This thesis proposes a new approach to the design of reliable VLSI interconnects, based on predictive failure models embedded in a software tool for reliability analysis. A method for predicting the failure rate of complex integrated circuit interconnects subject to electromigration, is presented. This method is based on the principle of fracturing an interconnect pattern into a number of statistically independent conductor segments. Five commonly-occurring segment types are identified: straight runs, steps resulting from a discontinuity in the wafer surface, contact windows, vias and bonding pads. The relationship between median time-to-failure (Mtf) of each segment and physical dimensions, temperature and current density are determined. This model includes the effect of time-varying current density. The standard deviation of lifetime is also determined as a function of dimensions. A· minimum order statistical method is used to compute the failure rate of the interconnect system. This method, which is applicable to current densities below 106 AI cm2 , combines mask layout and simulation data from the design data base with process data to calculate failure rates. A suite of software tools called Reliant (RELIability Analyzer for iNTerconnects) which implements the algorithms described above, is presented. Reliant fractures a conductor pattern into segments and extracts electrical equivalent circuits for each segment. The equivalent circuits are used in conjunction with a modified version of the SPICE circuit simulator to determine the currents in all segments and to compute reliability. An interface to a data base query system provides the capability to access reliability data interactively. The performance of Reliant is evaluated, based on two CMOS standard cell layouts. Test structures for the calibration of the reliability models are provided. Reliant is suitable for the analysis of leaf cells containing a few hundred transistors. For MOS VLSI circuits, an alternative approach based on the use of an event-driven switch-level simulator is presented. / Thesis (Ph.D.)-University of Natal, Durban, 1988.