Suppressing the current leakage and voltage drift in indium-modified electrical switching devices

Abstract

Ovonic threshold switching (OTS) devices are essential selectors for high-density memory arrays, but their operational reliability is often limited by the threshold-voltage (V_th) drift arising from structural relaxation in amorphous chalcogenides. Although widely reported, the atomic mechanism behind the V_th drift remains insufficiently understood, limiting the rational design of highly stable OTS devices. Here, we investigate the atomic origin of the V_th drift in In–Te OTS devices (InTe₄, InTe₃, and In₃Te₇) and demonstrate that reducing homopolar Te–Te bonding effectively suppresses the structural relaxation and V_th drift. A higher In content lowers the leakage current, increases V_th, and significantly suppresses the V_th drift. The optimized In₃Te₇ devices achieve an on/off ratio of ∼10⁵, a low V_th drift coefficient of 26 mV dec⁻¹ and endurance exceeding 10⁸ cycles. Ab initio molecular dynamics (AIMD) simulations further demonstrate that a portion of Te–Te homopolar bonds gradually disappear during structural relaxation and that the increased In content suppresses both homopolar bonding and the formation of over-coordinated Te atoms, which are responsible for defect states inside the bandgap. This study clarifies the origin of V_th drift and provides a practical material-engineering strategy for developing reliable OTS selectors for future 3D memory technologies.

---

1. Introduction

With the rapid development of artificial intelligence, cloud computing, and the Internet of Things (IoT), the demand for high-performance data storage technologies has increased dramatically.^1,2 However, conventional memories are approaching their physical and scaling limits, driving the exploration of emerging nonvolatile memory (NVM) technologies such as resistive random-access memory (RRAM), magnetic RAM (MRAM), ferroelectric RAM (FeRAM), and phase-change memory (PCM).^3–6 Among these candidates, three-dimensional PCM arrays with crossbar structures have attracted significant attention owing to their high scalability, fast operating speed, and excellent data retention, positioning them as strong contenders for next-generation storage-class memory.^7–9 Nevertheless, the reliable operation of PCM arrays faces a major challenge from leakage currents in unselected cells, which can severely degrade read accuracy and energy efficiency. To address this issue, ovonic threshold switching (OTS) devices have been introduced as selectors, exhibiting abrupt and reversible transitions between high- and low-conductance states under an electric field.¹⁰ Their simple structure, high current density, fast switching speed, and compatibility with back-end-of-line processes make OTS devices indispensable components for PCM applications.^11–14

Recently, OTS devices based on a variety of material systems have been developed, showing significant improvements in on/off ratios, endurance, and on-current.^15–17 Among various systems, binary OTS devices have attracted extensive attention due to their simple composition, environmental friendliness, and process compatibility.^18–23 For example, a binary InTe-based OTS device with an HfO₂ buffer layer has demonstrated excellent endurance of up to 10⁹ cycles and a highly stable threshold voltage (V_th).²⁴ However, chalcogenide-based OTS materials inevitably suffer from structural relaxation, during which properties such as bandgaps, capacitance, and resistance gradually evolve over time.^25–28 This intrinsic instability can induce V_th drift, ultimately degrading the read–write margin and overall stability of PCM arrays.²⁹ Consequently, understanding the mechanism behind V_th drift in OTS devices and developing effective strategies to suppress it remain critical challenges.³⁰ In GeSe-based systems, the V_th drift has been linked to the instability of Ge–Ge bonds.³¹ In Ge–As–Se OTS devices, the insertion of a carbon interlayer effectively suppresses both the V_th drift and the leakage current (I_off).³² Moreover, in Ge–Se–As devices, the V_th drift has been associated with defect states originating from Se²⁻ defects.³³

However, despite these advances, the atomic origin of the V_th drift in Te-based OTS devices remains poorly understood. In particular, how the amorphous structure of Te-based OTS materials evolves during aging and how such structural relaxations influence the V_th drift have not yet been clarified. Addressing this knowledge gap is essential for developing drift-resistant OTS selectors for reliable 3D PCM integration. In this work, we experimentally investigate In–Te OTS devices with different In/Te ratios by evaluating their electrical characteristics, V_th drift, optical bandgaps, and chemical bonding configurations. Electrical measurements reveal that increasing the In content effectively suppresses the leakage current and slows the drift of V_th. The optimized In₃Te₇ OTS devices ultimately achieve an on/off ratio of ∼10⁵, endurance beyond 10⁸ cycles, and a low V_th drift coefficient of 26 mV dec⁻¹. Ultraviolet-visible-near infrared (UV-vis-NIR) absorption spectroscopy and X-ray photoelectron spectroscopy (XPS) analyses further show that higher In concentrations widen the bandgap and reduce the presence of unstable Te–Te bonds in the amorphous network. To uncover the underlying mechanism, we employ ab initio molecular dynamics simulations (AIMD). The results demonstrate that during structural relaxation, the population of homopolar Te–Te bonds in a-In₃Te₇ decreases over time, which is related to the time-dependent V_th drift. Furthermore, increasing the In content reduces both the Te–Te homopolar bonds and the population of over-coordinated Te atoms in the amorphous network, which are closely tied to defect states in In–Te OTS materials. This suppression of defect-related structural units accounts for the reduced drift observed in In–Te OTS devices, which also provides a new approach for reducing the V_th drift in other Te-based OTS devices.

2. Results and discussion

2.1 Structural characterization of In–Te OTS devices

To investigate the mechanism of V_th drift in OTS devices, In–Te films were selected as the amorphous functional layers, and three different compositions were prepared for comparison. By tuning the sputtering powers of the InTe₉ and In targets during co-sputtering, we fabricated OTS devices with via-hole structures based on InTe₄, InTe₃, and In₃Te₇. Fig. 1a shows the transmission electron microscopy (TEM) image of In₃Te₇ devices. The functional layer has a thickness of approximately 40 nm, while both the top and bottom electrodes consist of 100 nm-thick W layers, and the via-hole diameter is about 800 nm. The energy dispersive spectroscopy (EDS) mapping of In₃Te₇ devices in Fig. 1b reveals the uniform distributions of W, In, and Te elements. The fast Fourier transform (FFT) pattern of the blue region in Fig. 1c shows diffuse diffraction rings, confirming the amorphous state of the as-deposited In₃Te₇ film, which meets the requirements for OTS devices. Furthermore, the EDS line scan profile in Fig. 1d provides additional evidence for the localized distributions of W, In, and Te across the device stack. The final three compositions of In–Te films were precisely determined by extracting the In/Te ratios from the EDS mappings (Fig. S1 and Table S1), ensuring the desired stoichiometry for OTS material optimization.

Fig. 1 Structural characteristics of In–Te OTS devices. (a) Cross-sectional TEM image of the via-hole In₃Te₇ OTS device. The amorphous In₃Te₇ layer is ∼40 nm thick and is sandwiched between 100 nm W electrodes, with a via-hole diameter of ∼800 nm. (b) EDS elemental mapping of In, Te and W for In₃Te₇ devices, confirming the uniform elemental distribution of all elements. (c) FFT pattern of the amorphous In₃Te₇ layer, showing diffuse rings typical of an amorphous structure. (d) EDS line-scan profile corresponding to (c), revealing the elemental distribution across the device stack.

2.2 Electrical characteristics of In–Te OTS devices

The electrical characteristics of the In–Te OTS devices were subsequently measured to verify the OTS performance. Before achieving reversible switching, OTS devices typically require an initial firing process, during which the firing voltage is higher than the V_th.³⁴ For all the In–Te devices we fabricated, the firing voltages are higher than V_th and range from 2 V to 3.5 V (Fig. S2). After the firing process, the DC I–V characteristics of the OTS devices with a compliance current of 1 mA are shown in Fig. 2a. Initially, the OTS devices remains in a highly resistive off-state. When the applied voltage exceeds V_th, the device undergoes an abrupt transition to a low-resistance on-state, allowing a large current to flow through the amorphous layer. During the reverse I–V sweep, the device reverts to the off-state once the voltage drops below the holding voltage (V_hold). The InTe₄ devices exhibited a V_th of approximately 1.8 V. With increasing In content, the V_th of the InTe₃ and In₃Te₇ devices increased to around 2.2 and 2.7 V, respectively. The I_off read at 1/2V_th is about 200 nA for the InTe₄ devices, whereas it decreases to about 30 nA for the InTe₃ and In₃Te₇ devices. The I_off of In₃Te₇ devices is nearly one order of magnitude lower than that of InTe₄ devices, and is lower than that of many other reported Te-based OTS devices.^19,35 This indicates that increasing the In concentration greatly enhances the device performance, which may be attributed to the modified bandgap and improved structural stability of the amorphous matrix in In–Te films. In addition, all devices with three compositions exhibited on-state currents exceeding 1 mA, corresponding to an on/off ratio approaching 10⁵ in the In₃Te₇ device, which is highly beneficial for enhancing the driving capability and density of 3D PCM arrays in practical applications.

Fig. 2 Electrical characteristics of In–Te OTS devices. (a) DC I–V curves of devices based on InTe₄, InTe₃, and In₃Te₇. A higher In content leads to lower I_off and increased V_th. (b) Pulsed switching tests under 50 consecutive triangular pulses, showing stable switching performance for all devices. (c) V_th distributions over 50 cycles. All devices exhibit low cycle-to-cycle variations, while the V_th increases with the higher In content. (d) Endurance performance under periodic square pulses, demonstrating over 10⁸ stable cycles.

To investigate the dynamic switching of In–Te OTS devices, the pulse-switching characteristics were measured. Triangular voltage pulses with 5 µs rising/falling edges were applied, and a 200 Ω series resistance was used to limit the current. As shown in Fig. 2b, all In–Te devices exhibit stable switching behavior under 50 consecutive triangular pulses, and the V_th values are consistent with those obtained from DC sweeps. The V_hold values of all devices remain nearly constant, which is consistent with the result that V_hold is largely determined by the series resistance in the circuit.³⁶ Furthermore, increasing the In content leads to a higher on-state current, rising from 5.5 mA in the InTe₄ device to 8.7 mA in the In₃Te₇ device. Fig. 2c summarizes the cycle-to-cycle variation of V_th over 50 pulses. For InTe₄ devices, V_th ranges from 1.6 V to 2.0 V, whereas for In₃Te₇ devices, it mainly falls between 2.65 and 3.1 V. All devices with three compositions exhibit relatively low cycle-to-cycle variability, with no significant dependence on In composition. The endurance characteristics of In–Te based OTS devices are shown in Fig. 2d. A sequence of periodic square pulses (a pulse width of 100 ns and a period of 1 µs) were applied, and after a certain number of pulses, triangular voltage pulses were used to verify the switching behavior while recording both the on-state and leakage currents. All In–Te devices achieved over 10⁸ switching cycles, meeting the endurance requirements for practical PCM applications.^37,38

2.3 Time-dependent threshold voltage drift in In–Te OTS devices

To investigate the differences in V_th drift among In–Te based OTS devices with three compositions, we performed pulse measurements using triangular pulses with varying time intervals. Prior to testing, each device was fired with a high-amplitude pulse, followed by 100 triangular pulses to stabilize its switching behavior.³⁹ Next, we switched the device using a single triangular pulse with an amplitude high enough to reliably turn on the device. After allowing the device to relax for a delay time (t_d), an identical triangular pulse was applied to read the V_th. Fig. 3a presents the pulse I–V characteristics of the InTe₄ device under triangular pulses with different t_d. The pulse interval increased from 10⁻⁵ s to 1 s, and the V_th drifted from 1.4 V to 1.7 V. To quantitatively analyze the impact of time on V_th drift, we statistically evaluate multiple V_th values for In–Te based OTS devices with three compositions, as shown in Fig. 3b–d. The degree of V_th drift is assessed using the relationship: V_th = V_th0 + γ log(t_d),⁴⁰ where V_th0 is a constant and γ represents the drift coefficient of the threshold voltage. All In–Te OTS devices exhibit an increase in V_th with longer times. Among them, the InTe₄ devices show the largest drift coefficient of 49 mV dec⁻¹, while the In₃Te₇ devices exhibit the smallest value of 26 mV dec⁻¹. These results indicate that increasing the In content effectively reduces the V_th drift coefficient, which is significantly lower than that of several previously reported OTS devices.^19,32,41

Fig. 3 Time-dependent V_th drift of In–Te OTS devices. (a) V_th evolution of InTe₄ from 10⁻⁵ s to 1 s, showing a drift from ∼1.4 V to ∼1.7 V. An initial high-amplitude triangular pulse turns the OTS device on, followed by a delay time t_d and a second triangular pulse for reading V_th. (b)–(d) V_th drift of InTe₄, InTe₃, and In₃Te₇ OTS devices. In₃Te₇ devices shows the lowest V_th drift, with a drift coefficient of 26 mV dec⁻¹.

2.4 Optical and structural properties of In–Te films

To investigate the mechanism behind the improved OTS performance, we characterized the In–Te films with three compositions (Fig. 4). The UV-vis-NIR absorption spectra of the In–Te films were measured to determine the optical band gap (E_g) (Fig. 4a). The E_g of each composition was extracted by linear fitting according to the Tauc plot: (αhν)^1/2 = C(hν − E_g),⁴² where C is a constant, α is the absorption coefficient, and hν is the photon energy. The optical E_g of the InTe₄ films was determined to be only 0.50 eV. With higher In content, the optical E_g increased to 0.71 and 0.77 eV for InTe₄ and In₃Te₇, respectively. The enlarged optical E_g of InTe₃ and In₃Te₇ can account for the observed increase of approximately 1 V in V_th and a one-order-of-magnitude reduction in I_off in Fig. 2a. This also indicates that the V_th is positively correlated with the film band gap, whereas I_off shows a negative correlation, consistent with previous studies.^43,44Fig. 4b presents the X-ray diffraction (XRD) patterns of the as-deposited and annealed In–Te films with the three compositions. The absence of discernible diffraction peaks in all as-deposited samples confirms their amorphous state. After annealing at 300 °C for 30 minutes, the InTe₄ films showed several peaks corresponding to hexagonal Te crystals, while the InTe₃ and In₃Te₇ films exhibited diffraction peaks corresponding to crystalline In₂Te₃ and Te.^45,46 XPS was employed to analyze the bonding states and binding energies of the In–Te films, as shown in Fig. 4c and d. In InTe₄ films, the In 3d spectrum exhibits two peaks at 444.8 and 452.4 eV, corresponding to the In 3d_5/2 and In 3d_3/2 states, with a spin–orbit separation of 7.6 eV. Meanwhile, two peaks at 572.9 and 583.3 eV are observed in the Te 3d region, attributed to Te 3d_5/2 and Te 3d_5/2, separated by 10.4 eV, which is consistent with previously reported XPS profiles of In–Te thin films.^47,48 With increasing In content, the In 3d peaks exhibit a shift toward higher binding energy. This arises from the strengthened interaction between In and the more electronegative Te atoms, which promotes the formation of more stable heteropolar In–Te bonds. Consequently, the proportion of unstable homopolar Te–Te bonds is reduced in In₃Te₇, enhancing the structural stability of the amorphous network and thereby suppressing the V_th drift.

Fig. 4 Characteristics of In–Te thin films. (a) UV-vis-NIR absorption spectra and Tauc plots, revealing increasing bandgaps from 0.50 eV (InTe₄) to 0.71 eV (InTe₃) and 0.77 eV (In₃Te₇). The larger bandgaps are consistent with higher V_th and reduced I_off. (b) XRD patterns of as-deposited and annealed films. The absence of discernible diffraction peaks in all as-deposited samples confirm their amorphous structures. (c) and (d) XPS In 3d and Te 3d spectra. Increasing In causes the In 3d peaks to shift to higher binding energies.

2.5 Mechanisms of performance improvement in In–Te OTS devices

To gain deeper insight into the effect of increasing the In content on the V_th drift coefficient of amorphous In–Te devices, we conducted first-principles calculations to analyze and compare the amorphous structures of In–Te materials. Amorphous models of a-InTe₄, a-InTe₃, and a-In₃Te₇ were generated using AIMD simulations, and their pair distribution functions (PDFs) were calculated to characterize the atomic-scale structural correlations (Fig. 5a). In a-InTe₄, the intensities of the In–Te and Te–Te peaks are comparable, while the In–In peak is relatively weak, indicating the coexistence of nearly equal amounts of heteropolar In–Te bonds and homopolar Te–Te bonds. As the In concentration increases, the number of Te–Te bonds decreases significantly in a-InTe₃ and a-In₃Te₇, whereas the In–Te bonds become dominant. This can be attributed to the higher number of In cations, which promote the formation of more stable heteropolar In–Te bonds rather than homopolar Te–Te bonds. Fig. 5b shows the coordination number (CN) distribution of Te atoms in a-InTe₄, a-InTe₃, and a-In₃Te₇. In all compositions, Te atoms are predominantly 3-coordinated, with small fractions of 2-, 4-, and a few 5-coordinated atoms. In In–Te systems, Te atoms with CN greater than two are considered over-coordinated species that deviate from the 8-N rule (which states that the ideal coordination number of an element is equal to 8 minus its number of valence electrons N). Mis-coordinated atoms are known to contribute significantly to the formation of defect states in amorphous materials,^49,50 as seen in mis-coordinated Ge and Se atoms in GeSe materials.⁵¹ With the increasing In content, the proportion of over-coordinated Te atoms decreases in a-In₃Te₇, while the number of 2-coordinated Te atoms exceeds the 4-coordinated Te atoms. This structural evolution indicates that higher In content suppresses the formation of unstable over-coordinated Te atoms, thereby reducing the defect density and slowing down atomic relaxation. The CN of In atoms shows no significant change and remains predominantly 4-coordinated with increasing In content (Fig. S3). Furthermore, the mean square displacement (MSD) of atoms in the amorphous In–Te models was calculated to evaluate their atomic mobility, as shown in Fig. 5c. The MSD of a-In₃Te₇ exhibits a pronounced reduction compared to those of a-InTe₄ and a-InTe₃, indicating slower atomic migration and a more rigid amorphous network. Both In and Te atoms show the same decreasing trend in their individual MSD (Fig. S4). As the In concentration increases, the reduced fraction of homopolar bonds and the further stabilization of the amorphous structure collectively contribute to the lower leakage current and suppressed the V_th drift observed in In–Te OTS devices.

Fig. 5 Structural properties, electronic structures and relaxation behavior of a-InTe₄, a-InTe₃, and a-In₃Te₇. (a) Partial pair distribution functions (PDFs), g(r). (b) Coordination-number distribution of Te atoms. The data show fewer over-coordinated Te atoms in a-In₃Te₇. (c) Mean square displacement (MSD). a-In₃Te₇ shows the lowest MSD, indicating slower atomic motion. (d) Density of states (DOS) and its normalized inverse participation ratio (IPR) of a-InTe₄, a-InTe₃, and a-In₃Te₇, respectively. (e) Real-space projections of defect states in a-In₃Te₇. The defect states arise from In–Te clusters and over-coordinated Te atoms. (f) AIMD bond evolution for 60 ps at 600 K in a-In₃Te₇.

Since the switching behavior of OTS devices is strongly correlated with defect states within the bandgap,^52,53 we calculated the electronic properties of the materials using density functional theory (DFT) models. The calculated electronic density of states (DOS) and the corresponding normalized inverse participation ratios (IPR) of a-InTe₄, a-InTe₃, and a-In₃Te₇ are shown in Fig. 5d. IPR serves as a metric for evaluating the localization of electronic states, where states with higher IPR values are more localized. In addition to providing insight into the degree of localization in defect states, the IPR enables the determination of the mobility gap in amorphous glass.⁵⁴ This mobility gap is derived from the energy difference between the valence and conduction band edges, defined by electronic states exhibiting relatively low IPR values. Our calculations show that the E_g values of a-InTe₄, a-InTe₃, and a-In₃Te₇ are 0.52, 0.57, and 0.61 eV, respectively, which follow the same trend as the experimental results in Fig. 4a. In all three amorphous In–Te composition models, defect states are observed, and these states consistently exhibit relatively large IPR values. This indicates that carriers trapped in such highly localized states possess low mobility and thus contribute negligibly to electrical conduction at room temperature. Under an applied electric field, however, these localized defect states become more delocalized, allowing carriers to tunnel into extended valence states and leading to a reduction in device resistance. To identify the structural origins of these defect states, we projected them in a-In₃Te₇ into real space by using the analytical methods for electron wave-functions, as shown in Fig. 5e. The defect states were found to originate from two types of atomic motifs: clusters composed of In and Te atoms and clusters formed by over-coordinated Te atoms. Therefore, over-coordinated Te atoms and homopolar Te–Te bonds play a critical role in the formation of defect states and strongly influence the switching behavior of OTS materials. This observation is consistent with previous findings on homopolar Ge–Ge bonds in GeSe-based OTS systems.^51,55,56 The real-space projection of defect states also shows that increasing the In content reduces defect states associated with over-coordinated Te atoms in the amorphous network, which in turn helps suppress the leakage current in OTS devices.

To further investigate how the amorphous structure evolves during relaxation, we performed AIMD simulations to track the structural changes of a-In₃Te₇ over 60 ps. The aging simulation was carried out at 600 K to accelerate the relaxation process, and the time-dependent evolution of Te–Te homopolar bonds and In–Te heteropolar bonds is shown in Fig. 5f. During structural aging, unstable Te–Te bonds are progressively replaced by more stable In–Te bonds, leading to changes in the defect states within the bandgap and consequently affecting the V_th. Combined with previous results showing that higher In content significantly increases the proportion of In–Te bonds, these findings explain why In₃Te₇-based OTS devices exhibits the slowest aging rate and the lowest V_th drift among InTe₄, InTe₃, and In₃Te₇.

3. Conclusions

In summary, we systematically explored the origin of V_th drift in In–Te OTS devices and demonstrated an effective strategy to suppress it by reducing homopolar bonding in the amorphous network. By tuning the In/Te ratio, we fabricated OTS devices based on InTe₄, InTe₃, and In₃Te₇ and revealed that higher In content leads to a larger optical bandgap, lower leakage current, and smaller V_th drift coefficient. The In₃Te₇ devices exhibit a high on/off ratio approaching 10⁵ and a low V_th drift coefficient of 26 mV dec⁻¹. Structural analyses and AIMD simulations show that increasing the In concentration markedly reduces the fraction of unstable Te–Te homopolar bonds and over-coordinated Te atoms while promoting the formation of stronger, more stable In–Te heteropolar bonds. This structural stabilization results in lower atomic mobility, fewer deep defect states, and slower structural relaxation, thereby improving both the reliability and switching stability of OTS devices. In addition, we found that the defect states are closely associated with Te–Te bonds, and that aging indeed leads to the breaking of Te–Te bonds accompanied by the formation of In–Te bonds. Overall, our findings clarify the atomic mechanism of V_th drift in Te-based OTS devices and establish composition engineering as a powerful route to achieving high-performance, drift-resistant OTS selectors for next-generation 3D PCM applications.

4. Experimental

4.1 Device fabrication and electrical measurements

In–Te OTS devices with a via-hole structure were prepared on SiO₂/Si (100) substrates. First, bottom electrodes of 10 nm Ti and 100 nm W were deposited sequentially. A 100 nm SiO₂ isolation layer was then grown using plasma-enhanced chemical vapor deposition (PECVD). Via-holes with diameters of 200, 400, 600, and 800 nm and a depth of 100 nm were patterned by electron-beam lithography (EBL) followed by inductively coupled plasma (ICP) etching. The OTS layer and top electrode patterns were defined using UV lithography. Subsequently, the OTS layer and a 100 nm W top electrode were deposited consecutively by DC magnetron sputtering. The final lift-off process created multiple isolated device units. The composition of the OTS material layer was controlled by adjusting the co-sputtering power of the InTe₉ and In targets. Electrical characterization, including DC I-V and pulse-switching measurements, was performed using an Agilent B1530A semiconductor analyzer.

4.2 Ab initio simulations

Ab initio simulations were performed using the Vienna Ab initio Simulation Package (VASP).⁵⁷ The calculations employed the projector augmented-wave (PAW) method and Perdew–Burke–Ernzerhof generalized gradient approximation (GGA-PBE) exchange functional.⁵⁸ Amorphous structures were generated through a melt-quench-relaxation procedure based on AIMD within DFT.⁵⁹ A time step of 3 fs was used, and the cutoff energy during AIMD was set to 300 eV. The Γ-point was used for Brillouin-zone sampling, and the simulations were carried out in the canonical (NVT) ensemble with a Nosé–Hoover thermostat. The starting model consisted of a cubic supercell containing 300 atoms, where In and Te atoms were randomly placed according to the experimental density. Each model was melted at 3000 K for 18 ps to remove memory effects, followed by rapid cooling to 300 K at a rate of 30 K ps⁻¹. The cell volume was subsequently adjusted to release residual pressure. The a-InTe₄, a-InTe₃, and a-In₃Te₇ structures were equilibrated for an additional 12 ps to obtain atomic trajectories. Afterward, all amorphous models were fully relaxed at 0 K with a force tolerance of 0.02 eV for electronic-structure calculations. For the electronic structure calculations, the energy cutoff was set to 500 eV, and a 2 × 2 × 2 k-point grid was used for Brillouin-zone sampling to achieve higher accuracy.

Conflicts of interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Data availability

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Supplementary information is available. See DOI: https://doi.org/10.1039/d5tc04468e.

Acknowledgements

This work was supported by the National Science and Technology Major Project of China (Grant No. 2022ZD0117600), the Regional Joint Fund of National Natural Science Foundation of China (No. U24A20303), the Hubei Key Laboratory of Advanced Memories, and the Fundamental Research Funds for the Central Universities.

References

1. S. Salahuddin, K. Ni and S. Datta, Nat. Electron., 2018, 1, 442–450.
2. W. Zuo, Q. Zhu, Y. Fu, Y. Zhang, T. Wan, Y. Li, M. Xu and X. Miao, J. Semicond., 2023, 44, 053102.
3. G. W. Burr, R. S. Shenoy, K. Virwani, P. Narayanan, A. Padilla, B. Kurdi and H. Hwang, J. Vac. Sci. Technol., B:Nanotechnol. Microelectron.:Mater., Process., Meas., Phenom., 2014, 32, 040802.
4. T. Endoh, H. Koike, S. Ikeda, T. Hanyu and H. Ohno, IEEE J. Emerg. Sel. Top. Circuits Syst., 2016, 6, 109–119.
5. J. Shen, W. Song, K. Ren, Z. Song, P. Zhou and M. Zhu, Adv. Mater., 2023, 35, 2208065.
6. M. Xu, S. Wang, Y. He, Y. Li, W. Zhang, M. Yang, X. Qi, Z. Wang, M. Xu, D. Shang, Q. Liu, X. Miao and M. Liu, Nat. Comput. Sci., 2025, 5, 1178–1191.
7. M. Xu, X. Mai, J. Lin, W. Zhang, Y. Li, Y. He, H. Tong, X. Hou, P. Zhou and X. Miao, Adv. Funct. Mater., 2020, 30, 2003419.
8. K. Ding, J. Wang, Y. Zhou, H. Tian, L. Lu, R. Mazzarello, C. Jia, W. Zhang, F. Rao and E. Ma, Science, 2019, 366, 210–215.
9. Q. Xu, M. Xu, S. Tang, S. Yuan, M. Xu, W. Zhang, X.-B. Li, Z. Wang, X. Miao, C. Wang and M. Wuttig, InfoMat, 2025, 7, e70006.
10. M. Zhu, K. Ren and Z. Song, MRS Bull., 2019, 44, 715–720.
11. S. Jia, H. Li, T. Gotoh, C. Longeaud, B. Zhang, J. Lyu, S. Lv, M. Zhu, Z. Song, Q. Liu, J. Robertson and M. Liu, Nat. Commun., 2020, 11, 4636.
12. D. Garbin, W. Devulder, R. Degraeve, G. L. Donadio, S. Clima, K. Opsomer, A. Fantini, D. Cellier, W. G. Kim, M. Pakala, A. Cockburn, C. Detavernier, R. Delhougne, L. Goux and G. S. Kar, IEEE Int. Electron Devices Meet., 2019, 35.31.31–35.31.34.
13. A. Verdy, M. Bernard, N. Castellani, P. Noé, J. Garrione, G. Bourgeois, M. C. Cyrille, G. Navarro and E. Nowak, IEEE 11th International Memory Workshop (IMW), 2019, 1–4.
14. J. Shen, S. Jia, N. Shi, Q. Ge, T. Gotoh, S. Lv, Q. Liu, R. Dronskowski, S. R. Elliott, Z. Song and M. Zhu, Science, 2021, 374, 1390–1394.
15. S. Ban, J. Lee, Y. Seo, W. Lee, T. Kim and H. Hwang, Adv. Electron. Mater., 2024, 11, 2400665.
16. Z. Zhao, S. Clima, D. Garbin, R. Degraeve, G. Pourtois, Z. Song and M. Zhu, Nanomicro Lett., 2024, 16, 81.
17. R. Wu, R. Gu, T. Gotoh, Z. Zhao, Y. Sun, S. Jia, X. Miao, S. R. Elliott, M. Zhu, M. Xu and Z. Song, Nat. Commun., 2023, 14, 6095.
18. Y. Koo, S. Lee, S. Park, M. Yang and H. Hwang, IEEE Electron Device Lett., 2017, 38, 568–571.
19. S. Ban, S. Lee, J. Lee and H. Hwang, IEEE Electron Device Lett., 2022, 43, 643–646.
20. R. Wu, Y. Sun, S. Zhang, Z. Zhao and Z. Song, Nanomaterials, 2023, 13, 1114.
21. Y. Koo and H. Hwang, Sci. Rep., 2018, 8, 11822.
22. Y. Saito, S. Hatayama, W. H. Chang, N. Okada, T. Irisawa, F. Uesugi, M. Takeguchi, Y. Sutou and P. Fons, Mater. Horiz., 2023, 10, 2254–2261.
23. Z. Zhao, M. Zhang, Q. Yang, T. Gotoh, Q. Ge, N. Shi, Y. Sun, J. Zhao, Y. Sui, R. Jiang, H. Yu, S. R. Elliott, Z. Song and M. Zhu, Adv. Funct. Mater., 2025, 35, 2570135.
24. S. Ban, J. Lee, T. Kim and H. Hwang, IEEE Symposium on VLSI Technology and Circuits, 2023, pp. 1–2.
25. W. Zhang and E. Ma, Mater. Today, 2020, 41, 156–176.
26. J. Yoo, I. Karpov, S. Lee, J. Jung, H. S. Kim and H. Hwang, IEEE Electron Device Lett., 2020, 41, 191–194.
27. Z. Chai, W. Zhang, R. Degraeve, S. Clima, F. Hatem, J. F. Zhang, P. Freitas, J. Marsland, A. Fantini, D. Garbin, L. Goux and G. S. Kar, IEEE Symposium on VLSI Technology, 2019, pp. T238–T239.
28. J. Y. Raty, W. Zhang, J. Luckas, C. Chen, R. Mazzarello, C. Bichara and M. Wuttig, Nat. Commun., 2015, 6, 7467.
29. H. Y. Cheng, W. C. Chien, I. T. Kuo, E. K. Lai, Y. Zhu, J. L. Jordan-Sweet, A. Ray, F. Carta, F. M. Lee, P. H. Tseng, M. H. Lee, Y. Y. Lin, W. Kim, R. Bruce, C. W. Yeh, C. H. Yang, M. BrightSky and H. L. Lung, IEEE Int. Electron Devices Meet., 2017, 2.2.1–2.2.4.
30. J. Lee, S. Kim, S. Lee, S. Ban, S. Heo, D. Lee, O. Mosendz and H. Hwang, IEEE Symposium on VLSI Technology and Circuits, 2022, pp. 320–321.
31. S. Clima, D. Garbin, W. Devulder, J. Keukelier, K. Opsomer, L. Goux, G. S. Kar and G. Pourtois, Microelectron. Eng., 2019, 215, 110996.
32. X. Li, Z. Yuan, Y. Xue, Y. Zhu, S. Song and Z. Song, Appl. Electron. Mater., 2023, 5, 3499–3506.
33. M. Choi, H.-J. Sung, B. Koo, J.-B. Park, W. Yang, Y. Kang, Y. Park, Y. Ham, D.-J. Yun, D. Ahn, K. Yang and C. S. Lee, Adv. Sci., 2024, 11, 2404035.
34. S. Lee, J. Lee, M. Kwak, O. Mosendz and H. Hwang, Appl. Phys. Lett., 2021, 118, 212103.
35. A. Velea, K. Opsomer, W. Devulder, J. Dumortier, J. Fan, C. Detavernier, M. Jurczak and B. Govoreanu, Sci. Rep., 2017, 7, 8103.
36. Z. Hu, W. Zhang, R. Degraeve, D. Garbin, Z. Chai, N. Saxena, P. Freitas, A. Fantini, T. Ravsher, S. Clima, J. F. Zhang, R. Delhougne, L. Goux and G. Kar, IEEE Trans. Electron Devices, 2023, 70, 812–818.
37. Z. Yang, B. Li, J. J. Wang, X. D. Wang, M. Xu, H. Tong, X. Cheng, L. Lu, C. Jia, M. Xu, X. Miao, W. Zhang and E. Ma, Adv. Sci., 2022, 9, 2103478.
38. M. Zhu, W. Song, P. M. Konze, T. Li, B. Gault, X. Chen, J. Shen, S. Lv, Z. Song, M. Wuttig and R. Dronskowski, Nat. Commun., 2019, 10, 3525.
39. X. Zhou, Z. Hu, Z. Chai, W. Zhang, S. Clima, R. Degraeve, J. F. Zhang, A. Fantini, D. Garbin, R. Delhougne, L. Goux and G. S. Kar, IEEE Electron Device Lett., 2022, 43, 1061–1064.
40. N. Ciocchini, M. Cassinerio, D. Fugazza and D. Ielmini, IEEE Trans. Electron Devices, 2012, 59, 3084–3090.
41. J. Wen, C. Yi, J. Chen, L. Wang, Z. Liu, Z. Chen, H. Tong and X. Miao, IEEE Electron Device Lett., 2025, 46, 115–118.
42. J. B. Coulter and D. P. Birnie Iii, Phys. Status Solidi B, 2018, 255, 1700393.
43. M. Lee, S. Lee, M. Kim, J. Lee, C. Kwon, C. Won, T. Kim, S. Lee, S. Cho, S. Na, S. Park, K. Yoon, H. Kim and T. Lee, J. Alloys Compd., 2024, 973, 172863.
44. S. Clima, D. Matsubayashi, T. Ravsher, D. Garbin, R. Delhougne, G. S. Kar and G. Pourtois, Solid-State Electron., 2024, 214, 108876.
45. X. Wang, Y. Xu, H. Zhu, R. Liu, H. Wang and Q. J. C. Li, CrystEngComm, 2011, 13, 2955–2959.
46. M. Emziane, J. C. Bernède, J. Ouerfelli, H. Essaidi and A. Barreau, Mater. Chem. Phys., 1999, 61, 229–236.
47. J. Wu, Z. Shao, B. Zheng, Y. Zhang, X. Yao, K. Huang and S. Feng, Nanoscale Adv., 2023, 5, 2418–2421.
48. C. Liu, Y. Yuan, X. Zhang, J. Su, X. Song, H. Ling, Y. Liao, H. Zhang, Y. Zheng and J. Li, Nanomaterials, 2020, 10, 1887.
49. S. Hatayama, Y. Saito, P. Fons, Y. Shuang, M. Kim and Y. Sutou, Acta Mater., 2023, 258, 119209.
50. J. Lee, S. Ban, T. H. Lee and H. Hwang, IEEE Electron Device Lett., 2023, 44, 1468–1471.
51. S. Clima, D. Garbin, K. Opsomer, N. S. Avasarala, W. Devulder, I. Shlyakhov, J. Keukelier, G. L. Donadio, T. Witters, S. Kundu, B. Govoreanu, L. Goux, C. Detavernier, V. Afanas'ev, G. S. Kar and G. Pourtois, Phys. Status Solidi RRL, 2020, 14, 1900672.
52. D. Ielmini and Y. Zhang, J. Appl. Phys., 2007, 102, 054517.
53. P. Fantini, N. Polino, A. Ghetti and D. Ielmini, Adv. Electron. Mater., 2023, 9, 2300037.
54. S. Clima, T. Ravsher, D. Garbin, R. Degraeve, A. Fantini, R. Delhougne, G. S. Kar and G. Pourtois, ACS Appl. Electron. Mater., 2023, 5, 461–469.
55. A. Slassi, L. S. Medondjio, A. Padovani, F. Tavanti, X. He, S. Clima, D. Garbin, B. Kaczer, L. Larcher, P. Ordejón and A. Calzolari, Adv. Electron. Mater., 2023, 9.
56. M. Moon, S. Hwang, J. Kim, Y. Park, C. Hong and S. Han, ACS Appl. Mater. Interfaces, 2025, 17, 57260–57272.
57. G. Kresse and J. Furthmüller, Phys. Rev. B: Condens. Matter Mater. Phys., 1996, 54, 11169–11186.
58. J. P. Perdew, K. Burke and M. Ernzerhof, Phys. Rev. Lett., 1996, 77, 3865–3868.
59. J. Hafner, J. Comput. Chem., 2008, 29, 2044–2078.

---