CAD computation for manufacturability: can we save VLSI technology from itself?

Every 18 to 24 months, the areal density of VLSI doubles and the predicted date for the End Of CMOS Scaling is pushed out approximately 18 to 24 months. This rate of growth has been controlled mainly by the increasing capabilities of lithographic patterning. However, the rate of improvement of lithography systems in key physical parameters such as illumination wavelength has begun to slow. To make up the shortfall, lithography has increasingly turned to using CAD tools to transform the geometric shapes in designs into shapes on photomasks which have been compensated for systematic pattern-distorting effects. As the requirements for this compensation grow, however, it becomes increasingly difficult to hide in post-tapeout data preparation, and must be considered in back-end design tools and methodologies. We describe a number of the challenges to lithographic patterning, highlighting the factors that limit "physical scaling" and introduce the layout-to-mask shape transformations that compensate for these limitations. We describe the implementation of these transformations in general-purpose and specialized CAD tools, pointing out challenges like growth of computation effort. Finally, we describe how limitations of post-tapeout compensation drive the need for "litho-aware" physical design tools, showing examples in cell design, place-and-route, and layout migration.