Every semiconductor company must decide whether to create a library in-house ("make") or outsource library creation ("buy" or royalty). Although this decision was formerly a technical one, short-term financial concerns have changed the process from finding the best solution to spending the least money right now. The economic crisis our industry experienced in 2001 increased interest in royalty-based libraries, but we are beginning to see signs that integrated device manufacturers (IDMs) and fabless semiconductor firms are deciding these libraries are more costly than they seemed to be, in terms of both financial and design impact.
The current situation is reminiscent of the trend in the 1990s when commercial libraries were first introduced. IDMs and fabless semiconductor companies were eager to eliminate the expense of internal library development in favor of buying custom libraries from commercial vendors. The honeymoon was short-lived, however, du e to inflated costs, missed delivery dates, and poor library quality. IDMs and fabless companies recognized their blunder and began taking back control of their libraries by automating the process of layout generation and characterization. As a result, library companies began to struggle. In what many view as an act of desperation, royalty-based libraries were introduced late in 1998, and the commercial library business was changed forever.
Library companies would receive back-end compensation (royalties) from fabs based on the real estate of the end-chip built using royalty libraries. This new business model did two things: it re-energized the commercial library market for both IDM's and fabless companies; and it put competing library companies out of business, creating a virtual monopoly. In reality, not all libraries were offered for royalties. In fact, the library services business still dominates revenues, supporting the theory that royalty libraries are more of a sales tactic than a sustainable bu siness model.
Make vs. buy library decision matrix
Once the dust settled, users began to question the competitiveness of the royalty libraries, and we saw a return to the "make" library market, which lasted through the first half of 2001. The economic downturn quickly changed that trend, and as semiconductor corporate losses piled up, IDMs delayed new processes/libraries and fabless companies with slashed budgets again turned to royalty libraries. This economy-fueled trend is already waning in 2002 with déjà vu all over again, as the semiconductor market recovery is on the horizon, and the design cycle is again revealing the negative aspects of royalty libraries.
Recent case studies have proven beyond a reasonable doubt that optimizing standard cell libraries for design requirements, process technologies and specific EDA tools yields double-digit gains in performance, area and power utilization. For example, simply increasing the drive strength count beyond those provided in the standard royalty libraries improves area and power utilization, and also reduces design iterations by having the correct cell available.
Budgeting for EDA tools has always been a corporate challenge. The lure of heavily discounted or no-up-front-cost products will always be strong, and the results will always be the same: Unless you're designing an IC to fail, the savings promised by "free" libraries are fool's gold indeed.
Dan Nenni is vice president of sales and marketing for IC library provider Copyright © 2003 CMP Media, LLC | Privacy Statement