but it still allowed for individual centers of excellence.

Jitu Khane, director of central engineering at Applied Micro Circuits, uses a slightly different approach. He said Applied also relied on modularity, but would separate design from verification. “If design is in San Jose, and verification is in Viet Nam or India, a problem that the verification team finds during the night the design team can fix the next day,” he explained.

The conclusion from these observations seems to be that the structure of a globally-dispersed design team should resemble the structure of the SoC: modular, with clearly-defined interfaces between the modules. The work flow can either be localized to within design sites, or it can be pipelined. But the team should be organized to minimize the need for high-bandwidth interactions across large distances-geographic or cultural-so there is less chance for inter-team delay and misunderstanding to impact the schedule.