Non-volatile memory technologies such as phase change memory (PCM) and magnetoresistive memory (MRAM) have seen various levels of success in manufacturing over the past decade.Intel/Micron’s Optane PCM memory and Everspin’s MRAM are just a few examples of recent products, while in more recent research, these same materials are also being explored as accelerator technologies for AI hardware applications. However, to enable these memory technologies, a significant understanding of plasma processing is required to achieve high fidelity pattern transfer while maintaining device performance. In this talk I will walk through first data highlighting the challenges of material volatility for both PCM and MRAM etching and how process induced modification of the material leads to deleterious device performance.Furthermore, we will review the promise of atomic layer etching in enabling new avenues to address patterning issues for MRAM and PCM and demonstrate how ALE can lead to new opportunities for sustainable future non-volatile memory processing applications.

In this study, we investigated the plasma enhanced atomic layer deposition (PEALD) of silicon nitride (SiN_x) using the VHF-CCP N₂ plasma and different aminosilane precursors at the process temperature range of 100–300 ^oC. The combinations of VHF-CCP plasma source instead of conventional HF-CCP plasma source and an adequate aminosilane precursor allowed a high growth rate and high quality films as well as excellent conformality close to 100% at substrate temperature of 300 ^oC. In addition, when external magnetic field was added during the VHF CCP N₂ plasma generation during the PEALD of SiN_x film, the magnetized VHF CCP N₂ plasma allowed better film quality and higher conformality at low temperature of 200 ^oC.

Continuously shrinking feature size in patterning imposes non-volatile memory processing challenges, particularly for phase change memory (PCM) materials, where damage mitigation is imperative. Optimization of etch process and chemistry in minimizing or eliminating structural or compositional damage is key for the success of this technology. Use of halogens is often needed for a better profile control, reduce redeposition, selective metal etching etc., but it can cause structural damage and elemental depletion of PCM materials leading to an increase in the recrystallization time. Ion sputtering with inert gases on the other hand, can cause material re-deposition on the sidewall, poor profile control and worse etch selectivity.

In addition to ion bombardment which is typically responsible for physical sputtering and increased roughness, plasma can generate short wavelength irradiation due to electron-neutral collisions. UV/VUV photons emitted in a plasma can reach high fluxes and thus become important in terms of plasma-surface interaction processes. Elemental depth profiling with ion beam analysis and time resolved laser reflectivity was done to study the phase transition behavior of GST when exposed to different chemistries, temperatures, plasma duration, and various reactor configurations.

Surface oxidation of PCM materials can substantially alter switching properties therefore, in-situ plasma enhanced CVD encapsulation has been viewed as a favorable solution. It is imperative that RIE and encapsulation mitigate damage and oxidation of PCM material during integration. In-situ encapsulation of GST and tuning of plasma parameters, caused controlled SiN film deposition with simultaneous selective etching and damage removal from GeSbTe-based PCM materials.

The recent development in data science and technology such as machine learning (ML) and artificial intelligence (AI) is now changing the way we perform various tasks, including research and development. Process development for semiconductor manufacturing is so complex that there is much room for improvement in its efficiency by ML and AI. Often one of the major problems in applying ML and AI for process development is the lack or insufficiency of experimental data. One possible remedy is to form “digital twins,”i.e., numerical simulation models of plasmas and plasma-interacting surfaces, and generate a large amount of data that can augment the shortage of experimental data. Even if the simulation results do not agree with experimental observations quantitatively, as long as the simulation data are correlated with experimental observations, such augmented data help us search optimal process conditions and perform a design of experiments. In this presentation, the recent development of data-driven plasma science [1] for low-temperature plasmas and their applications to material processing will be briefly reviewed. Then, more specifically, the prediction of sputtering yields/etch rates of materials by ion beams and the construction of plasma surrogate models will be discussed, which could allow real-time simulation of plasma processing systems for process control and fast survey of optimized process conditions.

[1] R. Anirudh, et al., “2022 Review of Data-Driven Plasma Science” IEEE Trans. Plasma Sci. (2023) to appear/ arXiv:2205.15832