Multi-Project Reticle Design and Wafer Dicing under Uncertain Demand∗

The pervasive use of advanced reticle enhancement technologies demanded by VLSI technology scaling leads to dramatic increases in mask costs. In response to this trend, multiple project wafers (MPW) have been proposed as an effective technique for sharing the cost of mask tooling among up to tens of prototype and low volume designs. Previous works on MPW reticle design and dicing have focused on the simple scenario in which production volumes are known a priori. However, this scenario does not apply for lowand medium-volume production, in which mask manufacturing is typically started when only rough estimates of future customer demands are available. In this paper we initiate the study of MPW use for production under demand uncertainty and propose efficient algorithms for two main optimizations that arise in this context: reticle design under demand uncertainty and ondemand wafer dicing. Preliminary experiments on simulated data show that our methods help reducing the cost overheads incurred by demand uncertainty, yielding solutions with a cost close to that achievable when a priori knowledge of production volumes is available.