Putting Design For Manufacturing Into Action

"Design for manufacturing," or DFM, is such a buzzword these days. Everybody seems to agree that DFM is critically important. However, there is also an ironic consensus that the term DFM is vaguely defined. This is especially true for those in the chip industry who are developing or employing tools to create cutting-edge semiconductors, from physical design through manufacturing of nanometer-scale processes.

A Web search yields many articles that offer opinions on "functional" definitions of DFM: the need for cooperation and collaboration between design and manufacturing communities, accounting for manufacturing effects early in the design flow, exploiting circuit information to make better manufacturing enhancements, etc. While all of these sound sensible, one cannot help but wonder why people aren't yet practicing these disciplines rigorously. Another Web search yields a wide collection of partial DFM applications: via-array generation, routing-for-OPC (optical proximity correction), planarity-fill generation, resolution enhancements, technology CAD, smart mask-fracturing, and so on.

Nevertheless, how each of these applications works with the others to form a true DFM solution remains vague. The problem is a lack of a heuristic, "working" definition of DFM and of key guiding principles that conjoin high-level goals and low-level implementations to spell out what a DFM process is and how it should work. To exploit the missing pieces and take on the daunting task of defining DFM in a heuristic way, we must identify the root causes that hinder the collaboration between design and manufacturing communities. They are embodied in the following observations:

The "languages" employed by the two sides differ fundamentally. The physical design team works on circuit functionality and performance (timing, signal integrity, power consumption, etc.). It is natural for design engineers and their tools to operate on circuit data (represented by nets) and geometry data (represented by polygons). On the other hand, the manufacturing team works on silicon patterning (on-silicon fidelity, uniformity, process window, etc.).

So, manufacturing engineers and their tools naturally operate on geometry data and process data. Thus, the common "language" between the two sides is only geometric and polygonal. This perfectly reflects the established practices of expressing manufacturing constraints in (geometric) design rules and representing tape-out data in (polygonal) GDSII format, which start to show their deficiencies.

Fundamental differences exist between the two sides' design methodologies. The design team performs its tasks hierarchically, incrementally, and concurrently with design reuse. The manufacturing team takes taped-in designs layer by layer, step by step, through sequential operations without much reuse. In other words, design engineers fix a large problem by first solving small problems and then reuse the small solutions to build a large solution incrementally. But manufacturing engineers rigorously tackle a large problem in sequential steps to ensure stability.

Thus, a DFM solution should offer much more than a sequence of disjointed applications invoked throughout the design-to-manufacturing flow. Collaboration requires a new layer of applications to properly "interpret" the exchanged information beyond just geometric data. Moreover, to enable a seamless DFM flow, manufacturing data-preparation operations (such as resolution enhancements) must be invoked in a hierarchical and incremental fashion to fit into a well-practiced design methodology.

Once these are available, the exchanged information—including circuit, geometry, and process data—can flow freely between both sides and be properly used to enable a true DFM foundation. Plus, many sequential, manufacturing-specific operations can be broken into smaller, more manageable pieces and then be reused to form a true DFM methodology. The second-generation DFM companies can figure out the implementation details of the tools.

The bottom line is that a good working definition of DFM should include the concepts of bidirectional communication of circuit, geometry, and process information between design and manufacturing. It must also include hierarchal and incremental methodologies that can be employed by both.