Electromigration critical length effect and early failures in Cu/oxide and Cu/low k interconnects
Electromigration (EM) reliability was investigated for dual-damascene Cu/oxide and Cu/low k interconnects. In addition to the basic characterization of EM behavior, statistical and physical analyses were performed to understand the effect of low k dielectric materials, interconnect geometry, and process control. In order to improve the EM reliability, EM early failure behavior was investigated using a statistical simulation, which was developed to separate the bimodal failure distributions and treat them individually. EM experiments were performed at temperatures between 190 and 400 o C and at current densities between 0.25 and 3.0 MA/cm2 . Temperature coefficient of resistivity (TCR) was found to be independent of inter-layer dielectric (ILD) materials simply because dielectric material does not contribute to electrical phenomena. However, temperature rise due to Joule heating was evaluated to be higher for low k ILD compared with oxide. The activation energies for Cu/oxide, Cu/porous MSQ (Methylsilsesquioxane) and Cu/organic polymer were found to be in the range of 0.8–1.0 eV, suggesting mass transport of Cu ions at the interface of Cu and SiNx cap-layer regardless of ILD material. This is because all Cu interconnect structures preserve the same Cu/Ta and Cu/SiNx interface for EM. The impact of low k ILD on EM characteristics was investigated using the critical product of current density and conductor length, (jL)c. Compared with oxide, a significantly smaller mechanical strength of low k ILD reduces the (jL)c and the thermomechanical confinement. Smaller thermomechanical confinement in Cu/low k interconnects reduces (jL)c and back-flow stress (∆σ), leading to a faster mass transport and a shorter EM lifetime. FIB observation revealed that extrinsic failure in Cu/low k interconnects can decrease the (jL)c even further. This can be a serious reliability issue. For both Cu/oxide and Cu/low k interconnects, (jL)c is independent of temperature in the test condition. However, an unpredicted current density dependence of (jL)c was observed in Cu/low k interconnects, which is still under investigation. Monte Carlo statistical simulation based on the weakest link approximation was developed to characterize early failures in dual-damascene Cu interconnects. This simulation can separate a bimodal failure distribution into two individual lognormal distributions. Therefore, early failures (the distribution with a smaller t50) can be treated as an independent event; it is now possible to evaluate the EM lifetime characteristics (t50 and σ) and the populations of early failures independently. Statistical analysis suggested that early failures were dependent on the process controls. Failure analysis using FIB and TEM confirmed the bimodal failure behaviors which were first detected by the statistical analysis. Void formation at the cathode via bottom was responsible for the early failures. Multi-via structures can reduce the effect of early failures and improve the interconnect reliability.