Resetting the Drift of Oxygen Vacancies in Ultrathin HZO Ferroelectric Memories by Electrical Pulse Engineering

Abstract

Ferroelectric HfO₂-based films incorporated in nonvolatile memory devices offer a low-energy, high-speed alternative to conventional memory systems. Oxygen vacancies have been rigorously cited in literature to be pivotal in stabilizing the polar noncentrosymmetric phase responsible for ferroelectricity in HfO₂-based films. Thus, the ability to regulate and control oxygen vacancy migration in operando in such materials would potentially offer step changing new functionalities, tunable electrical properties, and enhanced device lifespan. Herein, a novel in- operando approach to control both wake-up and fatigue device dynamics is reported. Via clever design of short ad hoc square electrical pulses, both wake-up can be sped up and both fatigue and leakage inside the film can be reduced, key factors for enhancing the performance of memory devices. Using plasmon-enhanced photoluminescence and dark-field spectroscopy (sensitive to <1% vacancy variation), evidence that the electrical pulses give rise to oxygen vacancy redistribution is provided and it is shown that pulse engineering effectively delays wake-up and reduces fatigue characteristics of the HfO₂-based films. Comprehensive analysis also includes impedance spectroscopy measurements, which exclude any influence of polarization reversal or domain wall movement in interpretation of results.

Keywords: HZO ultrathin ferroelectric random‐access memories; Raman and photoluminescence; dark‐field spectroscopies; domain wall motions; fatigues; oxygen vacancy migrations; wake‐up.

Figure 1

Pulse‐engineered PUND‐reset waveforms. a–i) PUND‐reset pulses. Every square reset pulse is fixed in the opposite direction of the preceding triangular PUND pulse. The amplitude of reset pulse is in the range 0.1E _C–0.3E _C. The duration of the pulse varies from 0.25 to 0.75 ms. All waveforms are at 1 kHz (1 ms duration).

Figure 2

Wake‐up effects with PUND‐reset waveform. a) Polarization versus switching cycles for PUND (black) and two representative PUND‐reset (blue: 0.1 V/0.25 ms, red: 0.3 V/0.75 ms) pulses. Symbols are for data points and dashed lines for the fit. All 9 PUND‐reset profiles can be found in Supporting Information (b). The characteristic parameter for PUND‐reset normalized to that of the PUND is defined as τ _normalized for comparing delay in time to wake‐up. b) Heatmap showing the τ _normalized time constants for all 9 PUND‐reset.

Figure 3

Reduced fatigue with PUND‐reset waveforms. a) Device with common wake‐up, that is, both cycled with PUND for 2000 cycles until wake‐up. Post 2000 cycles device cycled with PUND (black) and 0.1 V–0.25 ms PUND‐reset (red). The solid lines show linear fit (2000–10 000 cycles, after wake‐up (shaded area)) to extract slope of fatigue for each device. b) Fatigue rate (slope of polarization versus cycles during fatigue) for devices cycled with: PUND only (0.8 μC cm⁻² 1k cycles⁻¹), PUND (common wake‐up), PUND‐reset for the rest of cycles (0.2 μC cm⁻² 1k cycles⁻¹), and PUND‐reset only (0.3 μC cm⁻² 1k cycles⁻¹).

Figure 4

Leakage and polarization loss with PUND‐reset waveforms. a) Leakage current density curves for devices compared between pristine state and after cycling with PUND and PUND‐reset waveforms. b) P–V characteristics for PUND and PUND‐reset (0.3 V 0.5 ms) pulses. Inset: polarization loss as gap at 0 V. c) Heatmap shows a figure of merit ΔP _normalized defined as ΔP of PUND reset, normalized to conventional PUND. A reduction in leakage (ΔP _normalized < 1) is present for all PUND‐reset pulsed devices, with 0.3 V–0.5 ms showing least polarization loss.

Figure 5

In situ spectroscopy with PUND‐reset waveforms. a,c) Electrical and b,d) PL traces of devices cycled in NPoM geometry. The insets in a, c show the PUND waveform used to cycle the devices; cycles to wake‐up are shaded in orange. The PL compares conventional PUND (a,b) and PUND reset (c,d) with a delayed response in wake‐up visible both in electrical (c) and optical trace (d). e,f) DF traces of devices cycled in NPoM geometry. The top insets show the electrical trace and bottom insets show the PUND waveform used to cycle. The DF peaks near 700 nm for conventional PUND (e) show left shift and lower intensity after 200 cycles indicative of vacancy defect generation compared to no change in (f).

Figure 6

Impedance spectroscopy of 5 nm HZO FE capacitor stack. The effective dielectric permittivity of the HZO capacitor stack extracted at 1 and 2 kHz versus amplitude E ₀ of the AC electric field shows subthreshold (left of the black line) and Rayleigh region (right of the black line). The effective equivalent circuit to evaluate the measured impedance of the FE HZO capacitor stack is depicted in the bottom inset.