High-performance dynamic random access memory capacitor with an equivalent oxide thickness of 0.31 nm via stepwise cycling in Y-doped Hf_0.5Zr_0.5O₂ thin films

Abstract

This study investigates the field-induced ferroelectric (FFE) characteristics and dynamic random access memory (DRAM) performance of Y-doped Hf_0.5Zr_0.5O₂ (Y:Hf_0.5Zr_0.5O₂) thin films grown by atomic layer deposition (ALD). Compared with undoped Hf_0.5Zr_0.5O₂, the Y:Hf_0.5Zr_0.5O₂ film exhibited suppressed ferroelectric orthorhombic phase and stabilized FFE tetragonal phase, resulting in double hysteresis loop characteristics in the polarization–electric field curves. These changes were attributed to substitutional diffusion of Y ions introduced by a single ALD cycle of Y₂O₃ inserted in the middle of the film. However, the onset field of the FFE effect from the pristine film was too high for DRAM application. To address this issue, a stepwise cycling method was proposed, consisting of an initial short high-field cycling step (6 MV cm⁻¹, 10⁵ cycles, 1 second) followed by subsequent cycles at gradually decreased field amplitudes (5 MV cm⁻¹, 10⁵ cycles, 1 second → 4 MV cm⁻¹, 10⁷ cycles, 100 seconds). This approach effectively shifted the FFE switching peaks toward lower fields, enabling charge boosting at low voltage (±0.8 V) while minimizing increases in remanent polarization and leakage current density (J). Consequently, the stepwise cycled 5.5-nm-thick Y:Hf_0.5Zr_0.5O₂ film achieved a high dielectric permittivity (k) of ∼68 and a record-low equivalent oxide thickness (EOT) of ∼0.31 nm among dielectric thin films satisfying the DRAM J criterion (J < 10⁻⁷ A cm⁻² at 0.8 V). The substantial EOT reduction with stepwise cycling was enabled by the low-voltage charge-boosting effect, which enhanced the field-induced polarization response. These improvements were attributed to dopant-induced local structural inhomogeneity and effective redistribution of double positively charged oxygen vacancies. The EOT values were sustained with only slight degradation over 10⁹ cycles of 0.8 V operation and fully recovered after short high-field cycling.

---

New concepts

This work presents a transformative dielectric-scaling strategy for dynamic random access memory (DRAM) by integrating atomic-precision Y doping with a stepwise cycling approach. DRAM-compatible field-induced ferroelectric (FFE) Hf_1−xZr_xO₂ films face a critical bottleneck in achieving low-voltage charge boosting, as conventional composition tuning or prolonged high-field cycling to lower the onset field of the FFE effect inevitably increases remanent polarization, hysteresis loss, and leakage. Using minimal material addition, a single Y₂O₃ ALD cycle at the film's middle enables substitutional Y incorporation that stabilizes the FFE tetragonal phase while suppressing the polar orthorhombic phase, preserving crystallization quality and improving leakage. The combination of Y doping and the stepwise cycling strategy enables low voltage charge boosting, achieving symmetric k enhancement at ±0.8 V with low hysteresis energy loss and minimal leakage degradation, fundamentally overcoming the intrinsic trade-off. This material-cycling concept yields a record-low equivalent oxide thickness (∼0.31 nm), maintains performance within the 10–20 ns DRAM time window, and establishes a new operation scheme for repeatable and reliable FFE performance. It paves the way for advanced materials design in semiconductor applications, highlighting the pivotal role of innovative doping-cycling strategies in overcoming the scaling and reliability limits of existing high-k dielectrics.

Introduction

Dynamic random access memory (DRAM) has served as the main memory in computers, but the increasing demand for higher bit density and energy-efficient operation requires continuous device scaling.^1,2 A key metric for DRAM capacitors is the equivalent oxide thickness where POT is the physical oxide thickness of the dielectric film and k is the dielectric permittivity. To decrease the EOT while meeting the leakage current density (J) requirement (J < 10⁻⁷ A cm⁻² at 0.8 V), numerous high-k materials have been explored as next-generation DRAM dielectrics.^1,2

The ZrO₂/Al₂O₃/ZrO₂ stack has been widely employed in current DRAM capacitors.¹ However, its relatively low k value (∼30–40) prohibits the EOT scaling <∼0.5 nm. An EOT < 0.5 nm is necessary for sub-15 nm DRAM design rules.¹ Therefore, the Hf_1−xZr_xO₂ (HZO) solid solution system has emerged as a promising dielectric material because it may offer a higher k value than the ZrO₂/Al₂O₃/ZrO₂. It also has compatibility with the standard TiN electrodes, mature atomic layer deposition (ALD) processes, larger band gaps (∼5.5 eV) with low J, and relatively low crystallization temperatures (350–500 °C).^3–13 Depending on Zr content, HZO thin films display dielectric (DE) (x ∼ 0), ferroelectric (FE) (x ∼ 0.5), and field-induced ferroelectric (FFE) (x > ∼0.7) characteristics, corresponding to monoclinic (m-phase, space group: P2₁/c), polar orthorhombic (PO-phase, space group: Pca2₁), and tetragonal (t-phase, space group: P4₂/nmc) phases, respectively.^{3,5,8,13–16} FE characteristic is unsuitable for DRAM due to residual remnant polarization (P_r) after voltage removal.¹¹ In contrast, FFE behavior enables efficient charge–discharge operations without P_r, making it more desirable than the FE phase for DRAM applications.¹¹ Although the non-polar t-phase exhibits a relatively low k (∼30–40),^11,17,18 resulting in limited DE displacement charges, applying a sufficiently high electric field (E_t→PO) induces a reversible transition to the PO-phase, generating additional field-induced polarization charges.¹¹ When the field is decreased below a critical value (E_PO→t), the PO-phase reverts to the t-phase, discharging the field-induced polarization charges without residual P_r.¹¹

The k value of DRAM capacitor layers is typically extracted from small-signal (alternating current (AC) voltage amplitude of ∼50 mV) capacitance–voltage (C–V) measurements under quasi-static conditions, which reflect the capacitive response at a fixed direct current (DC) bias voltage, typically around 0 V.¹¹ However, actual DRAM operation involves repeated charging and discharging through the application and removal of voltage pulses with an amplitude of ∼1 V. Therefore, the conventional C–V measurements do not accurately capture the charge storage dynamics when reversible FFE switching dynamics are involved, which occur at voltages much higher than the AC signal amplitude.¹¹ To address this issue, Hyun et al. proposed non-switching pulse measurements as a more accurate method for evaluating the k of FFE HZO films by accounting for the reversible field-induced polarization during pulse application and removal.¹¹

In their study, they demonstrated that a pristine 6.5-nm-thick Hf_0.3Zr_0.7O₂ film exhibited a high k of ∼80 at an electric field of ∼4.0 MV cm⁻¹, attributed to the FFE-induced polarization above the E_t→PO threshold (∼2.5 MV cm⁻¹). However, at the DRAM operating voltage of 0.8 V for the 6.5-nm-thick film, the applied field (∼1.2 MV cm⁻¹) was insufficient to induce FFE switching, resulting in a significantly lower k of only ∼30.¹¹ To lower the E_t→PO and enable effective FFE polarization contributions at lower voltages, two strategies were suggested. The first involved decreasing the Zr content to 0.5, which effectively decreased E_t→PO and enabled FFE polarization contributions at 0.8 V, resulting in an EOT of 0.47 nm in a 6.7-nm-thick Hf_0.5Zr_0.5O₂ film. However, this approach increased P_r due to the coexistence of the t- and PO-phases.¹¹ Alternatively, field cycling was applied to the 6.5-nm-thick Hf_0.3Zr_0.7O₂ film to decrease the free energy difference between the t- and PO-phases, thereby lowering E_t→PO. While this method enabled FFE switching at lower voltages, it also induced significant P_r increase,¹¹ attributed to the irreversible t- to the PO-phase transition with the wake-up effect.¹⁹ Furthermore, extended cycling resulted in significant degradation of J characteristics.^11,19 In both approaches, the decreased E_t→PO led to significant energy loss in the polarization–electric field (P–E) hysteresis loops,¹¹ indicating the intrinsic limitations of undoped HZO films in DRAM applications and highlighting the need for further optimization of material and cycling conditions.

Shin et al. reported that Y-doping, achieved by inserting a Y₂O₃ layer into Hf_0.5Zr_0.5O₂ thin films, is promising for DRAM applications, as it stabilizes the t-phase and induces FFE characteristics.²⁰ Unlike other dopants that degrade the crystallization behavior of HfO₂- or ZrO₂-based thin films,^18,21 Y₂O₃ insertion does not hinder crystallization or grain growth.^18,20 In addition, Y₂O₃ insertion into ZrO₂ thin films improved the J characteristics, attributed to the acceptor p-type doping effect of trivalent Y ions,^18,22,23 suggesting similar improvements may be expected in Hf_0.5Zr_0.5O₂ thin films. Although the FE behavior and field cycling characteristics of doped HfO₂- and HZO-based films have been extensively studied,^19,21,24,25 the FFE characteristics remain relatively unexplored, particularly in the context of DRAM applications.

Therefore, this study investigated the FFE characteristics and DRAM performance of Y-doped Hf_0.5Zr_0.5O₂ (Y:Hf_0.5Zr_0.5O₂) thin films. The FFE HZO system faces a critical bottleneck in achieving efficient low-voltage charge boosting, as conventional approaches such as composition tuning or extended high-field cycling to lower the FFE onset field inevitably lead to increased P_r, J, and hysteresis loss.¹¹ A combined approach of Y-doping and stepwise cycling was introduced to overcome these limitations in this work. Substitutional diffusion of Y ions effectively stabilized the FFE t-phase while suppressing the FE PO-phase formation. The Y:Hf_0.5Zr_0.5O₂ film showed a lower J than the undoped films. The stepwise cycling method, which consisted of an initial short high-field cycling step (6 MV cm⁻¹, 10⁵ cycles, 1 second) followed by subsequent cycles at gradually reduced field amplitudes (5 MV cm⁻¹, 10⁵ cycles, 1 second → 4 MV cm⁻¹, 10⁷ cycles, 100 seconds), was adopted to shift the FFE switching peaks toward lower fields effectively. This method enabled the low-voltage charge boosting with minimized hysteresis, suppressed P_r, and mitigated J degradation. The optimized 5.5-nm-thick Y:Hf_0.5Zr_0.5O₂ film achieved a record-low EOT of ∼0.31 nm among DRAM-compatible dielectrics satisfying the DRAM J criterion. It also maintained stable dielectric properties within the DRAM read/write time window of 10–20 ns, estimated from experimentally obtained pulse durations with the measured DRAM resistance–capacitance (RC) ratio (∼1351). Furthermore, stable operation was sustained over 10⁹ cycles at 0.8 V. Nevertheless, a slight degradation in performance was observed with extended cycling. However, the degradation could be fully recovered by short high-field cycling, demonstrating a reliable and repeatable recovery strategy for long-term DRAM operation.

Results and discussion

Comparative characterization of Hf_0.5Zr_0.5O₂, Hf_0.3Zr_0.7O₂, and Y:Hf_0.5Zr_0.5O₂ films (5.3–5.5-nm-thick) at the pristine state

This section describes the structural, chemical, and electrical characteristics, as well as the DRAM performance, of the 5.5-nm-thick Y:Hf_0.5Zr_0.5O₂ film in comparison with the undoped 5.3-nm-thick Hf_0.5Zr_0.5O₂ and 5.4-nm-thick Hf_0.3Zr_0.7O₂ films at the pristine state. For the 5.5-nm-thick Y:Hf_0.5Zr_0.5O₂, a single ALD cycle of Y₂O₃ was inserted at the center location along the thickness direction of the Hf_0.5Zr_0.5O₂ layer with a Y-precursor feeding time of 3.5 seconds, resulting in a Y-concentration (Y/[Y + Hf + Zr]) of 2.60 atomic%, estimated by energy-dispersive X-ray fluorescence spectroscopy (EDXRF).

Grazing-angle incidence X-ray diffraction (GIXRD) characterization was conducted to investigate the effect of Y₂O₃ insertion on the phase evolution of Hf_0.5Zr_0.5O₂ thin films. Due to the larger ionic radius of Y³⁺ (90 pm) compared to Hf⁴⁺ and Zr⁴⁺ (72 pm),^18,20 Y substitution into the crystal lattice can be examined by monitoring the 2θ shifts of GIXRD peaks near ∼30.6° and ∼35.6°, corresponding to the o(111)/t(011) and o(002) planes, respectively.^20,26–28 Dopant diffusion also influences the polymorphism and phase transitions in HZO thin films, observable through variations in the aspect ratio (AR), unit cell volume, and m-phase ratio derived from the deconvoluted GIXRD spectra.^20,26–28 The FFE t-phase generally exhibits a lower AR and smaller unit cell volume than the FE PO-phase in the HfO₂-based fluorite material system, allowing the relative phase changes to be inferred from these parameters.^20,26–28 In a previous study, systematic variations in these parameters were observed by changing the number of Y₂O₃ insertion layer (IL) ALD cycles inserted at the center location of the ∼10-nm-thick Hf_0.5Zr_0.5O₂ films after the post-metallization annealing (PMA) at 525 °C.²⁰ The results indicated that the substitutional diffusion of larger Y³⁺ ions into the Hf⁴⁺ and Zr⁴⁺ sites induced lattice expansion, transitioning the m- and PO-phases to the FFE t-phase.²⁰ Fig. S1a–f present the reproduced data from the previous study along with detailed analysis.

Similar GIXRD analysis was performed for the 5.3-nm-thick Hf_0.5Zr_0.5O₂, 5.4-nm-thick Hf_0.3Zr_0.7O₂, and 5.5-nm-thick Y:Hf_0.5Zr_0.5O₂ films. Fig. 1a and b show the background-subtracted GIXRD patterns and Gaussian deconvoluted curves of the three films after PMA at 450 °C for 30 seconds. Unlike the ∼10-nm-thick films in Fig. S1b, no m-phase peaks were observed, attributed to the lower PMA temperature and suppressed grain growth in the thinner films.²⁰ The 2θ position of the o(002) peak (from the PO phase) was lower for the Y:Hf_0.5Zr_0.5O₂ film (35.34°) than for the Hf_0.5Zr_0.5O₂ film (35.42°). According to Bragg's law, a decrease in 2θ corresponds to an increase in interplanar spacing, indicating lattice expansion induced by the substitutional diffusion of larger Y³⁺ ions into the smaller Hf⁴⁺ and Zr⁴⁺ sites.²⁰ In contrast, the o(111)/t(011) peak (from PO and t-phases) exhibited a slightly higher 2θ value in the Y:Hf_0.5Zr_0.5O₂ (30.58°) than in the Hf_0.5Zr_0.5O₂ (30.56°), despite the expected decrease in 2θ values due to lattice expansion.²⁰ This trend coincides with previous reports that the o(111)/t(011) peak position is influenced not only by dopant-induced lattice expansion (or contraction), but also by the phase fraction changes, where an increased t-phase (and decreased PO-phase) increases the 2θ values.^4,6,20,26 Therefore, the increase in 2θ of the o(111)/t(011) peak is attributed to the increased t-phase fraction over the opposing effect of lattice expansion.

Fig. 1 Background-subtracted GIXRD (incident angle: 0.5°) patterns (grey) of 5.3-nm-thick Hf_0.5Zr_0.5O₂, 5.4-nm-thick Hf_0.3Zr_0.7O₂, and 5.5-nm-thick Y:Hf_0.5Zr_0.5O₂ films for (a) o(002) and TiN peaks and (b) o(111)/t(011) peaks, respectively. (c) The aspect ratio and unit cell volume for the three films. Error bars of the aspect ratio and unit cell volume are estimated from the error of the peak positions of the o(002) and o(111)/t(011) peaks from the standard deviation of the Gaussian fitting. (d) The Cs-TEM image of the 5.5-nm-thick Y:Hf_0.5Zr_0.5O₂ film, with the iFFT image of the region indicated by the red square.

To validate this argument, the AR and unit cell volume were calculated from the interplanar spacings d₁₁₁ and d₀₀₂ of the o(111)/t(011) and o(002) peaks, respectively, assuming negligible differences in the two shorter lattice parameters,^20,26,27 and Fig. 1c shows the results. Calculation methods and assumptions are reported elsewhere.^26,27 Both the Hf_0.3Zr_0.7O₂ and Y:Hf_0.5Zr_0.5O₂ films exhibited decreased AR and unit cell volume values compared to Hf_0.5Zr_0.5O₂, indicating FFE t-phase stabilization, attributed to the increased Zr content^8,10 and Y diffusion,²⁰ respectively. Notably, the AR decreased similarly by 0.65% and 0.79% for the Hf_0.3Zr_0.7O₂ and Y:Hf_0.5Zr_0.5O₂, respectively, compared to Hf_0.5Zr_0.5O₂, whereas the unit cell volume decreased more substantially in the Hf_0.3Zr_0.7O₂ (0.53%) than in the Y:Hf_0.5Zr_0.5O₂ (0.13%). This trend reflects the combined effects of t-phase stabilization, which decreases the unit cell volume, and lattice expansion from Y-diffusion, which offsets it, resulting in a smaller net decrease. The GIXRD changes observed in the 5.5-nm-thick Y:Hf_0.5Zr_0.5O₂ film were consistent with those in ∼10-nm-thick Hf_0.5Zr_0.5O₂ films with one or two Y₂O₃ ALD cycles inserted at the film's center location, as shown in Fig. S1a–f, confirming both t-phase stabilization and lattice expansion.

It was reported that the Y-diffusion does not disrupt the continuous growth of Hf_0.5Zr_0.5O₂ or HfO₂.^20,79Fig. 1d shows the spherical-aberration-corrected transmission electron microscopy (Cs-TEM) image of the 5.5-nm-thick Y:Hf_0.5Zr_0.5O₂ film, with the fast Fourier transform (FFT) pattern and magnified image of the region indicated by the red square, confirming a continuous, single-layer crystalline lattice.

The t-phase stabilization mechanism through Y-diffusion was further investigated. Shin et al. reported that the substitutional diffusion of Y³⁺ ions into the Hf⁴⁺ and Zr⁴⁺ sites created oxygen vacancies (V_O) to maintain charge neutrality, thereby enhancing the t-phase stabilization.²⁰ Fig. S2a–c show the reproduced X-ray photoelectron spectroscopy (XPS) data from the previous study.

Previous studies also suggested that dopants or defects may influence HfO₂ polymorphism by changing grain size.^9,28 Fig. S3a–d show the scanning electron microscopy (SEM) images and grain diameter distributions of the 5.3-nm-thick Hf_0.5Zr_0.5O₂, 5.4-nm-thick Hf_0.3Zr_0.7O₂, and 5.5-nm-thick Y:Hf_0.5Zr_0.5O₂ films after post-deposition annealing (PDA). However, all three films exhibited comparable grain diameter (20.3–21.3 nm), indicating that grain size variation was not the primary factor contributing to t-phase stabilization in the Y:Hf_0.5Zr_0.5O₂ film. A more detailed discussion on the effects of these structural and chemical factors on DRAM characteristics is provided in Section 2.2.

Next, the electrical characteristics relevant to the FFE DRAM application were examined. The following sections present transient current–electric field (I–E) and P–E curves measured using bipolar triangular pulses. Fig. 2a shows the schematic I–E and P–E curves of an ideal FFE film, where the FFE forward and reverse switching peaks, corresponding to the E_t→PO and E_PO→t, respectively, occur at the same electric field. In this case, the P–E loop, obtained by integrating the I–E curve, is non-hysteretic. However, in practical FFE films, the E_t→PO typically appears at a higher field than the E_PO→t peak, as shown in Fig. 2b, due to the local energy barrier during the switching process from the PO- to t-phase.¹¹ This field difference results in a hysteresis in the P–E curve. Furthermore, experimental FFE films exhibit broadened E_t→PO and E_PO→t distributions, as shown in Fig. 2c, resulting in finite FFE slopes in the P–E curves.²⁹ The distribution of E_t→PO and E_PO→t may arise from structural inhomogeneities such as non-uniform grain/crystallite size, defect distribution across the film, or the polycrystalline nature of the actual HZO films.^21,29–31

Fig. 2 Schematics of representative I–E and P–E behavior. (a)–(c) FFE characteristic films with (a) single E_t→PO and E_PO→t transitions without hysteresis, (b) single E_t→PO and E_PO→t transitions with hysteresis, and (c) distributed E_t→PO and E_PO→t transitions with hysteresis. (d) and (e) Pinched or broken hysteresis loops with changing P_r. (f) FE characteristic film. (g) Ideal I–E and P–E curves for DRAM application.

In the FFE cases shown in Fig. 2a–c, the switching peaks are confined to the first and third (E_t→PO) or second and fourth (E_PO→t) quadrants, resulting in P–E curves without P_r. However, when the E_PO→t peaks shift into the first and third quadrants, as shown in Fig. 2d and e, the P–E curve becomes pinched (or broken), and further shifts lead to increased P_r, which may indicate a higher fraction of the FE PO-phase.¹⁶ Such pinched P–E curves are typically observed in Hf_0.4Zr_0.6O₂ films¹⁶ or thin (<7 nm) Hf_0.5Zr_0.5O₂ films,^6,11 where t- and PO-phases coexist. Although referring to the switching peaks as E_t→PO and E_PO→t may not be strictly accurate in these cases, where the films no longer exhibit pure FFE characteristics, these terms are used for simplicity in this study. When both switching peaks merge in the first and third quadrants, as shown in Fig. 2f, a typical FE characteristic P–E curve with a large P_r is obtained.³²

Three requirements should be met for the practical application of FFE films as DRAM dielectric layers, as illustrated in Fig. 2g. First, since DRAM capacitor voltages are as low as ±0.8 V, the E_t→PO distribution should lie within this low-field (low-voltage) range to enable effective k enhancement through FFE charge boosting.¹¹ Second, the hysteresis loop should be minimized to decrease adverse heating and energy loss,^11,33 which requires minimal field difference between the E_t→PO and E_PO→t distributions. Third, the P_r should be minimal,¹¹ requiring that the E_PO→t should remain confined to the second and fourth quadrants.

Next, the electrical characteristics of the 5.5-nm-thick Y:Hf_0.5Zr_0.5O₂ film were compared with those of the 5.3-nm-thick Hf_0.5Zr_0.5O₂ and 5.4-nm-thick Hf_0.3Zr_0.7O₂ films. Fig. 3a and b show the I–E and P–E curves of the three pristine films, respectively. The Hf_0.5Zr_0.5O₂ film exhibited a pinched hysteresis loop and a non-negligible 2P_r (23.0 µC cm⁻²), due to the coexistence of t- and PO-phases at low thickness (<7 nm), as previously reported.¹¹ In contrast, the Hf_0.3Zr_0.7O₂ and Y:Hf_0.5Zr_0.5O₂ films displayed FFE I–E characteristics and double hysteresis loops in the P–E curves, consistent with the t-phase stabilization confirmed by the GIXRD analysis in Fig. 1a–c.

Fig. 3 The (a) I–E curves, (b) P–E curves, (c) non-switching P–E curves, (d) Q_c (open) and Q_d (closed) characteristics, (e) k–E characteristics, and (f) EOT–E characteristics, respectively, of the 5.3-nm-thick Hf_0.5Zr_0.5O₂ (pink), 5.4-nm-thick Hf_0.3Zr_0.7O₂ (light green), and 5.5-nm-thick Y:Hf_0.5Zr_0.5O₂ (sky blue) films at the pristine state.

The Y:Hf_0.5Zr_0.5O₂ film exhibited broader E_t→PO and E_PO→t distributions than Hf_0.3Zr_0.7O₂. Fig. S4b shows the Y doping profile of the 5.5-nm-thick Y:Hf_0.5Zr_0.5O₂ film after PDA, obtained by time-of-flight secondary ion mass spectrometry (ToF-SIMS). The Y₂O₃ IL at the center location of the Hf_0.5Zr_0.5O₂ film resulted in YO₂⁻ profile peaking near the center and decreasing toward both interfaces, indicating spatial inhomogeneity. Such inhomogeneity may contribute to the broadening of the switching distributions.^26,29 Notably, the field difference at which E_t→PO and E_PO→t reached maxima was smaller in Y:Hf_0.5Zr_0.5O₂ (sky blue dotted lines) than in Hf_0.3Zr_0.7O₂ (green dotted lines) in the I–E curves, indicating a lower local energy barrier for the PO- to t-phase transition,¹¹ resulting in slimmer hysteresis in the P–E curves. A detailed discussion on the decreased hysteresis is provided later.

Fig. 3c shows the non-switching P–E curves, with the films pre-poled using a 6 MV cm⁻¹ triangular pulse for both biases, followed by the same measurement pulse, as illustrated in the inset. The pre-poling aligns the remnant FE switching polarization, allowing for the measurement of only the field-induced polarization component relevant to FFE DRAM operation.¹¹ Due to the decreased Zr composition, the Hf_0.5Zr_0.5O₂ exhibited an abrupt polarization increase at a lower field than the Hf_0.3Zr_0.7O₂ for both biases due to the decreased E_t→PO, consistent with prior reports.¹¹ However, despite showing the largest saturated polarization (P_s) in Fig. 3b, the Hf_0.5Zr_0.5O₂ film displayed the smallest P_s in the non-switching P–E curve because the previously aligned remnant FE component could not contribute to the charging process.¹¹ Both the Hf_0.3Zr_0.7O₂ and Y:Hf_0.5Zr_0.5O₂ films displayed abrupt polarization increase at ∼3 MV cm⁻¹, with the latter exhibiting a smaller hysteresis loop. Hyun et al. reported that the P_r in the non-switching P–E curves indicates the differences in charging and discharging behavior.¹¹ A larger P_r was observed for the Hf_0.5Zr_0.5O₂ film, while the Hf_0.3Zr_0.7O₂ and Y:Hf_0.5Zr_0.5O₂ films exhibited minimal P_r, consistent with Fig. 3b.

Subsequently, non-switching pulse measurements were conducted to evaluate the charging and discharging behavior, which mimics DRAM operation, as previously reported.¹¹ The pulse application method, shown in the inset of Fig. 3d, involved a 6 MV cm⁻¹ rectangular pulse with a 20 µs length for pre-poling, followed by measurement pulses of the same length with incrementally increasing amplitude. The charging (Q_c) and discharging (Q_d) charges were extracted from the resulting current response. Fig. 3e and f show the k–E and EOT–E characteristics, respectively, calculated from Q_d because the discharging determines the reading margin in DRAM operation.¹¹ The previous report provided the detailed measurement and calculation processes, along with further discussion of the non-switching pulse measurements.¹¹ Section 2.4 provides detailed discussions regarding the pulse length relevant to DRAM read/write time.

Fig. 3d shows the Q_c (open) and Q_d (closed) values for the three films. The Hf_0.3Zr_0.7O₂ and Y:Hf_0.5Zr_0.5O₂ films exhibited nearly identical Q_c and Q_d values across the whole field range, whereas the Hf_0.5Zr_0.5O₂ film showed a smaller Q_d than Q_c above ∼2.5 MV cm⁻¹, consistent with the P_r behavior observed in Fig. 3c.¹¹ However, this difference was negligible within the DRAM operation voltage (∼1.5 MV cm⁻¹ ≈ 0.8 V/5.3 nm).

Fig. 3e shows the k–E characteristics of the three films. Due to their high onset fields for FFE switching, both Hf_0.3Zr_0.7O₂ and Y:Hf_0.5Zr_0.5O₂ exhibited an increased k only above ∼3.0 MV cm⁻¹, reaching values of ∼72 and ∼63, respectively, at ∼5 MV cm⁻¹. The Hf_0.5Zr_0.5O₂ film, with a lower FFE onset field, showed an earlier increase in k and reached ∼75 at ∼3 MV cm⁻¹. However, at ±0.8 V (±1.45–1.50 MV cm⁻¹, indicated by grey dashed lines), all three films exhibited low k values (Hf_0.5Zr_0.5O₂ ∼35; Hf_0.3Zr_0.7O₂ ∼30; Y:Hf_0.5Zr_0.5O₂ ∼33) due to the limited FFE charging contributions at such low fields. Fig. 3f shows the EOT–E characteristics. While all films exhibited EOT values below 0.3 nm in the high-field region, significantly large EOT values were observed at 0.8 V (Hf_0.5Zr_0.5O₂ ∼0.59 nm; Hf_0.3Zr_0.7O₂ ∼0.68 nm; Y:Hf_0.5Zr_0.5O₂ ∼0.65 nm), again indicating the limited FFE charging contributions.

Fig. S5 shows the J–voltage (J–V) curves of the 5.3-nm-thick Hf_0.5Zr_0.5O₂, 5.4-nm-thick Hf_0.3Zr_0.7O₂, and 5.5-nm-thick Y:Hf_0.5Zr_0.5O₂ films at the pristine state. All three films satisfied J < 10⁻⁷ A cm⁻² at 0.8 V. Notably, the Y:Hf_0.5Zr_0.5O₂ film exhibited a lower J at 0.8 V compared to the Hf_0.5Zr_0.5O₂ and Hf_0.3Zr_0.7O₂ films, attributed to the acceptor doping effect of the trivalent Y cations into the tetravalent Hf and Zr cation sites, as extensively reported previously.^18,22,23 However, despite the favorable J characteristics in the pristine state, the low k and high EOT observed in Fig. 3e and f were unsuitable for DRAM applications, necessitating additional optimization strategies.

Stepwise cycling

To enhance FFE charging at DRAM operating voltages and achieve higher k and lower EOT, shifting E_t→PO and E_PO→t to lower values is required. Shin et al. reported that modulation of E_t→PO and E_PO→t critically depends on the cycling field amplitude, with the distributions shifting to lower fields only when the cycling field (E_cycle) exceeds the E_t→PO value.²⁹ For instance, in the pristine Hf_0.3Zr_0.7O₂ and Y:Hf_0.5Zr_0.5O₂ films, E_t→PO is distributed between ∼2.5–6 MV cm⁻¹ (Fig. 3a), requiring E_cycle = 6 MV cm⁻¹ to shift the entire distribution to a lower-field region. In contrast, applying E_cycle = 4 MV cm⁻¹ would shift only the portion between ∼2.5–4 MV cm⁻¹.

However, extensive high-field cycling induces a wake-up effect due to the irreversible transition to the PO-phase, resulting in increased P_r.¹⁹ Fig. S6a–c show the I–E curves, and Fig. S6d–f show the P–E curves of the 5.3-nm-thick Hf_0.5Zr_0.5O₂, 5.4-nm-thick Hf_0.3Zr_0.7O₂, and 5.5-nm-thick Y:Hf_0.5Zr_0.5O₂ after field cycling at E_cycle = 6.0 MV cm⁻¹, with application of 100 kHz bipolar triangular pulses. In the Hf_0.5Zr_0.5O₂ film, only 10⁵ cycles (1 second of total cycling time) resulted in an FE-like P–E curve with significant 2P_r (46.7 µC cm⁻²), indicating loss of FFE characteristics. Hence, this film was excluded from further cycling evaluation. In the Hf_0.3Zr_0.7O₂ and Y:Hf_0.5Zr_0.5O₂ films, E_t→PO and E_PO→t gradually shifted to lower fields after 10⁷ cycles (100 seconds of total cycling time), but were also accompanied by notably increased P_r. High field-cycling also causes significant J degradation due to accumulated cycling stress.^11,19 Fig. S7a and b show the J–V curve changes of the Hf_0.3Zr_0.7O₂ and Y:Hf_0.5Zr_0.5O₂ films, where J significantly increased after field cycling at E_cycle = 6.0 MV cm⁻¹ for 10⁷ cycles, far exceeding the DRAM requirement (J < 10⁻⁷ A cm⁻² at 0.8 V). These results highlight the need for another cycling method that enables low-field FFE switching modulation without inducing excessive P_r or J degradation.

Hence, this work proposes a stepwise cycling method, where a short high-field cycling step (6 MV cm⁻¹, 10⁵ cycles, 1 second) is followed by subsequent cycles at gradually decreasing field amplitudes (5 MV cm⁻¹, 10⁵ cycles, 1 second → 4 MV cm⁻¹, 10⁷ cycles, 100 seconds). This approach aims to shift the FFE switching peaks toward lower fields, thereby improving low-voltage charge boosting while reducing the adverse effects of excessive high-field cycling on the film. This minimizes undesired increases in P_r and J degradation.

Fig. 4a shows the I–E curve changes of the 5.5-nm-thick Y:Hf_0.5Zr_0.5O₂ film after three sequential steps: (step 1) E_cycle = 6.0 MV cm⁻¹, 10⁵ cycles (1 second) → (step 2) E_cycle = 5.0 MV cm⁻¹, 10⁵ cycles (1 second) → (step 3) E_cycle = 4.0 MV cm⁻¹, 10⁷ cycles (100 seconds) (all 100 kHz bipolar triangular pulses). The I–E curves were measured using 6 MV cm⁻¹, 1 kHz bipolar triangular pulses between each cycling step. A schematic diagram of the stepwise cycling pulse sequence is shown in Fig. S8.

Fig. 4 The (a) I–E curve and (b) P–E curve changes of the 5.5-nm-thick Y:Hf_0.5Zr_0.5O₂ and (c) I–E curve and (d) P–E curve changes of the 5.4-nm-thick Hf_0.3Zr_0.7O₂ with stepwise cycling after sequential cycling steps of (step 1) E_cycle = 6.0 MV cm⁻¹, 10⁵ cycles → (step 2) E_cycle = 5.0 MV cm⁻¹, 10⁵ cycles → (step 3) E_cycle = 4.0 MV cm⁻¹, 10⁷ cycles.

For operation at ±0.8 V, the E_t→PO should be distributed within this voltage range, highlighted in yellow. In step 1, the pristine film (dotted black curve) exhibited E_t→PO values broadly distributed between ∼2.5 and 6.0 MV cm⁻¹ for both biases. Applying E_cycle = 6.0 MV cm⁻¹ (orange dashed line) for 10⁵ cycles shifted these peaks toward lower fields, but not fully into the ±0.8 V range. Since the E_t→PO distribution was shifted below 5.0 MV cm⁻¹ in step 1, a lower E_cycle = 5.0 MV cm⁻¹ (sky blue dashed line) for 10⁵ cycles was used to shift them further in step 2. After step 2, E_t→PO distribution is located below 4.0 MV cm⁻¹, so applying E_cycle = 4.0 MV cm⁻¹ (green dashed line) for 10⁷ cycles in step 3 shifted the majority of the peaks within ±0.8 V range. Consequently, the final I–E curve exhibited E_t→PO and E_PO→t distributed in the low-field region (indicating enhanced FFE charge boosting at low fields), a decreased average field gap between them (indicating low hysteresis), and minimal intrusion of E_PO→t peaks into the first and third quadrants (indicating low P_r), resembling the desired FFE DRAM behavior illustrated in Fig. 2g. Fig. 4b shows the P–E curve changes of the 5.5-nm-thick Y:Hf_0.5Zr_0.5O₂ after the same sequential stepwise cycling steps. Compared to extended high-field cycling at E_cycle = 6.0 MV cm⁻¹ for 10⁷ cycles (Fig. S6f), the stepwise cycling effectively diminished the increase in P_r and hysteresis while shifting the E_t→PO and E_PO→t to lower fields, resulting in a P–E curve similar to the desired configuration in Fig. 2g.

Next, the identical stepwise cycling steps (E_cycle = 6.0 MV cm⁻¹, 10⁵ cycles → E_cycle = 5.0 MV cm⁻¹, 10⁵ cycles → E_cycle = 4.0 MV cm⁻¹, 10⁷ cycles) were applied to the 5.4-nm-thick Hf_0.3Zr_0.7O₂ film, as shown in the I–E curves in Fig. 4c. In step 1, the pristine film (dotted black curve) exhibited E_t→PO values broadly distributed between ∼3.0 and 6.0 MV cm⁻¹ under both biases. Applying E_cycle = 6.0 MV cm⁻¹ (red dashed line) for 10⁵ cycles shifted the peaks toward lower fields, but a substantial portion of the E_PO→t peaks intruded into the first and third quadrants. Although the E_t→PO distributions were shifted below 5.0 MV cm⁻¹, subsequent steps applying E_cycle = 5.0 MV cm⁻¹, 10⁵ cycles (blue dashed line) and E_cycle = 4.0 MV cm⁻¹, 10⁷ cycles (brown dashed line) resulted in minimal changes. Consequently, the final I–E curve showed insufficient E_t→PO shifts into the ±0.8 V range, a larger field gap between E_t→PO and E_PO→t, and more pronounced E_PO→t intrusion into the first and third quadrants compared to the Y:Hf_0.5Zr_0.5O₂ case. These results indicated limited FFE charge boosting at low fields, larger hysteresis, and higher P_r. Also, unlike the relatively symmetric I–E curve in Y:Hf_0.5Zr_0.5O₂ after the stepwise cycling, the Hf_0.3Zr_0.7O₂ showed asymmetry with the positive bias E_t→PO not shifting as effectively into the ±0.8 V range as those at the negative bias. Fig. 4d shows the corresponding P–E curve changes. Although the stepwise cycling mitigated the P_r increase and leakage degradation compared to the extended high-field cycling case in Fig. S6e, both P_r and hysteresis were larger than those of the Y:Hf_0.5Zr_0.5O₂ in Fig. 4b. Fig. S9a and b show the positive-up-negative-down (PUND) measurements of the 5.4-nm-thick Hf_0.3Zr_0.7O₂ and 5.5-nm-thick Y:Hf_0.5Zr_0.5O₂ films after stepwise cycling, using 0.8 V and 6.0 MV cm⁻¹ triangular pulses, respectively. In both cases, the Hf_0.3Zr_0.7O₂ film exhibited larger 2P_r, confirming a more significant transition of the FFE portion into the FE component.⁸

Next, the charging and discharging behaviors of the stepwise cycled films were evaluated, along with the k–E and EOT–E characteristics extracted from Q_d. Fig. 5a and b show the non-switching P–E curves of the 5.4-nm-thick Hf_0.3Zr_0.7O₂ and 5.5-nm-thick Y:Hf_0.5Zr_0.5O₂ films, respectively, in the pristine and stepwise cycled states. Both films initially exhibited high FFE onset fields of ∼2.5–3 MV cm⁻¹ under both biases, which were lowered after stepwise cycling. However, Hf_0.3Zr_0.7O₂ showed asymmetric lowering, with the positive-bias FFE onset field remaining relatively high (∼1.5–2 MV cm⁻¹), whereas Y:Hf_0.5Zr_0.5O₂ exhibited more symmetric and pronounced reductions. These differences in FFE onset fields directly affected the DRAM characteristics.

Fig. 5 Non-switching P–E curves of the (a) 5.4-nm-thick Hf_0.3Zr_0.7O₂ and (b) 5.5-nm-thick Y:Hf_0.5Zr_0.5O₂ films at the pristine state and after stepwise cycling. The Q_c (open) and Q_d (closed) characteristics of the (c) 5.4-nm-thick Hf_0.3Zr_0.7O₂ and (d) 5.5-nm-thick Y:Hf_0.5Zr_0.5O₂ films at the pristine state and after stepwise cycling. The k–E characteristics of the (e) 5.4-nm-thick Hf_0.3Zr_0.7O₂ and (f) 5.5-nm-thick Y:Hf_0.5Zr_0.5O₂ films at the pristine state and after stepwise cycling.

Fig. 5c and d show the Q_c (open) and Q_d (closed) characteristics for the 5.4-nm-thick Hf_0.3Zr_0.7O₂ and 5.5-nm-thick Y:Hf_0.5Zr_0.5O₂ films, respectively, before and after stepwise cycling. In both cases, Q_c and Q_d at low-field regions increased after stepwise cycling. However, the increase in Hf_0.3Zr_0.7O₂ was less pronounced, particularly under positive bias, due to insufficient lowering of the FFE onset field. A noticeable increase was observed only above ∼2 MV cm⁻¹. In contrast, Y:Hf_0.5Zr_0.5O₂ exhibited a more abrupt and symmetric increase from low fields, confirming more effective FFE onset field modulation.

Fig. 5e and f show the k–E curves of the 5.4-nm-thick Hf_0.3Zr_0.7O₂ and 5.5-nm-thick Y:Hf_0.5Zr_0.5O₂ films, respectively. Hf_0.3Zr_0.7O₂ showed increased k values at ±0.8 V after stepwise cycling compared to the pristine state. However, due to the limited FFE onset field lowering and decreased FFE contributions resulting from a more significant transition to FE components (Fig. S9a and b), the resulting values remained asymmetric and relatively low (+0.8 V ∼42; −0.8 V ∼59). In contrast, Y:Hf_0.5Zr_0.5O₂ exhibited a symmetric k–E curve with higher values after stepwise cycling (+0.8 V ∼68; −0.8 V: ∼66), attributed to more effective FFE onset field lowering and retained FFE portions.

The EOT and J characteristics were evaluated to assess the suitability of the stepwise cycling strategy for practical DRAM applications. Fig. 6a and b show the EOT–E curves of the 5.4-nm-thick Hf_0.3Zr_0.7O₂ and 5.5-nm-thick Y:Hf_0.5Zr_0.5O₂ films, respectively, at the pristine state and after stepwise cycling. In 5.4-nm-thick Hf_0.3Zr_0.7O₂, EOT values decreased from +0.8 V ∼0.68 nm; −0.8 V ∼0.72 nm to +0.8 V ∼0.51 nm; −0.8 V ∼0.35 nm after stepwise cycling, but the reduction was relatively limited and asymmetric. In contrast, the Y:Hf_0.5Zr_0.5O₂ exhibited a more substantial decrease from ∼0.65–0.66 nm to ∼0.31–0.33 nm, confirming the effective low-field FFE charge boosting suitable for symmetric DRAM applications.

Fig. 6 The EOT–E characteristics of the (a) 5.4-nm-thick Hf_0.3Zr_0.7O₂ and (b) 5.5-nm-thick Y:Hf_0.5Zr_0.5O₂ films at the pristine state and after stepwise cycling. The J–V curves of the (c) 5.4-nm-thick Hf_0.3Zr_0.7O₂ and (d) 5.5-nm-thick Y:Hf_0.5Zr_0.5O₂ films at the pristine state and after stepwise cycling.

Fig. 6c and d show the J–V curves of the same films, before and after stepwise cycling. In Fig. S7, high field cycling at 6.0 MV cm⁻¹ for 10⁷ cycles (100 seconds) caused severe J degradation exceeding three orders of magnitude at ±0.8 V in both films. In contrast, stepwise cycling significantly mitigated the degradation. The 5.4 nm Hf_0.3Zr_0.7O₂ film exhibited an increase in J from +0.8 V ∼5.8 × 10⁻⁸ A cm⁻²; −0.8 V ∼1.9 × 10⁻⁷ A cm⁻² to +0.8 V ∼1.3 × 10⁻⁶ A cm⁻²; −0.8 V ∼1.5 × 10⁻⁶ A cm⁻² after stepwise cycling, failing to meet the DRAM J criterion. In contrast, the J of 5.5 nm Y:Hf_0.5Zr_0.5O₂ film increased from +0.8 V ∼6.3 × 10⁻⁹ A cm⁻²; −0.8 V ∼1.8 × 10⁻⁸ A cm⁻² to +0.8 V ∼7.7 × 10⁻⁸ A cm⁻²; −0.8 V ∼9.3 × 10⁻⁸ A cm⁻², meeting the DRAM specification.

The reasons for these differences between the 5.4-nm-thick Hf_0.3Zr_0.7O₂ and 5.5-nm-thick Y:Hf_0.5Zr_0.5O₂ are examined. The Y:Hf_0.5Zr_0.5O₂ film exhibited a slimmer hysteresis loop than Hf_0.3Zr_0.7O₂ in the pristine state (Fig. 3b) and retained a slimmer hysteresis and smaller P_r after cycling (Fig. 4b and d). The shifts in E_t→PO and E_PO→t towards low fields were more pronounced in Y:Hf_0.5Zr_0.5O₂ (Fig. 4a and c), resulting in enhanced DRAM characteristics at ±0.8 V (Fig. 5e, f and 6a, b).

Systematic studies on hysteresis reduction in FFE HZO-based thin films remain limited, whereas antiferroelectric (AFE) perovskite-based thin films were more extensively investigated.³⁴ In AFE perovskites, the application of a sufficiently large electric field (E_AFE→FE) aligns the antiparallel dipoles to induce a transient FE state, which reverts to the AFE state upon field removal at E_FE→AFE.³⁴ Hysterisis occurs due to a larger E_AFE→FE than E_FE→AFE,³⁴ as in the FFE system. Previous studies have extensively reported that heterovalent cation doping induces relaxor antiferroelectric (RAFE) characteristics by introducing local structural inhomogeneity and disrupting long-range ordered domains, thereby forming smaller nanodomains with weakened inter-domain interactions.^34–37 This structural change decreases the E_AFE→FE and E_FE→AFE difference, resulting in a slimmer hysteresis.³⁴ Similar effects were also reported in FE perovskites, where disrupted long-range order with doping leads to relaxor ferroelectric (RFE) behavior with decreased hysteresis and P_r.³⁸

Although the electrical characteristics of Y:Hf_0.5Zr_0.5O₂ do not fully correspond to RFE or RAFE, the incorporation of heterovalent Y ions at Hf and Zr sites may induce similar local structural inhomogeneity. Such inhomogeneity may inhibit continuous interface-to-interface nucleation of PO-phase domains from the t-phase matrix or provide defect-induced nucleation sites.³⁰ This change may decrease the transient FFE domain size, resulting in RFE- or RAFE-like slim hysteresis and lower P_r. Although Fig. 1a–c and Fig. S1a–e indicate lattice expansion from Y-diffusion, they do not provide direct evidence of local structural disorder. Fig. S10a–c, e and f show the Cs-TEM, iFFT, and line scan images for the 5.5-nm-thick Y:Hf_0.5Zr_0.5O₂ capacitors in the pristine and stepwise cycled states, respectively. No evident local inhomogeneity or distortion from Y diffusion could be observed, likely due to resolution limitations of Cs-TEM.

In perovskite-based RFE and RAFE systems, local structural inhomogeneity and domain-size reduction were directly confirmed by atomic-resolution imaging techniques such as high-angle annular dark-field scanning TEM or integrated differential phase contrast-STEM (iDPC-STEM), where alternating cation displacements generate antiparallel dipoles observable even in the unbiased pristine state.^37,39,40 However, FFE HZO films lack such dipoles in the non-polar t-phase. FFE domains are only observable upon electric-field-induced transition to the PO-phase by monitoring the oxygen ion displacement.^29,33,41 Therefore, direct confirmation of local structural inhomogeneity in Y:Hf_0.5Zr_0.5O₂ would require in situ biasing with advanced imaging techniques, which is beyond the scope of this study.

The enhanced DRAM characteristics in Y:Hf_0.5Zr_0.5O₂ should be further discussed, considering dopant valency and ionic radius, key factors that influence the electrical properties of doped HfO₂- and HZO-based thin films.^7,26,42,43 In terms of valency, previous studies indicated that substituting tetravalent Hf or Zr with trivalent (Al, Y, Gd, La) or bivalent (Sr, Ca, Mg) dopants decreases the formation energy of oxygen vacancies, generating one-half double positively charged V_O²⁺ per trivalent dopant or one V_O²⁺ per bivalent dopant to maintain charge neutrality.^20,26,44,45 These V_O²⁺ decrease the free energies of both t- and PO-phases, with a more substantial reduction for the t-phase, thereby promoting FFE behavior.^26,31,46 However, aliovalent doping does not always result in V_O²⁺ formation, as complex dopant diffusion behavior can vary significantly depending on concentration and insertion method, as previously reported.^8,20,28,47

The influence of oxygen vacancy distribution in the cycling-induced shifts of E_t→PO and E_PO→t requires further discussion. In HZO-based thin films, oxygen vacancies generally exist as either neutral V_O or double positively charged V_O²⁺ states.⁸⁰ Under an applied electric field during cycling, V_O may be transformed into V_O²⁺.⁸¹ Because of the lower diffusion activation energy of V_O²⁺ (∼0.8 eV) compared to V_O (∼1.0 eV),^46,82 this electric-field-induced transition may enhance their migration and redistribution. Previous studies have indicated that factors such as defect redistribution, depinning, grain size, strain, and internal bias may influence the E_t→PO and E_PO→t transitions in both pristine and cycled states.^8,30,48–51 V_O or V_O²⁺ defects present in the pristine state may serve as nucleation barriers, suppressing the PO-phase nucleation from the t-phase matrix and increasing the E_t→PO and E_PO→t. In contrast, the V_O → V_O²⁺ transition and their redistribution/depinning during cycling decreases E_t→PO and E_PO→t.^8,48

The influence of V_O²⁺ redistribution may be more pronounced in Y:Hf_0.5Zr_0.5O₂ compared to Hf_0.3Zr_0.7O₂ due to the introduction of trivalent Y doping. The migration behavior of V_O²⁺ can be significantly altered when stabilized by aliovalent dopants.^31,52 In such cases, additional V_O²⁺ tends to form near the dopant ion sites in the pristine state, where the V_O²⁺-dopant complex is energetically favorable,^31,53 and the nearby dopant influences the V_O²⁺ mobility.^31,52 A previous density functional theory study indicated that larger trivalent dopants increase the migration barrier for V_O²⁺ moving toward the dopant, but lower it for moving away from the dopant, thereby facilitating more active V_O²⁺ redistribution compared to the undoped case.⁵² Hence, Hf_0.3Zr_0.7O₂, with relatively weaker V_O²⁺ influence, exhibited limited low-field shifts in E_t→PO and E_PO→t. In contrast, Y:Hf_0.5Zr_0.5O₂ exhibited more pronounced reductions, likely due to more active V_O²⁺ redistribution and depinning effects,^8,48 enabling enhanced FFE switching at low voltages after stepwise cycling. Nonetheless, direct evidence of V_O²⁺ migration would require in situ biasing combined with advanced techniques such as iDPC-STEM, energy-dispersive spectroscopy, or electron energy-loss spectroscopy, which are beyond the scope of this study.^29,54–56

The impact of dopant ionic radius on the changes in E_t→PO and E_PO→t should also be considered. Although studies on FFE doped-HfO₂ or HZO-based thin films remain limited compared to those on FE behavior, it was reported that dopants with smaller ionic radii than Hf or Zr, such as Al and Si, effectively suppress P_r increase and minimize changes in E_t→PO and E_PO→t during cycling.^26,28,43 For instance, ∼9-nm-thick Si:HfO₂ displayed no P_r increase or noticeable shifts in E_t→PO and E_PO→t even after 10⁷ cycles.⁴³ Similarly, ∼10-nm-thick Al:Hf_0.5Zr_0.5O₂ exhibited minimal P_r increase and only slight E_t→PO changes even after 10⁹ high field cycles at 6 MV cm⁻¹, thereby retaining a high E_t→PO and demonstrating excellent energy storage reliability.²⁸

In contrast, Y-doping in the present study led to a significant decrease in E_t→PO. Previous experimental and kinetic studies reported that dopants with smaller ionic radii induce metal–oxygen bond contraction, thereby favoring stabilization of the t-phase (metal-oxygen bond lengths of ∼2.08 and ∼2.34 Å).²⁶ In contrast, larger dopants tend to elongate bonds and may favor the PO-phase with longer bond lengths (∼2.24, ∼2.32, and ∼2.72 Å).²⁶ Y-diffusion stabilized the t-phase despite causing lattice expansion (Fig. 1 and Fig. S1), due to V_O²⁺ formation for charge compensation (Fig. S2). However, compared to smaller dopants such as Al and Si, Y may be less effective in stabilizing the t-phase due to its larger size.^7,26,27,43 Consequently, Al- and Si-doped films in previous studies retained high E_t→PO even after cycling,^28,43 whereas Y:Hf_0.5Zr_0.5O₂ exhibited more pronounced reductions in E_t→PO and E_PO→t due to a more easily decreased free energy difference between the t- and PO-phases.²⁹ To further elucidate these trends, future studies should systematically investigate various dopants with different valencies and ionic radii under identical doping schemes and cycling conditions.

DRAM characteristics of Y:Hf_0.5Zr_0.5O₂ films with changing thickness

This section describes the performances of the Y:Hf_0.5Zr_0.5O₂ films with POT of 4.1, 4.6, 5.1, 5.5, 5.9, and 6.5 nm, each with a single ALD cycle of Y₂O₃ inserted at the center location of the film, as shown in Fig. 7. The 5.5-nm-thick film corresponds to the sample evaluated in Sections 2.1 and 2.2. Shin et al. reported that adjusting the number of Y₂O₃ ALD cycles in ∼10-nm-thick Hf_0.5Zr_0.5O₂ films modulates the Y concentration. Insufficient doping leads to limited FFE t-phase stabilization, whereas excessive doping induces structural mosaicity, thereby degrading FFE characteristics.²⁰ Therefore, appropriate Y doping is critical for optimal FFE characteristics.

Fig. 7 Schematics diagrams of the fabricated Y:Hf_0.5Zr_0.5O₂ capacitors with different POT, PMA temperature, Y feeding time and Y atomic concentration.

For the Y:Hf_0.5Zr_0.5O₂ films in this section (4.1–6.5 nm), a single Y₂O₃ ALD cycle exhibited optimal FFE characteristics (data not shown), with the Y feeding time adjusted for each POT to fine-tune the properties. The optimized Y feeding times for 4.1-, 4.6-, 5.1-, 5.5-, 5.9-, and 6.5-nm-thick films were 2.5, 2.5, 3.0, 3.5, 4.0, and 5.0 seconds, respectively. The corresponding Y-concentrations (Y/[Al + Hf + Zr]) measured by EDXRF were 2.52, 2.34, 2.66, 2.60, 2.67, and 2.78 atomic%, respectively. The ToF-SIMS results in Fig. S4a–c confirm similar Y ion profiles distributed within the Hf_0.5Zr_0.5O₂ for the 4.6-, 5.5-, and 6.5-nm-thick samples. PMA temperature was also adjusted for different POTs to ensure sufficient crystallization. The 5.1-, 5.5-, 5.9-, and 6.5-nm-thick films were sufficiently crystallized by PMA at 450 °C for 30 seconds, whereas the thinner 4.1- and 4.6-nm-thick films required 500 °C for 30 seconds for the crystallization. Adding dopants may increase the PMA temperature of HZO and deteriorate its crystallization.²¹ However, comparative GIXRD analysis of undoped 5.3-nm-thick Hf_0.5Zr_0.5O₂ and 5.5-nm-thick Y:Hf_0.5Zr_0.5O₂ films PMA-treated at 400, 450, and 500 °C confirmed that Y₂O₃ insertion did not increase the crystallization temperature but instead slightly promoted grain growth, as previously reported (Fig. S11).^18,20

For each sample, stepwise cycling was performed following the same cycling sequence of E_cycle = 6.0 MV cm⁻¹, 10⁵ cycles → E_cycle = 5.0 MV cm⁻¹, 10⁵ cycles → E_cycle = 4.0 MV cm⁻¹, 10⁷ cycles. The P–E and I–E curve changes, non-switching P–E curves, Q_c and Q_d characteristics from non-switching pulse measurements, k–E curves, EOT–E curves, and J–V curves before and after stepwise cycling were examined, as for the 5.5-nm-thick Y:Hf_0.5Zr_0.5O₂ film discussed in Section 2.2. Fig. S12–S15 showed the results for the 4.6-, 5.1-, 5.9-, and 6.5-nm-thick Y:Hf_0.5Zr_0.5O₂ films, respectively. The 4.1-nm-thick film exhibited high leakage contributions even in its pristine state, and cycling at 6.0 MV cm⁻¹ for only 10⁵ cycles (1 second) caused further degradation, as shown in Fig. S16. Hence, it was excluded from further analysis.

Fig. 8a shows the k and EOT values of the 4.6-, 5.1, 5.5-, 5.9-, and 6.5-nm-thick Y:Hf_0.5Zr_0.5O₂ films in the pristine (green stars) and stepwise cycled (purple circles) states at +0.8 V (closed) and −0.8 V (open), respectively, with the values extracted from the non-switching pulse measurements shown in Fig. 5f and Fig. S12–S15f for k, and Fig. 6b and Fig. S12–S15g for EOT. In the pristine state, the E_t→PO distributions in all samples were located at fields >∼2–2.5 MV cm⁻¹ in the I–E curves (Fig. 4a and Fig. S12–S15c). These high switching fields suppressed FFE switching at ±0.8 V, resulting in commonly low k values (∼28–34) (top part of Fig. 8a, green stars), which is consistent with previously reported values for the t-phase.^11,17,29

Fig. 8 (a) k and EOT values of the 4.6-, 5.1-, 5.5-, 5.9-, and 6.5-nm-thick Y:Hf_0.5Zr_0.5O₂ films in the pristine (green stars) and stepwise cycled (purple circles) states at +0.8 V (closed) and −0.8 V (open), respectively. (b) The Δk = k_{Stepwise cycled} − k_Pristine and ΔEOT = EOT_Pristine − EOT_{Stepwise cycled} values obtained from the k and EOT values in Fig. 8(a). (c) The J–EOT characteristics of the 4.6-, 5.1-, 5.5-, 5.9-, and 6.5-nm-thick Y:Hf_0.5Zr_0.5O₂ films in the pristine (green stars) and stepwise cycled (purple circles) states at +0.8 V (closed) and −0.8 V (open), respectively. (d) Comparison of the EOT–POT characteristics of past (Hf,Zr)O₂-, TiO₂-, and (Sr,Ti)O₃-based thin films with this study.

While DRAM dielectrics generally exhibit decreased k with decreasing POT due to the influence of low-k interfacial layers, Park et al. reported that Hf_0.5Zr_0.5O₂ films display increasing k with decreasing POT due to decreased lower-k PO-phase and increased higher-k t-phase.¹⁷ In this study, however, the Y-concentrations were optimized for each POT to stabilize the FFE t-phase, resulting in comparable k values across the samples. Consequently, with similar k, decreased from ∼0.86–0.88 nm to ∼0.58–0.61 nm as POT decreased (bottom part of Fig. 8a, green stars).

After stepwise cycling, the Q_c and Q_d values (Fig. 5d and Fig, S12–S15e) increased significantly from low fields, indicating lowered E_t→PO and E_PO→t. Consequently, the k values in the low fields, including ±0.8 V, increased in all cases compared to the pristine states, as shown in the k–E curves (Fig. 5f and Fig. S12–S15f). Notably, the 5.5-nm-thick Y:Hf_0.5Zr_0.5O₂ exhibited the highest k values at ±0.8 V (∼66–68), whereas both thinner (4.6 and 5.1 nm) and thicker (5.9 and 6.5 nm) films showed smaller values (top part of Fig. 8a, purple circles). Consequently, the EOT decreased with decreasing POT between 5.5–6.5 nm, but remained nearly constant or slightly increased between 4.6–5.5 nm (bottom part of Fig. 8a, purple circles).

The changes in k and EOT before and after stepwise cycling were calculated for each film as Δk = k_{Stepwise cycled} − k_Pristine and ΔEOT = EOT_Pristine − EOT_{Stepwise cycled}, from the data in Fig. 8a and b showed the results. Both Δk and ΔEOT reached their maximum values at 5.5 nm and decreased for both thinner and thicker films, indicating inefficient FFE charge boosting. The underlying reasons for this limited performance enhancement differed depending on the thickness. For the thicker 5.9- and 6.5-nm-thick Y:Hf_0.5Zr_0.5O₂ films (blue highlighted region), the limited enhancements were attributed to the increased FE P_r components after stepwise cycling. As shown in Fig. S17a and b, PUND measurements of the 5.5-, 5.9-, and 6.5-nm-thick Y:Hf_0.5Zr_0.5O₂ films with 0.8 V and 6.0 MV cm⁻¹ triangular pulses exhibited progressive increases in 2P_r with increasing POT. Although all films were optimized to exhibit comparable FFE characteristics in the pristine state by adopting controlled Y feeding times, previous studies reported that surface energy effects stabilize the t-phase in thinner films in the 5–10 nm HZO thickness range.^6,10 Consequently, the same stepwise cycling sequence induced a more significant transition from the FFE t-phase to the FE PO-phase in thicker films, limiting the enhancements in both k and EOT.

For the thinner 4.6- and 5.1-nm-thick Y:Hf_0.5Zr_0.5O₂ films (orange highlighted region), the lower enhancements were attributed to limited lowering of the E_t→PO and E_PO→t distributions. Fig. S18a and b show the I–E curve changes of the 4.6- and 5.1-nm-thick films, respectively, after sequential stepwise cycling for the 5.5-nm-thick case in Fig. 4a. Despite applying the same stepwise cycling sequence, the E_t→PO and E_PO→t distribution in the thinner films displayed a less pronounced shift into the ±0.8 V range, as further confirmed by comparing the stepwise cycled I–E curves of the 4.6-, 5.1-, and 5.5-nm-thick Y:Hf_0.5Zr_0.5O₂ films in Fig. S18c. Consequently, the k values at ±0.8 V were lower and the EOT values were larger for the thinner films (Fig. S19a and b) compared to the 5.5-nm-thick case.

Previous studies reported that the energy barrier between the t- and PO-phases increases as the HZO thickness decreases, with both E_t→PO and E_PO→t exhibiting a −2/3 power dependence on film thickness or grain size.^30,50 High-field cycling decreases the free energy difference between the t- and PO-phases, thereby decreasing this energy barrier and shifting the E_t→PO and E_PO→t to lower fields.²⁹ However, in thinner films, the larger free energy difference may prevent sufficient barrier reduction under the same stepwise cycling conditions.

To address this issue, step 3 of the stepwise cycling was extended to E_cycle = 4.0 MV cm⁻¹, 10⁸ cycles (1000 seconds), and the k–E and EOT–E characteristics of the 4.6-, 5.1-, and 5.5-nm-thick Y:Hf_0.5Zr_0.5O₂ films were compared, and Fig. S19c and d show the results. With prolonged cycling at step 3, the thinner 4.6- and 5.1-nm-thick films exhibited relatively increased k and decreased EOT values at ±0.8 V due to further decreased free energy difference between the t- and PO-phases.

Fig. 8c shows the J–EOT characteristics of the 4.6-, 5.1-, 5.5-, 5.9-, and 6.5-nm-thick Y:Hf_0.5Zr_0.5O₂ films in the pristine (green stars) and stepwise cycled (purple circles) states at ±0.8 V (closed: +0.8 V, open: −0.8 V), with the J values obtained from Fig. 6d and Fig. S12–S15h, and EOT values obtained from Fig. 6b and Fig. S12–S15g. In all cases, stepwise cycling increased J, but the increase remained within 1–2 orders of magnitude due to mitigated cycling stress. However, the EOT values were substantially lowered, resulting in significantly improved J–EOT characteristics compared to the pristine states. Notably, the stepwise cycled 5.5-nm-thick Y:Hf_0.5Zr_0.5O₂ film achieved an EOT of ∼0.31 nm while satisfying the DRAM J criterion (J < 10⁻⁷ A cm⁻² at 0.8 V). Notably, the 4.6-nm-thick film achieved an EOT of ∼0.27 nm at 0.8 V; however, the J value exceeds the DRAM operation limit.

Fig. 8d compares the EOT–POT performance of previously reported DRAM dielectrics that satisfy the DRAM J criterion, including ZrO₂-^18,22,23,57 HZO-,^{11,17,58–64} TiO₂-,^65–68 and SrTiO₃-^69–73 based thin films, with the results of this study. The purple star represents the stepwise-cycled 5.5-nm-thick Y:Hf_0.5Zr_0.5O₂ film in this study, which exhibited the best performance among all materials under the same comparison criterion. This result was achieved through the combined effects of Y-doping, which decreased J, and stepwise cycling, which mitigated J degradation and induced low-voltage charge-boosting effects, resulting in a high k (∼68) and record-low EOT (∼0.31 nm) at the smallest POT (∼5.5 nm).

DRAM read/write time and operation reliability

It should be considered whether the k and EOT values obtained from the 20 µs rectangular pulses in this study apply to practical DRAM systems, where write and read operations occur at time scales of 10–20 ns.¹¹ Hyun et al. reported that, when FFE HZO films are integrated into state-of-the-art DRAM capacitors, the RC time constants of the measurement system and practical DRAM system can be calculated using eqn (1),where R_ms, A_ms, C_ms, R_DRAM, A_DRAM, C_DRAM, ε_r, ε₀, and d are the total resistance of the measurement system (including parasitic resistances), capacitor area of the measurement system, total capacitance of the measurement system, total resistance of the DRAM system (including parasitic resistances), capacitor area of the DRAM system, total capacitance of the DRAM system, relative permittivity, vacuum permittivity, and film thickness, respectively.¹¹ In this study, R_ms and A_ms were ∼195 Ω and ∼88 000 µm², respectively, whereas the state-of-the-art DRAMs employ a pillar-type capacitor structure with R_DRAM and A_DRAM of ∼0.1 MΩ and ∼0.127 µm², respectively.^11,74 From eqn (1), the RC ratio between the measurement and practical DRAM systems was calculated as ∼1351. When the 20 µs measurement pulse width was normalized by this ratio, a corresponding DRAM RC time of ∼14.8 ns was obtained, which is within the DRAM read/write operation window (10–20 ns),¹¹ indicating that the DRAM characteristics in this study relevant to practical DRAM operation.

To further evaluate this aspect, non-switching pulse measurements were conducted using 0.8 V rectangular pulses with pulse lengths ranging from 1 to 40 µs on the stepwise-cycled 5.5-nm-thick Y:Hf_0.5Zr_0.5O₂ film. The resulting current curves are shown in Fig. 9a, where positive currents correspond to charging while negative currents correspond to discharging. Detailed discussions regarding the charging and discharging behavior are reported elsewhere.¹¹ Sufficient charging and discharging were observed for pulse lengths above 15 µs, whereas lengths of 10 µs or shorter led to progressively insufficient charging and discharging currents.

Fig. 9 (a) Charging and discharging currents of the 5.5-nm-thick Y:Hf_0.5Zr_0.5O₂ film with application of 0.8 V, 1–40 µs rectangular pulses. (b) k and EOT changes of the 5.5-nm-thick Y:Hf_0.5Zr_0.5O₂ plotted as a function of the DRAM RC time, calculated by normalizing the pulse width with the RC ratio.

The k and EOT values were extracted from the discharging currents using a previously reported method,¹¹ and plotted as a function of DRAM RC time in Fig. 9b, with pulse lengths normalized by an RC ratio of ∼1351. For pulse lengths of 15–40 µs, corresponding to DRAM RC times of 11.1–29.6 ns, minimal variation in k (67–70) and EOT (0.31–0.32 nm) was observed, confirming stable DRAM characteristics (highlighted in yellow) within 10–20 ns. Even at a 10 µs pulse length, equivalent to DRAM RC time of 7.4 ns, k and EOT values of ∼63 and ∼0.34 nm, respectively, were retained. However, further reduction in pulse length, corresponding to DRAM RC times of 3.7 ns or less, led to noticeable degradation in both parameters.

In Sections 2.2 and 2.3, stepwise cycling with high electric fields (6.0, 5.0, and 4.0 MV cm⁻¹) effectively shifted the E_t→PO and E_PO→t distributions to lower fields, thereby enhancing both k and EOT. However, it was reported that subsequent low-field cycling after high-field cycling may partially rejuvenate the E_t→PO and E_PO→t distributions to higher fields.²⁹ Since DRAM operation involves repeated application and removal of 0.8 V rectangular pulses, the reliability of the optimized k and EOT characteristics must be examined.

Therefore, DRAM reliability characteristics were investigated for the 5.5-nm and 4.6-nm-thick Y:Hf_0.5Zr_0.5O₂ films under bipolar 0.8 V rectangular pulses with 20 µs length, as shown in Fig. 10a and b, respectively. The pulse width of 20 µs was selected based on the results of Fig. 9, where normalization using the RC ratio (∼1351) corresponds to a DRAM RC time of ∼14.8 ns, within the 10–20 ns DRAM read/write window, thereby ensuring a practical reliability evaluation under DRAM-relevant time conditions. In step 1, stepwise cycling was conducted for the pristine films to enhance the k and EOT. For the 5.5-nm-thick film, the previous stepwise cycling was adopted, and a k of ∼68 and an EOT of ∼0.31 nm were acquired. For the 4.6-nm-thick film, extended cycling of 4.0 MV cm⁻¹ for 10⁸ cycles (1000 seconds) at step 1 was required to obtain k of ∼65 and an EOT of ∼0.27 nm. To test the endurance performance of the thin films, 5.5- and 4.6-nm-thick films were cycled in step 2 with 0.8 V bipolar rectangular pulses for 10⁹ cycles (∼11 hours). The 5.5-nm-thick film exhibited slight degradations in k (from 68 to 59) and EOT (from 0.31 to 0.36 nm), while the 4.6-nm-thick film showed similar degradation in k (from 65 to 55) and EOT (from 0.27 to 0.32 nm). These degradations were attributed to the partial rejuvenations of the E_t→PO and E_PO→t distributions to higher fields with low-field cycling at 0.8 V,²⁹ as shown in the left parts of Fig. S20a and b. Nevertheless, despite the significantly longer cycling duration at 0.8 V compared to the stepwise cycling in step 1, the extent of rejuvenation was markedly smaller than the initial low-field shift, consistent with the previous report.²⁹ Consequently, feasible k and EOT characteristics were still retained.

Fig. 10 Changes of k and EOT in (a) 5.5-nm-thick and (b) 4.6-nm-thick Y:Hf_0.5Zr_0.5O₂ films under pulse conditions similar to DRAM operation. (Step 1): stepwise cycling with 100 kHz bipolar triangular pulses at 6.0 MV cm⁻¹ (10⁵ cycles) → 5.0 MV cm⁻¹ (10⁵ cycles) → 4.0 MV cm⁻¹ (10⁷ cycles) for (a), and 6.0 MV cm⁻¹ (10⁵ cycles) → 5.0 MV cm⁻¹ (10⁵ cycles) → 4.0 MV cm⁻¹ (10⁸ cycles) for (b). (Step 2): bipolar rectangular pulses of 0.8 V with 20 µs width for 10⁹ cycles. (Step 3): 100 kHz bipolar triangular pulses at 4.0 MV cm⁻¹ for 10⁶ cycles. (Step 4): reapplication of 0.8 V bipolar rectangular pulses with 20 µs width for 10⁹ cycles.

To recover the degraded properties, bipolar triangular pulses of 4.0 MV cm⁻¹ were applied for 10⁶ cycles (10 seconds) in step 3 for both films. The right parts of Fig. S20a and b show that this short high-field cycling effectively restored the E_t→PO and E_PO→t distributions to their original lower-field positions. Consequently, both the 5.5- and 4.6-nm-thick films recovered their k and EOT values to those achieved after the initial stepwise cycling in step 1. This reversible restoration originated from the field-dependent reversibility of the FFE onset fields, as previously reported,²⁹ establishing a distinctive recoverable reliability mechanism in FFE HZO films that differentiates them from other DRAM dielectric films. In step 4, reapplying 0.8 V bipolar rectangular pulses for 10⁹ cycles again induced similar levels of slight degradation as in step 2.

This behavior is similar to the wake-up → fatigue → re-wake-up process commonly observed in ferroelectric RAM using HZO films, where initial high-field cycling increases P_r (wake-up), prolonged cycling operation leads to P_r degradation (fatigue), and subsequent high-field cycling restores P_r for further operation (re-wake-up).^19,75,76 In this study, a similar strategy was demonstrated in Y:Hf_0.5Zr_0.5O₂ films for FFE DRAM operation. Initial stepwise cycling achieved high k and low EOT values (step 1), which were slightly degraded by extended low voltage operations due to partial rejuvenation of the E_t→PO and E_PO→t distributions (step 2), but were effectively recovered with short high-field cycling (step 3). This rapid recovery capability confirms the feasibility of Y:Hf_0.5Zr_0.5O₂ films for prolonged and repeatable DRAM operation while maintaining optimized dielectric characteristics.

The feasibility of the proposed cycling method for both initial preconditioning and reconditioning during DRAM operation was further analyzed. State-of-the-art 16-gigabit DRAMs contain 32 independent banks, each with 8192 bitlines (BLs) and 65 536 wordlines (WLs).⁷⁷ While only a single WL can be accessed during DRAM read/write operations to avoid BL interference, 16 WLs per bank can be accessed simultaneously during non-operational activities such as stress testing or preconditioning.⁷⁸ In this study, stepwise cycling was performed using 100 kHz bipolar triangular pulses (10 µs duration). When normalized by the RC ratio of approximately 1351, this duration matches a DRAM time scale of around 7.4 ns, similar to typical DRAM operation (∼10 ns). Therefore, the total cycling time under actual DRAM conditions can be estimated by:

Based on eqn (2), the total time required for stepwise cycling (∼10⁷ cycles) to precondition all 16 GB of DRAM cells was estimated at ∼410 seconds (∼7 minutes), which is not a significant additional burden, since current industrial test procedures take up to ∼1 hour per chip. For reconditioning between DRAM operations, requiring 10⁶ cycles (Fig. 10), the estimated time is about 41 seconds. While preconditioning benefits from simultaneous cycling across all banks to reduce overall duration, reconditioning can be selectively applied to individual banks during idle periods, similar to refresh operations in conventional DRAM systems, reducing the cycling burden and making reconditioning more practical. Further reductions in cycling time, through pulse optimization or dopant engineering, could enable even faster preconditioning and reconditioning.

Hysteresis loss

The enclosed area of the P–E hysteresis loop in FFE HZO films corresponds to energy loss, which may cause undesirable heating and power dissipation during DRAM operation.¹¹ While energy loss was conventionally calculated from the area of P–E hysteresis loops measured with 1 kHz triangular pulses, Hyun et al. proposed that short rectangular pulses provide a more accurate evaluation, as they closely reflect actual DRAM operating conditions.¹¹ In the previous study, a significant energy loss of ∼60% was reported for field-cycled 6.5-nm-thick Hf_0.3Zr_0.7O₂ films.¹¹

In this study, energy loss was evaluated using the same method with rectangular pulses of ±0.8 V and a duration of 20 µs for both biases. Fig. S21a–f show the transient current–time, voltage–time, and charge–time curves for obtaining the P–V curves. Fig. 11a–c show the calculated P–V curves, and Fig. 11d–f show the normalized energy-charge curves for the 5.3-nm-thick Hf_0.5Zr_0.5O₂ (pristine), 5.4-nm-thick Hf_0.3Zr_0.7O₂ (pristine and stepwise cycled), and 5.5-nm-thick Y:Hf_0.5Zr_0.5O₂ (pristine and stepwise cycled) films. The corresponding energy loss values are summarized in Table 1. The details of the calculation and discussion on this evaluation method are provided in the previous report.¹¹

Fig. 11 The calculated polarization–voltage curves of the (a) 5.3-nm-thick Hf_0.5Zr_0.5O₂ (pristine), (b) 5.4-nm-thick Hf_0.3Zr_0.7O₂ (pristine and stepwise cycled), and (c) 5.5-nm-thick Y:Hf_0.5Zr_0.5O₂ (pristine and stepwise cycled) films, respectively. The normalized energy-charge curves during charging and discharging of the (a) 5.3-nm-thick Hf_0.5Zr_0.5O₂ (pristine), (b) 5.4-nm-thick Hf_0.3Zr_0.7O₂ (pristine and stepwise cycled), and (c) 5.5-nm-thick Y:Hf_0.5Zr_0.5O₂ (pristine and stepwise cycled) films, respectively.

Table 1 The energy loss values of the 5.3-nm-thick Hf_0.5Zr_0.5O₂ (pristine), 5.4-nm-thick Hf_0.3Zr_0.7O₂ (pristine and stepwise cycled), and 5.5-nm-thick Y:Hf_0.5Zr_0.5O₂ (pristine and stepwise cycled) films at ±0.8 V

		5.3-nm-thick Hf0.5Zr0.5O2	5.4-nm-thick Hf0.3Zr0.7O2		5.5-nm-thick Y:Hf0.5Zr0.5O2
		Pristine	Pristine	Stepwise cycled	Pristine	Stepwise cycled
Energy loss (%)	+0.8 V	19.9	0.2	26.0	0.9	27.8
	−0.8 V	17.6	0.9	44.8	0.6	28.2

---

In the pristine state, the 5.3-nm-thick Hf_0.5Zr_0.5O₂ film exhibited energy loss of 17.6–19.9%, whereas the 5.4-nm-thick Hf_0.3Zr_0.7O₂ and 5.5-nm-thick Y:Hf_0.5Zr_0.5O₂ films showed minimal energy loss (<1%) for both biases. This difference is attributed to their distinct FFE onset fields.¹¹ The Hf_0.3Zr_0.7O₂ and Y:Hf_0.5Zr_0.5O₂ films exhibited high FFE onset fields (∼2.5–3 MV cm⁻¹), exceeding the applied field (±0.8 V/5.4–5.5 nm ≈ ±1.45–1.48 MV cm⁻¹), inhibiting FFE switching and resulting in low k values (∼30 and ∼33, respectively, Fig. 3e). In contrast, the 5.3-nm-thick Hf_0.5Zr_0.5O₂ films exhibited a lower FFE onset field due to decreased Zr content,¹¹ enabling partial FFE switching, resulting in slightly higher k (∼35, Fig. 3e), but also increased energy loss.

After stepwise cycling, both the 5.4-nm-thick Hf_0.3Zr_0.7O₂ and 5.5-nm-thick Y:Hf_0.5Zr_0.5O₂ films exhibited increased energy loss due to the enabled FFE switching at ±0.8 V. In 5.4-nm-thick Hf_0.3Zr_0.7O₂, asymmetric onset field modulation was previously observed, with a higher onset field under positive bias (Fig. 5a and c). Consequently, the energy loss was ∼26.0% for the positive bias (corresponding k: ∼42, Fig. 5e), while a higher loss of ∼44.8% was observed for the negative bias (corresponding k: ∼59, Fig. 5e) due to enhanced FFE switching.¹¹ In contrast, the 5.5-nm-thick Y:Hf_0.5Zr_0.5O₂ film displayed more symmetric onset field reduction (Fig. 5b and d). Although higher and symmetric k values were achieved at ±0.8 V (∼66–68) (Fig. 5f), the energy loss remained relatively lower for both biases (∼27.8–28.2%). This trend is attributed to the more minor field separation between the E_t→PO and E_PO→t distributions in the I–E curves after stepwise cycling (Fig. 4a), resulting in decreased hysteresis (Fig. 4b and 5b).

Fig. S22a–h show the calculated P–V and normalized energy-charge curves, respectively, for the 4.6-, 5.1-, 5.9-, and 6.5-nm-thick Y:Hf_0.5Zr_0.5O₂ films in the pristine and stepwise cycled states. The corresponding energy loss values are summarized in Table S1. All pristine films showed minimal energy loss, whereas stepwise cycled films exhibited values ranging from 11.4% to 30.2%. The optimized Y:Hf_0.5Zr_0.5O₂ films achieved high k and low EOT values, with energy losses decreased to below ∼30%, significantly lower than those in the previous study.¹¹ These results highlight the effectiveness of the proposed Y-doping and cycling strategy in reducing hysteresis for DRAM applications. Future studies should further minimize energy loss through additional doping or cycling optimization methods.

Experimental section/methods

Sample preparation

A TiN bottom electrode (BE) (50 nm) was deposited on a thermally oxidized SiO₂ (100 nm)/Si substrate via sputtering (ENDURA 5500, Applied Materials). All HZO thin films were deposited via thermal ALD at a substrate temperature of 285 °C. The Hf and Zr precursors used were Hf[N(C₂H₅)CH₃]₄ (TEMA-Hf) and Zr[N(C₂H₅)CH₃]₄ (TEMA-Zr), respectively, with ozone (concentration: 180 g m⁻³) as the oxygen source. The growth rates per cycle were ∼0.105 nm for both HfO₂ and ZrO₂. Due to the almost identical ALD growth rates of HfO₂ and ZrO₂, the Hf:Zr composition could be controlled by changing the ALD cycle ratios for HfO₂ and ZrO₂. Y₂O₃ dopants were ex situ deposited using a separate thermal ALD at a substrate temperature of 250 °C. Y(EtCp)₂(iPr-amd) and ozone (concentration: 260 g m⁻³) were used as Y precursor and oxygen source, respectively. The Y doping concentration in Y:Hf_0.5Zr_0.5O₂ films (4.1–6.5 nm) was modulated by varying the Y precursor feeding time between 2.5 and 5 seconds within the Y₂O₃ precursor pulse-purge-reactant pulse-purge sequence (x-17-3-5 s), as illustrated in Fig. 7. Circular Pt (30 nm)/TiN (7 nm) top electrodes (TEs) (∼90 000 µm²) were deposited by sputtering using metal shadow masks (CDS 5000, SNTEK), and the electrode areas were accurately measured using an optical microscope. PMA and PDA were performed at 400, 450, or 500 °C for 30 seconds under an N₂ atmosphere to induce crystallization. Films with a thickness of 5.3 nm or greater were sufficiently crystallized at 450 °C, whereas thinner films (4.1 and 4.6 nm) required a higher temperature of 500 °C for crystallization.

Physical and electrical characterization

Film thicknesses were determined using spectroscopic ellipsometry (SE; M-2000, J. A. Woollam). The layer densities of Hf, Zr, and Y (µg cm⁻²), cation composition, and Y-concentration (Y/[Y + Hf + Zr]) were measured using EDXRF (Quant'X, Thermo SCIENTIFIC). Structural and crystallization characteristics were examined using GIXRD (X'pert Pro, PANalytical, incidence angle = 0.5°) and Cs-TEM (JEM-ARM200F, JEOL). Depth profiling was performed using ToF-SIMS (TOF.SIMS 5, ION-TOF). XPS (Axis Supra, Kratos) was used to analyze the chemical bonding states of Y₂O₃ and HZO films, with binding energies calibrated using the adventitious carbon signal (284.5 eV). Grain radii of the HZO films were analyzed using the watershed method implemented in Gwyddion software, based on plane-view images obtained from SEM (SIGMA, Carl Zeiss). P–E and I–E measurements were conducted at 1 kHz using bipolar or unipolar triangular pulses with a ferroelectric tester (TF Analyzer 2000, Aixacct Systems). Field cycling and stepwise cycling with 100 kHz bipolar triangular pulses were performed using the same ferroelectric tester. Non-switching pulse measurements were conducted with rectangular pulses (rising and falling time: 10 ns) applied using a pulse generator (81110A, Agilent) with a 50 Ω internal resistance, and the current responses were monitored via an oscilloscope (TDS684Dm, Tektronix) with a 50 Ω internal resistance. J–V measurements were conducted using a pico-ammeter/DC voltage source (HP 4140B, Hewlett-Packard). Bipolar 0.8 V rectangular pulses with 20 µs width were applied using a waveform generator (PicoScope 3403D, Pico Technology). The device's TE was connected to the waveform generator, and the BE was grounded to establish the potential difference. The PicoScope was operated via PicoScope 7 T&M software.

Conclusions

This study systematically investigated the FFE behavior of Y:Hf_0.5Zr_0.5O₂ thin films for DRAM applications, focusing on enhancing low-voltage charge boosting, decreasing hysteresis, suppressing P_r increase, and decreasing J degradation. A single ALD cycle of Y₂O₃, inserted in the center location of 4.1–6.5 nm Hf_0.5Zr_0.5O₂ films with optimized Y precursor feeding time for each thickness, resulted in the substitutional diffusion of Y ions. This diffusion suppressed the FE polar orthorhombic phase and stabilized the FFE tetragonal phase.

A novel stepwise cycling strategy was proposed to decrease the high onset field of FFE films. By progressively lowering the cycling field amplitude, the FFE switching peak distributions were effectively shifted toward lower fields, enhancing charge boosting at low voltages while suppressing the increase in P_r and mitigating J degradation. Consequently, the stepwise cycled 5.5-nm-thick Y:Hf_0.5Zr_0.5O₂ film exhibited optimized DRAM characteristics, with a high k of ∼68 and a remarkably low EOT of ∼0.31 nm. To the authors’ knowledge, this is the lowest reported EOT among all DRAM dielectrics, including (Hf,Zr)O₂-, TiO₂-, and (Sr,Ti)O₃-based thin films, while satisfying the DRAM leakage criterion (J < 10⁻⁷ A cm⁻² at 0.8 V). Also, an even lower EOT of ∼0.27 nm was achieved for the stepwise cycled 4.6-nm-thick Y:Hf_0.5Zr_0.5O₂ film, but the J level was higher than the criterion. The hysteresis loss of the stepwise cycled Y:Hf_0.5Zr_0.5O₂ films remained below 30%, marking a significant improvement over the previously reported value of ∼60% for field cycled Hf_0.3Zr_0.7O₂ thin films. These advantages were further highlighted when compared to the undoped 5.4-nm-thick Hf_0.3Zr_0.7O₂ film, which exhibited larger hysteresis, increased P_r, and limited modulation of FFE switching peak positions toward lower fields, resulting in relatively degraded DRAM characteristics.

The enhanced DRAM characteristics in Y:Hf_0.5Zr_0.5O₂ may be attributed to dopant-induced local structural inhomogeneity, which disrupts long-range FFE domain nucleation, resulting in slim hysteresis. In addition, the effective redistribution of V_O²⁺ during cycling, generated by the charge neutrality effect from Y-diffusion, and the larger ionic radius of Y compared to Hf and Zr, may have facilitated more pronounced shifts in the FFE switching peak distributions towards lower fields, improving DRAM characteristics at low voltages. However, direct evidence of these mechanisms was not obtained in this study, highlighting the need for future investigations using in situ biasing, high-resolution structural analysis, and systematic comparisons across dopants with varying valence states and ionic radii.

The practical applicability of the optimized Y:Hf_0.5Zr_0.5O₂ films under practical DRAM operating conditions was further confirmed by analyzing the pulse duration dependence of the extracted dielectric parameters and correlating them with the corresponding DRAM RC time constants. The stepwise cycled 5.5-nm-thick Y:Hf_0.5Zr_0.5O₂ film retained stable k and EOT values across the DRAM read/write time window of 10–20 ns, confirming its suitability for high-speed DRAM operation. In terms of reliability, only minor degradation in DRAM characteristics was observed, even after 10⁹ cycles (∼11 hours) of 0.8 V bipolar rectangular pulses, attributed to the partial rejuvenation of the FFE switching peak distributions. Notably, the degraded characteristics were fully recovered with short high-field cycling (10 seconds).

Author contributions

J. Shin designed and performed the experiments and wrote the manuscript. H. Seo, T. K. Kim, H. S. Park, and S. J. Shin helped with the maintenance of the ALD equipment. H. Paik, H. Song, H. Yoon, K. D. Kim, J. H. Song, S. Lee, and S. Choi contributed to the discussion. J. Han, D. H. Shin, and J. Choi assisted with the electrical measurement setup and interpretation. C. S. Hwang supervised the whole research and manuscript preparation.

Conflicts of interest

There are no conflicts to declare.

Data availability

The data supporting this article have been included as part of the supplementary information (SI). Supplementary information is available. See DOI: https://doi.org/10.1039/d5mh01563d.

Acknowledgements

This work was supported by the Next-generation Intelligence Semiconductor Foundation (2022M3F3A2A01079620) and SK hynix Inc. under its Center of Material Research for Semiconductors (C-MRS) program.

Notes and references

1. S. K. Kim, S. W. Lee, J. H. Han, B. Lee, S. Han and C. S. Hwang, Adv. Funct. Mater., 2010, 20, 2989–3003.
2. C. S. Hwang, Adv. Electron. Mater., 2015, 1, 1400056.
3. M. H. Park, H. J. Kim, Y. J. Kim, T. Moon and C. S. Hwang, Appl. Phys. Lett., 2014, 104, 072901.
4. H. J. Kim, M. H. Park, Y. J. Kim, Y. H. Lee, W. Jeon, T. Gwon, T. Moon, K. D. Kim and C. S. Hwang, Appl. Phys. Lett., 2014, 105, 192903.
5. M. H. Park, H. J. Kim, Y. J. Kim, T. Moon, K. D. Kim and C. S. Hwang, Adv. Energy Mater., 2014, 4, 1400610.
6. M. H. Park, H. J. Kim, Y. J. Kim, Y. H. Lee, T. Moon, K. D. Kim, S. D. Hyun and C. S. Hwang, Appl. Phys. Lett., 2015, 107, 192907.
7. M. H. Park, Y. H. Lee, H. J. Kim, Y. J. Kim, T. Moon, K. D. Kim, J. Mueller, A. Kersch, U. Schroeder and T. Mikolajick, Adv. Mater., 2015, 27, 1811–1831.
8. M. H. Park, H. J. Kim, Y. J. Kim, Y. H. Lee, T. Moon, K. D. Kim, S. D. Hyun, F. P. G. Fengler, U. Schroeder and C. S. Hwang, ACS Appl. Mater. Interfaces, 2016, 8, 15466.
9. K. D. Kim, Y. H. Lee, T. Gwon, Y. J. Kim, H. J. Kim, T. Moon, S. D. Hyun, H. W. Park, M. H. Park and C. S. Hwang, Nano Energy, 2017, 39, 390–399.
10. M. H. Park, Y. H. Lee, H. J. Kim, T. Schenk, W. Lee, K. Do Kim, F. P. Fengler, T. Mikolajick, U. Schroeder and C. S. Hwang, Nanoscale, 2017, 9, 9973–9986.
11. S. D. Hyun, H. W. Park, M. H. Park, Y. H. Lee, Y. B. Lee, B. Y. Kim, H. H. Kim, B. S. Kim and C. S. Hwang, Adv. Electron. Mater., 2020, 6, 2000631.
12. M. H. Park, D. H. Lee, K. Yang, J.-Y. Park, G. T. Yu, H. W. Park, M. Materano, T. Mittmann, P. D. Lomenzo and T. Mikolajick, J. Mater. Chem. C, 2020, 8, 10526–10550.
13. U. Schroeder, M. H. Park, T. Mikolajick and C. S. Hwang, Nat. Rev. Mater., 2022, 7, 653–669.
14. J. Muller, T. S. Boscke, U. Schroder, S. Mueller, D. Brauhaus, U. Bottger, L. Frey and T. Mikolajick, Nano Lett., 2012, 12, 4318–4323.
15. R. Materlik, C. Künneth and A. Kersch, J. Appl. Phys., 2015, 117, 134109.
16. M. H. Park, H. J. Kim, Y. H. Lee, Y. J. Kim, T. Moon, K. Do Kim, S. D. Hyun and C. S. Hwang, Nanoscale, 2016, 8, 13898–13907.
17. M. H. Park, Y. H. Lee, H. J. Kim, Y. J. Kim, T. Moon, K. D. Kim, S. D. Hyun and C. S. Hwang, ACS Appl. Mater. Interfaces, 2018, 10, 42666–42673.
18. H. Seo, I. W. Yeu, D. S. Kwon, D. G. Kim, J. Lim, T. K. Kim, H. Paik, J. H. Choi and C. S. Hwang, Adv. Electron. Mater., 2022, 8, 2200099.
19. M. Pešić, F. P. G. Fengler, L. Larcher, A. Padovani, T. Schenk, E. D. Grimley, X. Sang, J. M. LeBeau, S. Slesazeck and U. Schroeder, Adv. Funct. Mater., 2016, 26, 4601–4612.
20. J. Shin, H. Seo, K. H. Ye, Y. H. Jang, D. S. Kwon, J. Lim, T. K. Kim, H. Paik, H. Song and H. N. Kim, J. Mater. Chem. C, 2024, 12, 5035–5046.
21. A. G. Chernikova, M. G. Kozodaev, D. V. Negrov, E. V. Korostylev, M. H. Park, U. Schroeder, C. S. Hwang and A. M. Markeev, ACS Appl. Mater. Interfaces, 2018, 10, 2701–2708.
22. S. Lee, J. Nam, Y. Choi, J. Jeong, M. K. Nam, H. Oh, H. Kim, Y. Park, Y. Kim and W. Jeon, ACS Appl. Electron. Mater., 2024, 6, 8979–8986.
23. J. Nam, S. Lee, H. Oh, H. Kim, Y. Park, I.-H. Baek and W. Jeon, Mater. Sci. Semicond. Process., 2025, 192, 109485.
24. E. D. Grimley, T. Schenk, X. Sang, M. Pešić, U. Schroeder, T. Mikolajick and J. M. LeBeau, Adv. Electron. Mater., 2016, 2, 1600173.
25. C. Richter, T. Schenk, M. H. Park, F. A. Tscharntke, E. D. Grimley, J. M. LeBeau, C. Zhou, C. M. Fancher, J. L. Jones and T. Mikolajick, Adv. Electron. Mater., 2017, 3, 1700131.
26. M. H. Park, T. Schenk, C. M. Fancher, E. D. Grimley, C. Zhou, C. Richter, J. M. LeBeau, J. L. Jones, T. Mikolajick and U. Schroeder, J. Mater. Chem. C, 2017, 5, 4677–4690.
27. M. H. Park, C. C. Chung, T. Schenk, C. Richter, K. Opsomer, C. Detavernier, C. Adelmann, J. L. Jones, T. Mikolajick and U. Schroeder, Adv. Electron. Mater., 2018, 4, 1800091.
28. J. Shin, D. H. Shin, H. Seo, K. Do Kim, S. Choi, T. K. Kim, H. Paik, H. Song, S. Byun and I. S. Lee, Energy Storage Mater., 2025, 104306.
29. J. Shin, D. H. Shin, K. Do Kim, H. Seo, K. H. Ye, J. W. Jeon, T. K. Kim, H. Paik, H. Song and S. H. Lee, J. Mater. Chem. C, 2024, 12, 15423–15434.
30. N. Tasneem, Y. M. Yousry, M. Tian, M. Dopita, S. E. Reyes-Lillo, J. Kacher, N. Bassiri-Gharb and A. I. Khan, Adv. Electron. Mater., 2021, 7, 2100485.
31. J. Lee, K. Yang, J. Y. Kwon, J. E. Kim, D. I. Han, D. H. Lee, J. H. Yoon and M. H. Park, Nano Converge., 2023, 10, 55.
32. T. Schenk, E. Yurchuk, S. Mueller, U. Schroeder, S. Starschich, U. Böttger and T. Mikolajick, Appl. Phys. Rev., 2014, 1, 041103.
33. G. H. Park, D. H. Lee, H. Choi, T. Kwon, Y. H. Cho, S. H. Kim and M. H. Park, ACS Appl. Electron. Mater., 2023, 5, 642–663.
34. Y. Si, T. Zhang, C. Liu, S. Das, B. Xu, R. G. Burkovsky, X.-K. Wei and Z. Chen, Prog. Mater. Sci., 2024, 101231.
35. F. Li, S. Zhang, D. Damjanovic, L. Q. Chen and T. R. Shrout, Adv. Funct. Mater., 2018, 28, 1801504.
36. H. Qi, R. Zuo, A. Xie, A. Tian, J. Fu, Y. Zhang and S. Zhang, Adv. Funct. Mater., 2019, 29, 1903877.
37. L.-F. Zhu, S. Deng, L. Zhao, G. Li, Q. Wang, L. Li, Y. Yan, H. Qi, B.-P. Zhang and J. Chen, Nat. Commun., 2023, 14, 1166.
38. H. Pan, F. Li, Y. Liu, Q. Zhang, M. Wang, S. Lan, Y. Zheng, J. Ma, L. Gu and Y. Shen, Science, 2019, 365, 578–582.
39. T. Zhang, Y. Si, S. Deng, H. Wang, T. Wang, J. Shao, Y. Li, X. Li, Q. Chen and C. Liu, Adv. Funct. Mater., 2024, 34, 2311160.
40. C. Xu, N. Luo, C. Zhong, G. Luo, R. Che, X. Guo, C. Chen, S. Zhang and Y. Zhu, Nat. Commun., 2025, 16, 5438.
41. Y. Cheng, Z. M. Gao, K. H. Ye, H. W. Park, Y. H. Zheng, Y. Z. Zheng, J. F. Gao, M. H. Park, J. H. Choi, K. H. Xue, C. S. Hwang and H. B. Lyu, Nat. Commun., 2022, 13, 645.
42. C.-K. Lee, E. Cho, H.-S. Lee, C. S. Hwang and S. Han, Phys. Rev. B: Condens. Matter Mater. Phys., 2008, 78, 012102.
43. U. Schroeder, E. Yurchuk, J. Müller, D. Martin, T. Schenk, P. Polakowski, C. Adelmann, M. I. Popovici, S. V. Kalinin and T. Mikolajick, Jpn. J. Appl. Phys., 2014, 53, 08LE02.
44. F. Mehmood, T. Mikolajick and U. Schroeder, Appl. Phys. Lett., 2020, 117, 092902.
45. L. Yin, S. Gong, X. Li, B. Lu, Q. Peng, S. Zheng, M. Liao and Y. Zhou, J. Alloys Compd., 2022, 914, 165301.
46. R. He, H. Wu, S. Liu, H. Liu and Z. Zhong, Phys. Rev. B, 2021, 104, L180102.
47. H. Seo, J. Shin, H. S. Park, T. K. Kim, H. Paik, H. Song and C. S. Hwang, J. Mater. Chem. C, 2025, 13, 6257–6266.
48. P. D. Lomenzo, C.-C. Chung, C. Zhou, J. L. Jones and T. Nishida, Appl. Phys. Lett., 2017, 110, 232904.
49. P. D. Lomenzo, M. Materano, T. Mittmann, P. Buragohain, A. Gruverman, T. Kiguchi, T. Mikolajick and U. Schroeder, Adv. Electron. Mater., 2022, 8, 2100556.
50. R. Ganser, P. D. Lomenzo, L. Collins, B. Xu, L. A. Antunes, T. Mikolajick, U. Schroeder and A. Kersch, Adv. Funct. Mater., 2024, 34, 2405513.
51. J. H. Song, K. Do Kim, J. Shin, S. J. Shin, S. H. Lee, S. Y. Byun, I. S. Lee, H. S. Park, Y. J. Kim and H. W. Nam, J. Mater. Chem. C, 2025, 13, 14751–14766.
52. H. Zhang, B. Gao, S. Yu, L. Lai, L. Zeng, B. Sun, L. Liu, X. Liu, J. Lu and R. Han, presented at 2009 International Conference on Simulation of Semiconductor Processes and Devices, San Diego, CA, USA, 10, 2009.
53. J. Yang, Y. Xie, C. Zhu, S. Chen, J. Wei, Y. Liu, M. Chen and D. Cao, Nanotechnology, 2024, 35, 235704.
54. P. Nukala, M. Ahmadi, Y. Wei, S. De Graaf, E. Stylianidis, T. Chakrabortty, S. Matzen, H. W. Zandbergen, A. Björling and D. Mannix, Science, 2021, 372, 630–635.
55. Y. Zheng, Y. Zheng, Z. Gao, J.-H. Yuan, Y. Cheng, Q. Zhong, T. Xin, Y. Wang, C. Liu and Y. Huang, presented at 2021 IEEE International Electron Devices Meeting (IEDM), San Francisco, CA, USA, 12, 2021.
56. L. Chen, Z. Liang, S. Shao, Q. Huang, K. Tang and R. Huang, Nanoscale, 2023, 15, 7014–7022.
57. S. H. Cha, C. H. An, S. T. Cho, D.-G. Kim, D. S. Kwon, J. I. Lim, W. Jeon and C. S. Hwang, Phys. Status Solidi RRL, 2019, 13, 1900282.
58. D. Das and S. Jeon, IEEE Trans. Electron Devices, 2020, 67, 2489–2494.
59. D. Das, B. Buyantogtokh, V. Gaddam and S. Jeon, IEEE Trans. Electron Devices, 2021, 69, 103–108.
60. V. Gaddam, G. Kim, T. Kim, M. Jung, C. Kim and S. Jeon, ACS Appl. Mater. Interfaces, 2022, 14, 43463–43473.
61. S. Oh, H. Jang and H. Hwang, J. Appl. Phys., 2023, 133, 154102.
62. S. Oh, H. Jang and H. Hwang, IEEE Electron Device Lett., 2023, 45, 28–31.
63. H. Shin, G. Kim, S. Lee, H. Choi, S. Lee, S. Lee, Y. Nam, G. Kang, H. Kim and J. Ahn, Phys. Status Solidi RRL, 2024, 18, 2400108.
64. J. Y. Park, H. Choi, J. Lee, K. Yang, S. Y. Lee, D. I. Han, I. Jeon, C. H. Jung, H. Lim and W. Lee, Appl. Surf. Sci. Adv., 2025, 27, 100733.
65. S. K. Kim, G. J. Choi, S. Y. Lee, M. Seo, S. W. Lee, J. H. Han, H. S. Ahn, S. Han and C. S. Hwang, Adv. Mater., 2008, 20, 1429–1435.
66. W. Jeon, S. Yoo, H. K. Kim, W. Lee, C. H. An, M. J. Chung, C. J. Cho, S. K. Kim and C. S. Hwang, ACS Appl. Mater. Interfaces, 2014, 6, 21632–21637.
67. J. Jeon, T. Kim, M. Jang, H. K. Chung, S.-C. Kim, S. O. Won, Y. Park, B. J. Choi, Y. J. Chung and S. K. Kim, Chem. Mater., 2024, 36, 3326–3333.
68. T. K. Kim, H. Seo, J. Lim, H. Paik, J. Shin, H. Song, D. S. Kwon and C. S. Hwang, J. Mater. Chem. C, 2025, 13, 16969–16980.
69. S. W. Lee, J. H. Han, S. Han, W. Lee, J. H. Jang, M. Seo, S. K. Kim, C. Dussarrat, J. Gatineau and Y.-S. Min, Chem. Mater., 2011, 23, 2227–2236.
70. W. Lee, J. H. Han, W. Jeon, Y. W. Yoo, S. W. Lee, S. K. Kim, C.-H. Ko, C. Lansalot-Matras and C. S. Hwang, Chem. Mater., 2013, 25, 953–961.
71. W. Lee, S. Yoo, W. Jeon, Y. W. Yoo, C. H. An, M. J. Chung, H. J. Kim, S. W. Lee and C. S. Hwang, Thin Solid Films, 2015, 589, 723–729.
72. H. K. Chung, J. Jeon, H. Kim, M. Jang, S.-C. Kim, S. O. Won, I.-H. Baek, Y. J. Chung, J. H. Han and S. H. Cho, Appl. Surf. Sci., 2024, 664, 160243.
73. H. Paik, J. Lim, H. Seo, T. K. Kim, J. Shin, H. Song, D. G. Kim, W. Lee, D. S. Kwon and C. S. Hwang, Mater. Horiz., 2025, 12, 7305–7317.
74. M. Popovici, A. Belmonte, H. Oh, G. Potoms, J. Meersschaut, O. Richard, H. Hody, S. Van Elshocht, R. Delhougne and L. Goux, presented at 2018 IEEE International Electron Devices Meeting (IEDM), San Francisco, CA, USA, 12, 2018.
75. Y. Cheng, Z. Gao, K. H. Ye, H. W. Park, Y. Zheng, Y. Zheng, J. Gao, M. H. Park, J.-H. Choi and K.-H. Xue, Nat. Commun., 2022, 13, 645.
76. H. S. Park, S. Choi, K. D. Kim, M. K. Yeom, S. H. Lee, S. K. Ryoo and C. S. Hwang, ACS Appl. Electron. Mater., 2024, 6, 6826–6836.
77. D. Kim, M. Park, S. Jang, J.-Y. Song, H. Chi, G. Choi, S. Choi, J. Kim, C. Kim and K. Kim, presented at 2019 IEEE International Solid-State Circuits Conference (ISSCC), San Francisco, CA, USA, 2, 2019.
78. K. Min, H. Kwon, S. Lee, H. Hong, S. Yim, J. Kim and J.-B. Lee, presented at International Symposium for Testing and Failure Analysis, 12, 2020.
79. M. G. Farahani, C. Magen, A. Quintana, I. Fina and F. Sanchez, ACS Appl. Electron. Mater., 2025, 7, 6628–6634.
80. Y. Lee, S. H. Kim, H. W. Jeong, G. H. Park, J. Lee, Y. Y. Kim and M. H. Park, Appl. Surf. Sci., 2024, 648, 158948.
81. L. Hu, J. Yang, J. Wang, P. Cheng, L. O. Chua and F. Zhuge, Adv. Funct. Mater., 2021, 31, 2005582.
82. M. Materano, P. D. Lomenzo, A. Kersch, M. H. Park, T. Mikolajick and U. Schroeder, Inorg. Chem. Front., 2021, 8, 2650–2672.

---