SAN MATEO, Calif. Debate continues on whether a single universal data model can accelerate the lithography process for ever smaller circuits. A recent panel on design/process integration at SPIE 2004 representing users and EDA vendors offered something between a panel discussion and a staged endorsement of the Silicon Integration Initiative's Universal Data Model program.
There was uniform consensus on the nature of the problem UDM is meant to solve, and remarkable consensus on the solution perhaps aided by the absence of Synopsys and Mentor Graphics from the discussion.
The problem was succinctly outlined by the first panelist to speak, Motorola director of design-to-manufacturing solutions Warren Grobman. Describing the complex interactions that now occur between physical design and mask-making, Grobman observed that it is no longer feasible to build the silicon before getting feedback about the manufacturability of the design. Instead, he argued, designers and mask-makers and eventually process integration engineers must have common models of the lithography process that all can share.
That was the goal of UDM, according to the second speaker, SI2 fellow Don Cottrell. Cottrell described how a group of interested parties had come together under the aegis of SI2 to extend work already begun at Semi. "We are trying to springboard off the Open Access program," Cottrell said. The aim, he claimed, was to extend the set of standard APIs and data models now in Open Access to include models that would permit bi-directional communication between design and mask-making.
"This is the low-hanging fruit-something tangible that looks achievable," Cottrell said. "Our goal is to lower the design and manufacturing costs related to lithography, to improve qualification and to provide better, more accurate data management in the design-mask interface."
Then, avoiding the entire issue of whether a unified data model was the correct approach and whether Open Access was the best platform to build on, the panel turned to the question of where to start.
Luigi Capodieci of AMD addressed the question through a rather obscure reference to Pierre Abelard's "Logico Ingredientibus." Capodieci said that he asked designers at AMD what they needed first from a UDM effort, and they expressed indifference to the whole idea. So Capodiece suggested focusing first on a few concrete design-for-manufacturing applications that could do useful things with the model. These, he said, should be built at Universities on a contract that stipulated no IP retention by the university and no IPO for the team that did the work. "We need no more Numeritechs," he quipped.
Rahul Goyal, director of EDA business at Intel, took a skeptical view. "Designers will not adopt a new flow just because it works a lot better," he warned. "If UDM is going to succeed, it will have to be the only way to get the design done within constraints." Goyal urged an initial limited focus on enabling "lithography-friendly design."
"We must demonstrate concrete value to design teams, and that means working applications," he concluded.
Jim Hogan, general partner at Telos Venture Partners, took a similar view. "This will work if we get rid of pain for the designers," he said. "The industry is moving from a wafer-cost model to a good-die cost model, and that represents an opportunity for us." Hogan concluded that, in his estimate, Cadence had spent $25 million over three years developing the database that underlies Open Access. "No one company has the resources to undertake something like that now. It will have to be a joint effort."
Eric Leavitt, chief architect on the Open Access change team at Cadence-essentially the gatekeeper to the Open Access definition-argued that Cadence had spent $20 million, not $25 million on developing the software. He then gave a pitch for the program, saying that chips have already taped out using an Open Access-based flow, and that the elimination of file translations and job decks brought about by the database would result in dramatic savings.
Questions from the audience predictably zeroed in on the notion that a single universal data model was feasible when there are still not accurate local models of advanced resolution enhancement technology (RET) and lithographic processes. It was also asked if a single standard API implemented in code controlled by one major EDA player would stifle entrepreneurial development, and for that matter whether Cadence's data base was the right basis for such drastic extensions as are envisioned by UDA.
The panel responded with defenses of the current plan, to extend Open Access using the Cadence database into the realm of design-mask interactions.