A BKM by Analytical TEM to Effectively Characterize Electromigration Voids versus Grain Orientations in the Integration and Evaluation of Subtractive Ru BEOL Interconnect

Tirelessly exploring revolutionary integration schemes in driving the miniaturization of transistors is a constant theme in Semiconductor Research, and Technology Development (TD), towards “more Moore”. With the FinFET transistor technology nodes shrinking to the physical limit, gate-all-around (GAA) architecture opens a new era, especially in high-end applications, e.g., Artificial Intelligence (AI), etc[1]. To effectively deliver electrical performance gained from the GAA device advancement in the front-end-of-line (FEOL), innovations in the back-end-of-line (BEOL) interconnects also must be synchronized to address the associated technological challenges in device performance, such as to minimize RC delay [2-3]. As a Pathfinding, Ru has been under evaluations to outperform Cu damascene processes in BEOL. With intrinsic properties of a low bulk resistivity, and a good oxidation resistance, Ru demonstrates a high conductivity in narrow wires, better electromigration durability, compatible with liner-less integration, etc[2-3]. These entitle Ru suitable for below 5nm advanced technology nodes, thanks to the advent of the extreme ultra-violet (EUV) lithography and subtractive etch for 300 mm wafers that makes possible to directly patterning Ru lines to 10nm~ width. The early reliability assessment (ERA) tests in evaluations of semiconductor device endurance are critical milestones to qualify process maturity before the technology transfer from TD to High-Volume Manufacturing (HVM), such as Electromigration (EM) Stress-migration (SM) and the Time-Delayed-Device-Breakdown (TDDB) [2-3]. Effectively and timely characterizations of defect initiations and propagations habits in shrinkage of BEOL is an inevitable task for failure analysis (FA) to discern, using various state-of-the-art analytical transmission electron microscopy (TEM) techniques. Challenges are associated with how to quickly and effectively differentiate and pinpoint the true defects, such as early stage of nanoscale EM voids initiations, versus grain orientations induced TEM / STEM images contrast variations. These efforts will provide guidelines to drive process improvement for yield learning and yield enhancement during TD.
In this paper, we report a Best-Known-Method (BKM) to EFFECTIVE and Quickly differentiate micro voids versus ambiguity from grain orientations. The experiments were conducted on a TEM Metrios by Thermo Fisher Scientific, running at 200KV with a STEM probe corrector well aligned. Figures-1a1b are HAADF-STEM vs. LAADF-STEM images. By toggling between the two STEM modes, diffraction contrast was introduced on purpose. STEM image contrasts vary for crystalline Ru grains, while no change for the true void! Figure 1c is a higher-Mag STEM lattice imaging that undoubtedly convinced the presence of the EM void. Figures 1d1e are an x-ray energy dispersive spectroscopy (EDS) mapping and an electron energy loss spectroscopy (EELS) spectrum, that further proved a crevice initiation, post the EM tests. Due to diffraction effect, if only using low-Mag TEM, it is challenging to reliably judge if a grey-scale feature is due to grain orientations or a void. Even zoom-in to do HRTEM, still having to adjust zone-axis, since crystal orientations vary from one grain to another. The Effectiveness of this BKM is self-explaining by introducing diffraction effects in STEM mode and toggling between HAADF-STEM vs. LAADF-STEM, as aforementioned. Even though this seems contradictory to the Z-contrast concept described in TEM microscopy textbooks. What truly matters most is how to flexibly apply scientific principles to timely resolve the Real-World problems for engineering solutions. Ru’s EM is known to be almost “immortal”, with a much longer lifetime expected than Cu. IBM team confirmed the robustness of Ru against EM. However, for the first time, recently IBM team observed the void formation (missing grain) in the Ru stress line, after the long EM stress tests. More in-depth studies are underway, to be published separately, later.

References: [1] R. Bao, et al., IDEM, (2024), pp. 2-3. [2] K. Motoyama, et al., IDEM, (2024), pp. 39-1 [3] T. Naogami, 2022 VLSI Technology & Circuits Digest of Technical Papers, pp. 423~424.