| SAN JOSE, Calif. — Which is the better model for ensuring silicon intellectual property (IP) quality — an off-the-shelf approach like K-Mart, or one that involves a more intimate customer/provider partnership, like Nordstrom? In a panel at the International Symposium on the Quality of Electronic Design (ISQED) here March 23, sentiment clearly rested with the Nordstrom approach. |
Tim Holden, vice president of marketing at ARM, Ltd., offered a framework for discussing the development and use of IP. "We identify three phases in the IP chain," Holden suggested, "creation, implementation and integration."
Holden said that ARM uses a number of models in engaging with customers. The IP team may do the entire process, including integration, for an internal customer. Or, at the other extreme, they may do only the creation of an IP core, and turn the implementation of a physical design and the integration into an SoC over to a partner.
But in no case is there a simple hand-off, Holden said. Every model is a partnership in which ARM continues to participate. The situation is much like the range of choices at Nordstrom, where a customer may choose to buy a carefully-selected item off the rack, or may consign his fate to a personal shopper. But either way, the retailer's commitment to quality and satisfaction applies.
Joachim Kunkel, vice president of Synopsys' DesignWare engineering group, commented on the importance of IP quality to his customers. "IP needs to be not only functionally correct, but also ready to work with the customer's tool flow and compatible with the rest of their design," he said. "Evaluating all these dimensions is much more than a question of correctness of the original design intent."
Kunkel said that evaluating IP was difficult enough when the IP was a finished product, ready to license. "But often with leading-edge requirements, the IP is being developed in parallel with the chip into which it will be designed," he said. "This is an extremely difficult situation. It requires complete transparency on the part of the IP developer, because the user must understand the developer's methodology and watch the progress of the design in detail."
Kunkel said that the only major difference between such an IP relationship and a joint development partnership was that at the end, the IP vendor would retain some rights to the IP.
Rafi Kedem, senior director of the CoreWare program at LSI Logic, took a more pessimistic view. For digital IP, Kedem said, a tight three-way relationship between IP developer, foundry and SoC integrator is necessary. He made the point that this relationship could not be entirely mutual — it had to be managed by a party with the leverage to make things happen.
But for the most challenging IP, such as high-speed mixed-signal blocks, Kedem argued that trusting a third-party developer remote from the process knowledge was simply not an acceptable risk. Yet while recognizing the need for intimate involvement between parties in the IP chain, Kedem said that the K-Mart option was still necessary in order to restrain prices and to provide mature, stable IP that did not require so much effort.
Bhusan Gupta, director of the ST Microelectronics Berkeley Lab, said that in a market where much of the IP must conform to industry standards or commonly-accepted definitions, quality was becoming the only differentiation. He concurred with Kunkel that the most challenging situation for an IP user was to need a piece of IP that was still under development. "This becomes a partnership, not a licensing agreement," Gupta said.
Mamta Bansal, manager of foundation IP at PMC Sierra, homed in on the question of quality as a differentiator. She pointed out that there is a serious lack of quality metrics that would make comparison of IP offerings possible. She suggested that the work being done by the Virtual Socket Interface Alliance (VSIA) and Fabless Semiconductor Association (FSA) was a promising start.
Finally, Herb Gebhart, director of technical marketing at Artisan Components, offered an independent IP vendor's view. He said the IP business was a matter of balancing the increasing cost of quality IP development with economies of scale. He said that the gradual convergence of digital CMOS processes had helped IP vendors by narrowing the range of processes over which they had to qualify their IP.
But at the same time, he said, the number of variants in each process had increased, and the rate at which new processes changed had increased. These factors made it a struggle for IP vendors to maintain correlation between the IP versions they were delivering and the current processes offered by the foundries.
During the question and answer session, panelists made important distinctions between commodity IP, leading-edge digital IP and analog IP, each of which presented different problems for the evaluation of quality. But they returned again to the theme that IP users cannot escape the ultimate responsibility to assess not only the quality of the IP offered to them, but the capabilities and commitment of the vendor itself.
It was the relationship with the vendor, more than the initial condition of the IP, that determined success in SoC integration, speakers said. In the end, panelists saw little room for K-Mart.