Engineering High-K Switching Devices for in-Memory Computing

Selecting and processing of appropriate stack materials for HfO 2 -based RRAM devices significantly influences the distribution of defects and oxygen vacancies (V o ) within the switching layer, thereby playing an important role in their switching behavior (1). Controlling this distribution to have more oxygen vacancies near the top electrode (TE) while reducing them near the bottom electrode (BE) results in reduced switching power and enhanced multilevel cell characteristics (MLC) of RRAM devices (2) and that makes them as promising options for in-memory computing. Since RRAM devices fabricated as metal-insulator-metal stack, exploring the impact of each layer remains an active area of research. In this work, we compared different devices by varying the materials of each layer. The first comparison is by varying the oxide layer with same TE (Ru) and BE (TiN) by (i) using stoichiometric HfO 2 and (ii) employing HfO 2 treated with hydrogen plasma at the midpoint. The device treated with hydrogen plasma consumed nanowatts power compared to the stoichiometric device that consumed hundreds of microwatts to be switched. The introduction of hydrogen plasma treatment at the midpoint of the HfO 2 layer redirects more V o towards the TE, explaining the reduction in the switching power (1). While applying pulse SET operation, the hydrogen plasma-treated device exhibited excellent conductance quantization, whereas the stoichiometric device quickly saturated and remaining stuck in a low-resistance state (LRS) due to a lack of V o (3). Introducing a bilayer Al 2 O 3 /HfO 2 dielectric significantly decreased switching power consumption to a few microwatts range compared to the stoichiometric device. This reduction is attributed to the rapid formation of V o within the Al 2 O 3 layer, subsequently stimulating V o formation within the HfO 2 layer (4). However, the hydrogen plasma-treated device still maintains the lowest power consumption. Further investigation involved in changing the TE material, Ru and TiN to understand the impact of the TE, given its working function and interface barriers significantly affect the distribution of V o near the TE. The device using Ru as the TE switched with a 3nA compliance current (1). Introducing nitrogen plasma treatment to the bottom electrode of the TiN device improved its switching behavior due to the reduction V o near the BE because of formation of the Hf-N bond and led to increase the V o near the TE (5). This enhancement was marked by a reduction in switching power to picowatts and stabilization of the conductive filament within the oxide. This device demonstrated improved conductance quantization. REFERENCES Patel, Y.; Misra, D.; Triyoso, D.; Clark, R.; Tapily, K.; Consiglio, S.; Wajda, C.; and Leusink, G., "RRAM devices with plasma treated hfo2 with Ru as top electrode for in-memory computing hardware," ECS Transactions , 104(3), p. 35, 2021, DOI: 10.1149/10403.0035ecst. Wong, H.-S. P.; Lee, H. Y.; Yu, Sh.; Chen,Y.; Wu, Y.; Chen, P. Ch.; Lee, B.; Chen, F.T.; Tsai, M. G., "Metal–Oxide RRAM," Proceedings of the IEEE , 2012, DOI: 10.1109/JPROC.2012.2190369. Zeinati, A., Misra, D., Triyoso, D.H., Clark, R.D., Tapily, K., Consiglio, S., Wajda, C.S., Leusink, G.J. "Process Optimization to Reduce Power in hfo2-Based RRAM Devices for in-Memory Computing," ECS Meeting Abstracts , vol. MA2022-02, 806 (2022). Doi: 10.1149/MA2022-0215806mtgabs Misra, D.; Zhao, P.; Triyoso, D.; Clark, R.; Tapily, K.; Consiglio, S.; Wajda, C.; Leusink, G., "Dielectrics and Metal Stack Engineering for Multilevel Resistive Random-Access Memory," ECS Journal of Solid-State Science and Technology , 9, 05300, 2020. DOI: 10.1149/2162-8777/ab9dc5. Wang, M.-H.; Chang, T.-Ch.; Shih, Ch.-Ch., "Performance improvement after nitridation treatment in hfo2-based resistance random-access memory," Applied Physics Express , 2018. DOI: 10.7567/APEX.11.084101.