To 32nm and beyond: a design-manufacturing symbiosis
11/01/2008
At the 32nm node, technology enablers such as double-patterning, strain enhancement, advanced OPC, etc., require design for manufacturing (DFM) to support a design through manufacturing (D2M) methodology that is correct-by-construction. Such a D2M flow primarily consists of three stages ??? build, verify, and correct. The build stage incorporates manufacturing know-how and constraints such that designs are built to prevent subsequent issues in manufacturing. But as designs are built and integrated, the key is to verify and detect potential areas of yield or functionality concerns. This verification, in turn, drives a design correction or a design-specific manufacturing optimization. At advanced technology nodes, complex and demanding designs may go through this flow iteratively until yield goals are achieved.
In the specific case of double patterning (DP), for instance, layouts will need to be free of native DP splitting conflicts at the build stage of library or memory-bit cell design. At every subsequent stage of physical design such as block formation, placement and route, up to final chip integration, such automated DP-compliance will be necessary. Furthermore, post design tape out (TO), DP data decomposition will need to consider proximity and lithographic effects and guarantee a manufacturable process. Finally, DP-specific silicon analysis has to be done at TO such that design and DP-specific yield concerns are pro-actively highlighted to manufacturing. In other words, a successful and timely transition from single to double patterning would require a comprehensive and holistic view of the D2M flow.
Manufacturing guardbanding methodology, used for many process technology generations has also been scrutinized beyond the 65nm node. Although large guardbands minimize design dependence on manufacturing variations, they negatively impact technology entitlement in terms of area savings and performance. Moreover, rule-based approaches leave benefits on the table, or are just inadequate. Consequently, model-based approaches that simultaneously optimize accuracy and performance have gained rapid acceptance. Statistical modeling; hot-spot analysis for process effects such as stress, CMP, and etch; and computational lithography are examples of model-based approaches increasing in use.
Previous generation design-for-yield (DFY) methodologies relied primarily on random defects and isolated DFM hotspot analyses, modeling them as independent yield loss mechanisms. However, in the case of advanced technology nodes, yield loss can be predominantly due to design-specific systematic defects. Therefore, yield issues need to be more effectively understood, modeled and mitigated in the design flow itself. Alternately, these areas of concern need to be translated to design-specific marching orders for manufacturing. For example, silicon analysis of designs can be used to automate the extent and granularity of the CD-SEM inspection process resulting in increased productivity and higher yielding products.
Next-generation DFM will need to focus on several areas to meet the challenges posed by advanced technology nodes. First, a higher level of integration will be needed to capture not just individual effects based on point tools, but also interactions of multiple process steps. EDA software to support such integration will need to be modular, scalable, and easy to use while rich with automation and computational modeling. DFM will have to drive designs to make process-aware choices, such as in the areas of lithography, stress, etch, CMP, and reliability. DFM tools relying on model-based approaches for advanced nodes will be heavy on compute demands and will therefore need to leverage distributed processing.
So, are we there yet? The answer is a mixed bag. DFM is still in its infancy, where it means different things to different people. Usage of statistical and compute-intensive approaches is on the rise. Although designers are not expected to be process engineers and vice versa, there is clearly an osmosis of each others’ care-abouts and “pinch points.” DFM is no longer just a best-practices objective.Studies show quantifiable benefits to yield or product delivery time. But just as the industry buttons up solutions for 32nm, the 22nm node will bring in new challenges pushing this design-manufacturing symbiosis to the next level.
 |
Contact Milind Weling, at Cadence Design Systems, 555 River Oaks Parkway, San Jose, CA 95134, USA; ph.: (408) 914-6139; email [email protected].