Research On Improving 3D IC Yield Based On TSV Fault-Tolerance

Stacking multiple chips in vertical direction,three-dimensional integrated circuits(3D IC)realize electrical connections between multi-layer chips by through-siliconvias(TSVs),and package them into a single-or multi-functional whole,which greatly improves the integration of transistors.Compared with traditional two-dimensional integrated circuits that are gradually approaching the limit size,three-dimensional integrated circuits support heterogeneous integration,and have many advantages such as high interconnect density,high-speed communication,small package size,low cost,and low power consumption.Therefore,three-dimensional integrated circuits are considered to be a continuation of Moore's Law.However,due to the immature process technology,various defects are easily generated in TSVs during the manufacturing,using and aging of three-dimensional integrated circuits.These defects may affect electrical parameters of TSVs and even cause open or short circuit faults in TSVs.In addition,during the stacking and operation of three-dimensional integrated circuits,if a TSV is defective,its adjacent TSVs are likely to be defective,due to the influence of wafer roughness,curvature,and thermal stress between chips.TSV faults appear to be clustered,which greatly reduce the yield of TSVs,thereby reducing the yield of threedimensional integrated circuits.In order to repair the clustered TSV faults and improve the yield of TSV,adding idle redundant TSVs into the TSV block is considered in this dissertation.After TSV faults are detected during the test of TSVs,the TSV fault-tolerant architectures replace the failed TSVs with redundant TSVs to transmit signals,so as to achieve the purpose of repairing the faulty TSVs.The main contributions of this dissertation are as follows:(1)For grid-type TSV blocks,a new redundancy architecture for TSV clustered faults based on interval grouping is proposed.First,the grid-type TSV block is divided into several groups according to a certain rule,and the positions of redundant TSVs in each group are determined;then the redundant paths for each group are designed,so that the signal TSV fault can be tolerated by the RTSV in the same group;finally,multiplexer chains are utilized to realize the sharing of all redundant TSVs.The redundant repair paths of the architecture are very flexible,which makes a significant contribution to the high repair capabilities.The experiment results show that the proposed architecture exhibits repair rates of more than 99.5% for both uniformly distributed and highly clustered TSV faults,under the premise of low hardware overhead and short latency.Compared with router-based,ring-based and switchshifting redundancy architecture,the repair rate is increased by 27.5%,62.7% and 11.4%,respectively.In addition,in the severest clustered situation,the repair rates of the proposed architecture remain close to 100%.(2)For honeycomb-type TSV blocks,a new TSV redundancy architecture for clustered faults based on diamond burster is proposed.The honeycomb-type TSV block is divided into several TSV groups by diamond burster,and corresponding repair resources are configured for each group,so as to disperse the clustered TSV faults and repair them one by one.Multiplexer chains are used to share the repair resources,so as to make full use of all redundant TSVs.According to analysis and evaluation,this architecture consumes low hardware overhead and short latency.Experimental results show that the repair rates of the proposed architecture for clustered or uniformly distributed TSV faults maintain above 99%,far higher than similar method,which indicates that the proposed architecture is well applicable to repairing clustered and uniformly distributed TSV faults.