A New Accurate Yield Prediction Method for System-LSI Embedded Memories

The authors propose a new accurate yield prediction method for system-LSI embedded memories to improve the productivity of chips. Their new method is based on the failure-related yield prediction method in which failure bits in memory are tested to see whether they are repairable or not by using built-in redundancies. The important concept of the new method is called “repairable matrix” (RM). In RM, = 1 means that i row redundancy sets and j column redundancy sets are needed for repair, where is an element of the matrix. Here, RM can indicate all the candidate combinations of the number of row and column redundancy sets for repair. The new yield prediction method using RM solves two problems, “asymmetric repair” and “link set.” These have a significant effect on accurate yield prediction but have not yet been approached by conventional analytical methods. The calculation of yield by the new method is demonstrated in two kinds of advanced memory devices that have different design rules, failure situations, and redundancy designs. The calculated results are consistent with the actual yield. On average, the difference in accuracy between the new method and conventional analytical methods is about 5%.