Design and process related MIM cap reliability improvement

Abstract

A root cause failure investigation was performed on anomalous (early) MIM capacitor failures on an HBT MMIC process. These failures were only observed on capacitors in the actual MMICs; process control monitor (PCM) capacitors were nominal. Multiple failure analysis techniques were employed to determine the most probable root cause of the early failures. The root cause was determined to be etching of the capacitor dielectric by a chemical used in the MMIC fabrication process, at a step after the capacitors themselves were fully built. It was determined that only CAD layouts with certain features were susceptible to the etching. These features were not present on the PCM capacitors.A design based corrective action was recommended to eliminate the failure mechanism. The effectiveness of the corrective actions was verified with several designed processing experiments. These experiments also demonstrated that the failure rate of the faulty parts increased with the time (even if the parts were not in operation). Finally, the experiments showed that the reliability of a nominal capacitor with a BCB layer on top was better than that of an identical capacitor without the BCB layer. Capacitor reliability modelFor the sake of completeness, the following discussion on ramped voltage TDDB testing of capacitors is included. In this work, ramped voltage data was not used to predict capacitor lifetimes, but it was used to quantitatively compare capacitor reliability by plotting percent cumulative failure versus failure voltage.Normally, destructive ramped voltage testing and the linear field model are employed in order to estimate the lifetime of capacitors.This method has been described in detail elsewhere for intrinsic capacitors (Cramer et al. [1,2,3]). It is generally assumed that defects on or in the capacitor bottom plate or in dielectric itself causes a localized thinning of the dielectric which shortens the intrinsic lifetime of the capacitor, as illustrated in Fig. 2.Extrinsic capacitors require a refinement of the Nominal Thickness Method to the Effective Thickness Method.Typically, plots of percent cumulative failure versus failure voltage will be transformed into plots of percent cumulative failure versus lifetime via the above equations. The shapes of these two plots are nearly identical, only the values on the abscissa are changed from (linear) voltage to (logarithmic) lifetime. Since, in this work, the effort was directed at root cause determination of anomalous failures rather than predictions of expected lifetimes, the linear field model was never employed and the data was always plotted as a function of failure voltage.