Exponential growth of the artificial intelligence computing industry triggered by the ‘ChatGPT moment’ has accelerated the necessity of not only high memory capacity for big data but also fast computing speed. The conventional Von Neumann architecture has bottlenecks on processing speeds between classes of memory hierarchy and is increasingly challenged by the demands for higher processing speed and energy efficiency. Interest in alternative computing architectures such as in-memory computing and neuromorphic computing to reduce the bottleneck with parallel processing and physical proximities has increased. To realize these emerging architectures, a cross-point array is promising due to its ability to achieve a high-density 4F2 cell architecture and its potential for 3D vertical stacking [1,2]. However, leakage current occurs through unselected cells in the cross-point array, a phenomenon known as a sneak path issue, significantly affecting the reliability and performance of memory devices [1–3].

Selector devices have been introduced to suppress the undesired current while ensuring a sufficient current density on selected cells due to the volatile resistive switching with high selectivity [3,4]. Among the various types of selector devices, ovonic threshold switching (OTS) selectors have emerged as a promising solution due to their unique advantages such as a high on/off ratio, bidirectional operation, fast response time, and a simple structure [3,4]. OTS materials are typically amorphous chalcogenides and require an initial activation process known as the first-fire (ff) process, which initiates their switching behavior [5].

Although various mechanisms have been proposed to explain the OTS behavior, no single model reported to date has comprehensively accounted for all observed phenomena of the chalcogenides [6–9]. However, it is well established that conduction in the subthreshold region below the threshold voltage (V_th) can be effectively described by a model suggested by D. Ielmini based on a Poole-Frenkel (PF) mechanism [3,10–16]. According to this model, the conduction mechanism is dominated by trap-assisted hopping where electrons move between localized trap states within the amorphous material under an electric field barrier lowering effect [10,11].

In this study, the subthreshold conduction of OTS devices under different temperatures is investigated to induce the activation energy of the material based on this model. Amorphous silicon telluride (SiTe) is chosen as a strategic base material and nitrogen as a dopant to enhance the selector performance for the experimental and theoretical investigations. We previously reported the effect of nitrogen doping on selector performance and analyzed it in terms of trap states and electron localization [17]. In the present study, the temperature dependency on the subthreshold conduction of the selectors is reported and explained by the aforementioned model in terms of activation energy (E_A), inter-trap distance (Δz), and effective trap concentration (N_T,tot) to provide a comprehensive understanding of the materials.

Amorphous chalcogenide films with a thickness of 20 nm were deposited between tungsten (W) top and bottom electrodes with a thickness of 100 nm, as shown in a cross-sectional transmission electron microscope image in our previous report [17]. The SiTe films were deposited using radio frequency (RF) co-sputtering from Si and Te targets, and the nitrogen doped SiTe (N-SiTe) films were deposited via a reactive co-sputtering process under N2 and Ar gas flow. The N doping concentration of the N-SiTe film was 25 %, with a similar Si:Te composition of 44:56 compared to the SiTe film stoichiometry of 47:53 [17]. The compositions of the films were determined on the basis of peaks of N 1s, Si 2p, and Te 3d orbitals measured by X-ray photoelectron spectroscopy (XPS) (Thermo Fisher Scientific, K-Alpha) [17]. A working pressure of 10 mTorr was set for the OTS film deposition under a base pressure of < 5 × 10−6 Torr for uniform layer deposition. Direct current (DC) sputtering under 3 mTorr working pressure for the W top electrodes was followed by RF sputtering in a vacuum environment to prevent oxidation of the chalcogenides. All sputtering processes were conducted under room temperature.

Electrical properties were evaluated with DC voltage sweeps using a semiconductor characterization system (Keithley 4200-SCS) without an external resistor. During the DC voltage sweeps, the compliance current was set to 1 mA to avoid degradation of the devices to solely evaluate the temperature dependency. The temperatures of the selector devices were controlled using a hot-chuck controller with a heating plate (MS Tech, MST-1000H), and the DC I-V curves were measured under stabilized temperature after heater overshoot, as shown in Fig. 1(a).

Figure 1. (a) Photographs of heating plate and hot-chuck controller with semiconductor characterization system. (b) Schematic illustration of OTS selector device structure. Deposition methods for each layer are denoted with corresponding color in the schematics. Effective size and thickness of the devices are also represented to calculate the area of contact (A) and the chalcogenide thickness (u_a).

Figure 1(b) shows schematic illustrations of the fabricated OTS selector devices. The devices were fabricated as a T-shaped type with an effective device size of 3 μm. The OTS layer was deposited between W electrode layers, and the effective area became a square hole region of the silicon oxide (SiO_x) layer. The W layers were deposited with DC sputtering, the SiO_x layer with plasma enhanced chemical vapor deposition (PE-CVD), and the holes were etched with an inductively coupled plasma reactive ion etching process, as represented in Fig. 1(b).

Figure 2 shows typical DC I−V curves of the fabricated OTS selector devices. Pristine OTS selectors are known to activate after an initial ‘ff ’ process, which typically occurs at a voltage higher than the V_th, as represented by the black curves in Figs. 2(a) and 2(b) showing the ff voltages (V_ff). The ff process is a phenomenon that generates a metastable conductive filament formed by activated defects [18,19]. Due to the activated defects or trap states, subthreshold currents were increased and threshold switching occurred at lower voltages (V_th < V_ff) after the first-firing process. Since ideal OTS selectors should exhibit bidirectional operation, both positive and negative voltage sweep I−V curves were measured. As the voltage increases during the sweep, the current through the OTS materials sharply rises at a specific voltage, known as the V_th, leading to a low resistance state (LRS). The N-SiTe selector device required higher energy of V_th = 0.99 V to realize a LRS compared to the value of 0.89 V for SiTe. Conversely, the current rapidly decreases at a hold voltage (V_h) during measurement of the voltage decrease, returning the device to a high resistance OFF state (HRS). The off-current (I_off) is defined as the current value at half the threshold voltage (1/2 V_th), considering the operation of the OTS selector in a cell architecture. The leakage current could be suppressed to 1.3 μA by nitrogen doping on SiTe with a value of 29 μA. In this study, a compliance current (I_cc) of 1 mA was applied during DC I−V measurements to prevent potential damage to the selector devices, and thus the selectivity (I_on/I_off) of N-SiTe was calculated as 770 and that of SiTe as 34.

Figure 2. Bidirectional DC I−V curves of (a) SiTe and (b) N-SiTe OTS selector devices with full voltage ranges. Both I−V curves were measured at 298 K under 1 mA compliance current (I_cc).

In Figs. 3(a) and 3(b), I−V curves at the subthreshold regime of the SiTe and N-SiTe OTS selector devices at temperatures of 298, 318, 338, and 358 K are shown. In the subthreshold region, the off-current is known to be governed by a model based on the PF conduction model, as delineated in Eq. (1) [10].

Figure 3. Off-current measurement and subthreshold I−V curves at 298, 318, 338, and 358 K for (a) SiTe and (b) N-SiTe. (c) Arrhenius plot of SiTe and N-SiTe at different voltages. (d) Calculation of activation energy at applied voltages derived from the Arrhenius plot. (e) Inter-trap distances and total trap densities of SiTe and N-SiTe (f) Schematics of trap states for SiTe and N-SiTe with and without bias.

In Eq. (1), q is the elementary charge, A is the area of contact, N_T,tot is the effective trap concentration, Δz is an inter-trap distance, τ0 is the attempt-to-escape time for trapped electrons, E_C is the conduction band level, E_F is the Fermi level, k is the Boltzmann constant, T is temperature, V_A is applied voltage, and u_a is the amorphous chalcogenide thickness [10]. To analyze the effect of nitrogen doping on amorphous SiTe in Fig. 2, parameters of Δz, N_T,tot, and E_A (E_C − E_F ) should be derived.

Figure 3(c) shows Arrhenius plots of the natural logarithmic current and 1/kT derived from Fig. 3(a) for SiTe and Fig. 3(b) for N-SiTe selector devices at voltages of 0.2, 0.3, 0.4, and 0.5 V in the region of exponentially increasing current and at measured temperatures of 298, 318, 338, and 358 K. E_A (activation energies at certain voltages) values of SiTe and N-SiTe are given by the slopes of the Arrhenius plots with Eq. (2) [10].

Figure 3(d) shows the slopes derived from all curves of Fig. 3(c) with voltages of 0.2, 0.3, 0.4, and 0.5 V, and the slopes were determined by a linear fitting method to extract E_A of SiTe and N-SiTe. Each point on Fig. 3(d) corresponds to the linearly fitted slopes of Fig. 3(c) of SiTe and N-SiTe. It is noteworthy that the E_A values decrease with increased voltages, indicating that electrons can easily hop through nearby traps due to a lowered energy barrier leading to increased I_off. Activation energies of the chalcogenide materials E_A (E_C − E_F ) are given by the y-intercept of the linear fitted E_A values of various voltages from Eq. (2), which equate with the activation energies of the SiTe and N-SiTe without bias (V_A = 0 V). As shown in the Fig. 3(d), the E_A (E_C − E_F ) values of SiTe and N-SiTe were derived as 0.15 and 0.33 eV, respectively. The trap energy barrier was increased two-fold by N doping on SiTe, which correlate well with the observation of lowered I_off enhancing the selector device performance in Fig. 2.

Δz can be calculated from the subthreshold slope (STS) of the exponentially increasing current region with the relation given by Eq. (3) [10].

From Eq. (3), Δz values of SiTe and N-SiTe selectors were calculated as 1.98 and 2.69 nm, respectively, using linear fitted STS values from the 0.2 to 0.5 V region, u_a = 20 nm, and T = 298 K. N_T,tot was estimated as 1.286×10²⁰ traps/cm³ for SiTe and 5.112 × 10¹⁹ traps/cm³ for N-SiTe under an assumption of N_T,tot ≈ 1/(Δz)³ [11,12,15,16], as shown in Fig. 3(e). Increased inter-trap distance and decreased trap density by nitrogen doping also explain the lowered I_off of the N-SiTe devices, since e− hopping among traps was hampered. Figure 3(f) shows the schematic illustrations of the energy band of traps for SiTe and N-SiTe with voltage that explain subthreshold conduction of the OTS selectors. As illustrated in the energy band diagram, more bias could lower the activation energy due to a field-induced energy barrier lowering effect that increase I_off. Trap parameters changed by N doping, which increased Δz, decreased N_T,tot, and increased E_A, not only explain the effectively suppressed I_off but also account for the increased V_th of N-SiTe compared to SiTe OTS selector devices. These results are consistent with our previous results of spectroscopically measured trap states and optical bandgap energies [17]. The SiTe is affected by N doping via decreased trap states (N_T,tot, XPS, and Raman spectroscopy), increased bandgap energies [E_A (E_C − E_F ) and Tauc plot], and localized e− states (first-principles simulation), which result in suppression of I_off and thereby enhanced selector device performance [17].

In this study, the subthreshold region of SiTe based OTS selector devices is investigated with the aim of suppressing leakage current. By introducing nitrogen to amorphous SiTe, the I_off was effectively lowered, leading to enhanced selectivity. We analyzed the effect of nitrogen doping with DC I−V curves at various temperatures to derive the trap state parameters. Since the subthreshold current is governed by electron hopping through trap states, increased inter-trap distance, lowered trap density, and raised activation energy by N doping explain the observed I_off well.

This research was supported by Samsung Electronics Co., Ltd. under grant IO2102021-08356-01 and was also supported by the BK21 FOUR (Fostering Outstanding Universities for Research) funded by the Ministry of Education (MOE) of Korea and National Research Foundation (NRF) of Korea.

The authors declare no conflicts of interest.