Session:

Oxide-based non-volatile memories have great potential for neuromorphic computing applications.¹ In HfO_xfilamentary memory, when bias is applied, a conductive filament (CF) is formed, within which O ion motion in and out causes rupture and reformation of the CF, and this process primarily governs the switching. However, challenges such as high forming voltage (V_F), abrupt resistance change, and high off-state current prohibit its widespread adoption.²

In this study, we independently assess the impact of Ti doping and electrode material optimization on HfO_x devices with the goal of overcoming the problems in HfO_x filamentary memory. The Ti-doped HfO_x devices were fabricated, where the ~5 nm oxide was deposited via atomic layer deposition. Electrical characterization shows that the V_Fdecreases (~2.1 V) with the increase of Ti doping.³ In addition, analog pulse measurements show that a 25% Ti doping improves the linearity in resistance change during the reset. However, the switching is not substantially impacted, and the off-current is high.⁴ A few O ion motion at the CF break region governs the switching. We determined the location of the break via a direct tunneling electron transport model and a finite element analysis. Both analyses show that the break is close to the reset anode electrode; thus, the electrode has a significant impact on the switching. We hypothesize that a reduction in thermal conductivity (K) of the electrode will lower the off-current. A low electrode will cause less heat removal and more oxygen ion motion, thus will result in a thicker break in CF and less off-current. However, the challenge is that most of the electrodes have high K (e.g., K_Au = 310 W/m.K., K_TiN = 29 W/m.K). Ti₂AlN, like many other MAX phase materials, has unique metallic (high electrical conductivity) and ceramic properties (low thermal conductivity, K_{MAX phase}~4.63 W/m.K) as it contains alternating layers of metallic and covalent bonds.⁵ Thus, HfO_xdevices with Ti₂AlN MAX phase as the bottom electrode were fabricated. Alternating layers of Ti and AlN were first deposited at room temperature via magnetron sputtering and then thermally annealed to obtain Ti₂AlN. Raman spectroscopy of the electrode showed the characteristic spectra of Ti₂AlN. The HfO_x devices with MAX phase BE show a one-order-of-magnitude improvement in the off-current (HRS ~ 450 kΩ), a large switching window (~200), good endurance (>100) and retention (~10⁴s) of states. Furthermore, analog pulsing measurements are conducted to analyze the synaptic behavior of the devices. A COMSOL Multiphysics^® model is used to develop a fundamental understanding of the observed results.
Overall, this study demonstrates that through optimization of the materials in HfO_x memory, the electrical responses can be tailored for neuromorphic application.

1. Yang, J. J., et al. Nature nanotechnology, 8(1), 13-24.
2. K. Moon, et al., Faraday discussions 213, 421-451 (2019).
3. Athena, F. F., et al. Journal of Materials Chemistry C, 10(15), 5896-5904.
4. Athena, F. F., et al. Journal of Applied Physics, 131(20), 204901.
5. Zhang, T., et al. J. Ceram. Process. Res, 13, S149-S153.

Acknowledgements: This work was supported by the AFOSR MURI under Award No. FA9550-18-1-0024 to Eric M. Vogel. In part at the Georgia Tech Institute for Electronics and Nanotechnology, a member of the NNCI, which is supported by the NSF (ECCS-2025462). This material is based upon the work supported by the 2022-23 IBM PhD Fellowship Award to Fabia F. Athena.

Co-Author(s): Moses Nnaji (Georgia Institute of Technology), Matthew P. West (Georgia Institute of Technology), Samuel Graham (University of Maryland), Eric Vogel (Georgia Institute of Technology)

Fabia Farlin Athena

Appearances