The failure mode of multilayered A1 interconnects changes from shorts due to extrusions to failures due to resistance increase when the overlying dielectric is modified. A thick rigid dielectric favors failures due to exuusions whereas a more flexible dielectric favors resistance increase. The dielectric effectively changes the mechanical behavior of the interconnect without changing the overall kinetics of mass accumulation and depletion and suggests a model where the final mode of failure is determined by the response of the interconnect to mechanical stresses.Interconnects made from A1 alloy thin films sandwiched between refractory metal thin films, typically fail during electromigration due to an increase in resistance (1). However, intralevel and interlevel shorts due to extrusions have also been reported (2) and the type of refractory metal underlying the Aluminum has been shown to play a major role (3). In previous reports the underlying factors that determine the dominant mode of failure, resistance increase or shorts, have not been clearly elucidated. In this paper we show that the mode of failure of the metallization can be changed by modifying the overlying dielectric film, through two different intermetal oxide processes, both of which are widely used in the industry. The results provide a basis for a model that can be used to predict the mode of failures for other situations and allow a reliable metallization scheme to be designed.
Exoerimental detailsThe electromigration studies were carried out on interconnects, consisting of a multilayer stack 200nmTiW/400nm All%Cu/"OnmTiW that formed metal 2 of a 0.61 triple level metallization scheme. Wafers with the metal 2 lines fabricated at the same time were split between two intermetal oxides (IMO) A and B, depicted in fig 1. IMO A was formed by an "etchback" process. In this process 0.61 PECVD oxide was deposited, 300nm of a siloxane spin on glass (SOG) was spun on and the SOG together with any exposed PECVD etched back Approximately, 0.31 of oxide was left on top of an isolated metal line leaving S O G only in small pockets between lines. 0 . 4~ of PECVD oxide was then deposited on top. IMO B was formed by a "non etchback" process. This consisted of depositing 0.251 of PECVD oxide, spinning 300nm of the SOG and depositing 0 . 4~ of PECVD oxide directly on the SOG. Metal lines 2 . 4~ and 1 . 0~ wide and 2.4mm long were stressed at 200°C with a current density of 3 MA/cm2. A comb structure with a spacing of 0.71.1 to the stressed line (SL) was used to monitor extrusions through an extrusion monitor line (EML). Electromigration test circuit schematic is shown in fig 2. A lOKR resistance was kept in series with the extrusion monitor line to limit the maximum current to be drawn by the monitor the moment an extrusion connected it to the stressed line. Thus, the current through the stressed line would remain the same even after a single extrusion was formed. The resistance of the metal line as well as the monitor line was measured and stored as a function of t...