Reticle Floorplanning and Simulated Wafer Dicing for Multiple-project Wafers by Simulated Annealing

As semiconductor process technology relentlessly advances into deeper submicron feature sizes following the Moore’s Law, the cost of mask tooling is growing inexorably, up to 1, 1.5, and 3 million dollars for 90nm, 65nm, and 32nm process technology, respectively (LaPedus, 2006). Basically, the majority of smaller fabless integrated circuit (IC) design houses can hardly afford to have one mask set per design just for prototyping or low-volume production. In this circumstance, multiple project wafer (MPW) fabrication (or called shuttle run), long being used as a low-cost mechanism by the academics or industries (Pina, 2001; Morse, 2003) for prototyping their innovative designs, has become an indispensable chip fabrication vehicle. By way of an MPW program, the mask cost can be amortized among various designs placed in the same reticle (i.e., the same mask). Despite of assuming a lower mask cost per design, MPW requires each design to share more wafer fabrication cost. To minimize MPW wafer fabrication cost, the chips participating in a shuttle run should be properly placed in a reticle. This gives rise to the reticle floorplanning problem. Moreover, the wafers must be properly sawn to maximize the dicing yield. This gives rise to the simulated wafer dicing problem. In this chapter, we propose several approaches based on simulated annealing (SA) to solving reticle floorplanning and simulated wafer dicing problems. Since SA’s introduction (Kirkpatrick et al., 1983), it has played an important role in electronic design automation (Wong et al., 1988) such as circuit partitioning, placement, routing, etc. Many commercial physical design tools of this sort often employ SA as the last resort to optimize a design. The reasons for using SA are due to its ease of handling hard-to-be-satisfied constraints by transforming them into part of the objective function and a higher probability of finding a global optimum solution enabled by the capability of escaping local optima in practical implementations. Besides, an objective function for SA can be non-analytic or even does not have a closed-form expression so that it can only be evaluated using a non-calculus approach. Our simulated wafer dicing problem, though not having any hard-to-be-satisfied constraints, has a non-analytic objective function which makes SA quite suitable for solving this problem. Our reticle floorplanning problem has an even more difficult objective function which is the number of wafers required to be fabricated for a shuttle run and can only be evaluated using simulated wafer dicing. Despite of being able to handle nonO pe n A cc es s D at ab as e w w w .ite ch on lin e. co m