Filamentary oxide-based Resistive Random Access Memories (RRAM) are strong candidates for future Non-Volatile Memory (NVM) because of their simple structure, ease of fabrication, high-speed, high density and scalability [1]. However, the reliance on a highly localized nanometer scale conductive filament through an insulator leads to serious challenges such as the stochastic nature of the resistance state and the stochastic fluctuation of the state’s resistance value after it is programmed. A potentially promising solution to the resistance state control issue is “program-verify”, which checks to make sure that the desired resistance value is achieved before each program operation is considered finished. However, recent reports in the literature suggest that this approach does not work because of post-programming relaxation of the filament [2, 3]. This result is truly problematic and deserves closer scrutiny to understand the root cause. In this work, we investigate in detail the instability of the post-programmed state of HfO₂-based RRAM devices using a newly developed programming setup. Our novel Compliance-free Ultra-short Smart Pulse Programming (CUSPP) method supports program-verify operation, automatic cycling of SET/RESET states at high rates, and continuous resistance state monitoring with nanosecond measurement time resolution.

We investigated the short-term instability of HfO₂-based RRAM devices during programming in the time range of 100 ns to 1 s. Our measurements demonstrate conclusively that it is fluctuation, not relaxation, that significantly degrades the effectiveness of the program-verify approach. While this new insight is valuable for understanding the problem as well as searching for a solution, we uncovered additional troubling details. We observe that the occurrence probability (frequency) of these fluctuations initially decreases (over 100 μs) before reaching a non-zero steady state value that never vanished over the time duration of our study. Even more alarming is the observation that the fluctuation amplitude never decreases within the same time duration. These observations indicate that that techniques such as delay reading or averaging will not relieve this fundamental problem [4].

References:

[1] H.-S. P. Wong, H.-Y. Lee, S. Yu, Y.-S. Chen, Y. Wu, and P.-S. Chen, "Metal-oxide RRAM," Proc IEEE., vol. 100, 2012.

[2] A. Fantini, G. Gorine, R. Degraeve, L. Goux, C. Y. Chen, A. Redolfi, et al., "Intrinsic program instability in HfO2 RRAM and consequences on program algorithms," in 2015 IEEE International Electron Devices Meeting (IEDM), 2015, pp. 7.5.1-7.5.4.

[3] X. Li, H. Wu, B. Gao, N. Deng, and H. Qian, "Short Time High-Resistance State Instability of TaOx-Based RRAM Devices," IEEE Electron Device Letters, vol. 38, pp. 32-35, 2017.

[4] D. M. Nminibapiel, D. Veksler, P. R. Shrestha, J. H. Kim, J. P. Campbell, J. T. Ryan, et al., "Characteristics of Resistive Memory Read Fluctuations in Endurance Cycling," IEEE Electron Device Letters, vol. 38, pp. 326-329, 2017.