Impact of Embedment of Cu/TaOx/Ru on Its Device Performance

Wednesday, 4 October 2017: 10:20
Camellia 4 (Gaylord National Resort and Convention Center)
M. Al-Mamun (Virginia Tech), S. W. King (Intel Corporation), and M. K. Orlowski (ECE Department Virginia Tech)
In our previous work (ECS Trans.75 (32), 13-23, 2017), we have compared the electric performance of Cu/TaOx/Ru and Cu/TaOx/Pt devices finding that the performance of the Ru device is far inferior to that of the Pt device. The poor resistive switching (RS) properties of the Ru device were imputed to the greatly degraded inertness properties of the Ru electrode as a stopping barrier for Cu drift-diffusion. The degraded inertness of the Ru electrode, in turn, has been attributed to larger Cu surface and/or Cu bulk diffusion in Ru than in Pt. The result is surprising as both Pt and Ru are known to be excellent diffusion barriers for Cu in CMOS backend applications. Here, we have manufactured two nominally identical Cu/TaOx/Ru devices, though, differently embedded on the Si wafer. While Ru device A has been manufactured on oxidized Si wafers with 730 nm SiO2 followed by 20 nm of Ti, 55 nm of Ru, 25 nm of TaOx, and 150 nm of Cu, Ru device B has been manufactured in the same way except that a 30 nm TaOx layer has been inserted between SiO2 and Ti layers. Electrical analysis shows that the RS behavior of the two devices is markedly different. Thus the way, nominally identical RS devices are embedded into a technology module has a major impact on the intrinsic device performance. We have identified three major reasons for this disparity. First, the layers of Ti and Ru involved are very thin (20nm and 55 nm). Second, during resistive switching local temperatures at the filament during the form/set and reset operations can exceed 600oC i.e. far higher than the maximum temperature which CMOS backend may experience when the underlying processor is in operation. This temperature is estimated to be ~120oC. Indeed, up to 120oC and above, amorphous Ru remains an excellent diffusion barrier for Cu. The third reason is the crystallinity phase of Ru thin layers that depends on the deposition method and the subsequent ‘unintended’, i.e. related to the RS, local anneals.

We find that Ru device A can be switched a limited number of times at high compliance currents Icc (~0.5mA) while the device B at such Icc levels shows either volatile filament formation or when a filament forms, it is permanent. Conversely, at Icc of ~5uA Ru device B shows fairly good RS behavior, while Ru device A shows mostly volatile behavior. Besides Icc, the other critical parameter is the ramp rate rr during the voltage sweep. Both parameters determine effectively the thermal budget during the set and reset operations and are thus responsible for: 1) chemical reactions taking place between the thin metal and dielectric layers and formation of chemical compounds. 2) Cu surface diffusion on the Ru electrode and Cu penetration through the Ru electrode which appears to include compound formation of Ru2Si3 and Cu3Si in case of Ru device A. Therefore, the declared purpose of inserting the TaOx layer between SiO2 and Ti was to prevent those silicidation reactions. In this work, we show that reports in the literature on a number of Ru metallization issues such as: a) thin Ru amorphous films crystallize into a polycrystalline phase at ~550oC; b) observation of a strong diffusive penetration of Cu through the increased number of grain boundaries within Ru layer at 475oC; c) Ru transforms into a less dense Ru2Si3 at ~500oC; d) Ru2Si3 triggers Cu3Si formation at low temperature (~200oC) and degrades the barrier functionality of Ru electrode, are indeed correlated with the differences between the device I-V characteristics of the two Ru devices.

Because of the chemical reactions triggered by high local temperatures during cell switching, the Ru devices show new properties over the stable Pt devices. In contrast to Pt devices, the voltage ramp rate has a major impact on the resistance of the on-state, Ron, in the Ru devices: lower ramp rate leads to a higher Ron values because the forming filament is exposed for a longer time to elevated temperatures and suffers from Cu out-diffusion. The observation that Ron in Ru device A is significantly higher than in the Ru device B at the same Icc and rr values, demonstrates that Ru electrode in Ru device B has better inertness properties than in the Ru device A. In summary, this work shows that nominally same device with excellent characteristics in a specific experimental environment, may show vastly degraded characteristics when embedded into a module such as CMOS metallization backend – an issue that has largely been disregarded so far.