The effect of defect clustering on WASP device yields

Abstract

Monolithic wafer scale devices have many advantages over hybrid wafer scale devices, but continue to attract far less commercial interest. This is mainly due to the overheads associated with incorporating defect tolerance and re-configuration circuitry into large and complex designs. The WASP series of monolithic wafer scale massively parallel computers have overcome many of these obstacles by adopting a hierarchical defect tolerance strategy. The paper presents an analysis of the effect of defect clustering on the target WASP device yields. The yield model uses extrapolated parameters from earlier WASP devices. and assumes a 0.7 /spl mu/m process. The results show that the target WASP yield predictions are generally very high, and largely independent of defect clustering. However, for high defect densities and high harvest requirements, the predicted yields vary considerably depending on the degree of defect clustering.<>