Detection And Diagnosis Of Through Silicon Via Multi-fault In 3D Integrated Circuits

Three-dimensional integration technology uses through-silicon vias(TSVs)that penetrate the silicon substrate to achieve electrical connection between chip-level devices to complete one or more functions together.The existence of TSVs gives three dimensional integrated circuits(3D ICs)the advantages of high density,low latency,and low bandwidth,as well as provides a possible technical solution for solving the interconnection problem and continuing Moore's Law.Because the manufacturing process of TSV at this stage cannot meet the current demand,during the manufacturing process,it is prone to cause various defects.The article carries out modeling analysis,fault test and fault diagnosis on a resistive open fault caused by void defect in TSV center conductor,a short fault caused by pinhole defect in insulation layer and multifault formed when both occur simultaneously.The faulty TSV will reduce 3D IC yield and performance reliability,so it is very necessary to test the TSV on pre-bond.Traditional probe test can cause artificial stress damage to TSV.This paper proposes a non-probe TSV compound faults test method,the main research work is as follows:1.The void and pinhole in the center conductor of TSV are used as research object,establishing 3D simulation models of TSV resistive open fault,short fault and compound faults in the three-dimensional full-wave electromagnetic simulation software(HFSS).According to different fault types,the impact of changing the structural parameters of the faulty TSV on the signal transmission is analyzed by the S parameter.Based on the high-speed circuit theory and parameter extraction method,the equivalent circuits of these three faults are established in the high-speed circuit simulation software(ADS),and the respective S-parameters are obtained.The accuracy of each equivalent circuit model is verified via the S-parameters obtained in the simulation environment of the two.2.A method of TSV non-probe test based on ring oscillator is proposed.It takes the pending test TSV as the load of the ring oscillator,and takes the ring oscillator's oscillation period and duty cycle as the test parameters.Due to the various parameter changes caused by different fault types,available from the simulation results,this method can detect multiple fault types to a certain extent and expand the coverage of fault test.3.The machine learning is combined with the ring oscillator test structure,and the oscillation period and duty cycle of different faults are used as feature vectors,intelligent algorithms are used to detect the fault feature sets.According to Particle swarm optimization(PSO)to optimize Least squares support vector machine(LSSVM),this paper proposes a TSV fault diagnosis model.As PSO is prone to local optimization and premature convergence problems,it leads to misjudgment of TSV fault.Based on this situation,W-PSO that dynamically changes the inertia weight is used to optimize the LSSVM,and another fault diagnosis model is proposed.The simulation results show that the improved fault diagnosis model is superior to PSO-LSSVM in diagnosis accuracy,and the test method proposed in this paper can effectively diagnosis resistive open fault,short fault and multi-fault of different degrees.