A process-variation-cognizant efficient MOL and BEOL TDDB evaluation method

With increasingly complex manufacturing processes in advanced technology nodes, time-to-breakdown (Tbd) of MOL and BEOL dielectrics are significantly affected by dielectric thickness variations across the wafer. such variations can affect Tbd distributions in an adverse manner such that Tbd distributions can be multi-modal non-Weibull. Also, a very large magnitude difference in Tbd can be observed for a given stress voltage. When TDDB samples are randomly chosen from a given wafer for different stress voltages, the amount of thickness variation can vary from one sample group to another. As different groups of samples are subjected to different stress voltages for TDDB model parameter extraction, it is critically important for groups of samples have similar amount of thickness variation to minimize possible skewness effect on Tbd distributions of sample groups. We present efficient TDDB evaluation methods to optimize test time and improve accuracy of TDDB model parameter estimation. The proposed methods include four techniques: (1) dual ramped voltage stress (DRVS) for estimation of voltage acceleration and Tbd prediction prior to TDDB stress for optimal choice of TDDB stress voltages, (2) Vbd-based sample grouping to minimize skewness of thickness variation effect among test sample groups, (3) TDDB stress sequencing to optimize test duration, and (4) prediction of TDDB area scaling behavior with 7-parameter distributed bimodal clustering model to treat bimodality of Tbd/Vbd distributions.