Zn1−xTex Ovonic Threshold Switching Device Performance and its Correlation to Material Parameters

We have experimentally demonstrated a strong correlation between the electrical properties of Zn1−xTex Ovonic threshold switching (OTS) selector device and the material properties analysed by X-ray diffraction (XRD), spectroscopic ellipsometry, and X-ray photoelectron spectroscopy (XPS). The correlation and the key material parameters determining the device performances were investigated. By comparing the experimental data with the calculation results from various analytical models previously developed for OTS materials, the electrical properties of the device were shown to be dependent on the key material parameters; the concentration of sub-gap trap states and the bandgap energy of the OTS material. This study also experimentally demonstrated that those key parameters have determined the device performance as expected from the analytical model. The origin of the OTS phenomenon and conduction mechanism were explained both experimentally and theoretically. This leads to better understanding of the conduction mechanism of OTS devices, and an insight for process improvement to optimize device performance for selector application.

on state with low resistance; thus current through the device in the on state (I on ) increased considerably. The volatile on state is maintained as long as high voltage is supplied. The optimized ZnTe device showed high selectivity (the ratio of I on at V th to I off at 1/2 V th ) of 10 5 , which is exceptionally high compared to various types of selector devices [5][6][7][8] , and thus promising for selector device application in high-density X-point memory arrays.
Experimental observations of ZnTe with compositional change. Figures 2 and 3 show experimental results of various material properties of binary OTS material Zn 1−x Te x according to composition. X-ray diffraction (XRD) analysis results are shown in Fig. 2a. XRD peaks representing crystalline phase were only detected in highly Zn-rich samples (0 < x < 0.5), and all the crystalline phases detected were Zn crystals. This shows that in highly Zn-rich samples, the excessive Zn atoms are present in the form of Zn crystalline clusters. In other composition ranges, no XRD peak was detected, indicating that Zn 1−x Te x is amorphous, except for the highly Zn-rich composition. The spectroscopic ellipsometry results show the bandgap energy (E g ) of the material as a function of composition (Fig. 2b). The E g of pure Te was measured to be 0.6 eV. As the Zn content increases, E g increases gradually, reaching the maximum value of 1.6 eV at Zn 0.35 Te 0.65 and decreases again as the Zn content increases further. Figure 3c,d show X-ray photoelectron spectroscopy (XPS) spectra of pure Te, pure Zn, and Zn 0.35 Te 0.65 . Pure Te sample showed Te 3d 5/2 peak at 572.85 eV, which is well known as the bulk Te peak. Similarly, pure Zn sample also showed Zn 2p 3/2 peak at 1021.8 eV, which is well known as the bulk Zn peak. Binding energy shift measured in the XPS analysis indicates how the electronic states of elements in Zn 0.35 Te 0.65 exist. All Zn 2p 3/2 spectra detected from Zn 0.35 Te 0.65 were Zn 2+ peak at 1022.43 eV, while Te 3d 5/2 spectra were sum of Te bulk peak at 572.85 eV and Te* peak at 573.18 eV. In other words, all the Zn atoms in the material are present in Zn 2+ form having chemical bonding to Te atoms, while large proportion of Te remains as bulk Te and does not have a chemical bond with Zn, thus providing a large number of lone-pairs in the material. We could not find any reference for Te* peak, but as Te is known to have lots of oxidation states such as 6, 5, 4, 3, 2, 1, −1, −2, it is not strange to see the peak at 573.18 eV. Figure 3 shows the electrical response of the Zn 1−x Te x devices and corresponding selector performances according to the material composition. The Zn 1−x Te x devices showed OTS behaviour in Te-rich composition (0.5 < x ≤ 1.0). Mixing Te with an appropriate amount of Zn (x ~ 0.65) dramatically improved the selector performance by reducing I off , but excessive Zn content (x > 0.65) caused deterioration of device performance and even metallic failure. The Zn-rich devices (0 < x < 0.5) exhibited high off state resistance in the low voltage region in the initial state, but became permanently metallic after external bias (~1 V) was applied. The electrical characteristics are closely related to those material properties shown above. The OTS behaviour observed from pure Te (Fig. 3a) and the large amount of unbound Te lone-pairs detected in the XPS spectra of the Te-rich Zn 1−x Te x device (Fig. 2c) show the correlation of Te atoms and OTS phenomenon. This is consistent with the previous theoretical studies suggesting that the sub-gap trap states originating from the lone-pairs of Te atoms trigger the OTS phenomenon [14][15][16][17][18][19][20] . The off state resistance of the OTS behaviour shown in the devices is strongly correlated with the E g measured in the spectroscopic ellipsometry analysis. The off state resistance increases as E g increases as a function of the composition change until it reaches the maximum value at Zn 0.35 Te 0.65 , and decreases again as E g decreases with further compositional change. Pure Te alone cannot be used as selector application due to its high leakage current, while too much Zn may be a threat to the device. However, material composition optimization Analytical modelling and performance-determining parameters. Analytical modelling on the OTS phenomenon has been widely developed in order to predict the electrical characteristics of the OTS devices. According to the previous literatures, candidate explanations for the conduction mechanism include Poole-Frenkel emission, Schottky emission, space-charge limited currents, optimum channel hopping, field-induced delocalization of tail states, percolation band conduction, transport through crystalline inclusions, and thermally assisted hopping 14,19,20 . The expressions for each model from previous literatures are summarized in Table 1. Figure 4 shows the prediction of each model and their respective errors compared to the experimental data of Zn 0.35 Te 0.65 OTS device. Among those analytical models, thermally assisted hopping (TAH) model provided remarkably small error compared to the others. According to TAH model, the conduction in the OTS device could be understood by thermally assisted hopping conduction from localized traps to extended states, following eq. (1) 14,19,20 : where q is the elementary charge, A is the current path area, N T is the density of deep traps responsible for the off state conduction, ∆z is the average distance between deep traps, τ 0 is the attempt-to-escape time from a trapping site, (E C − E F ) is the energy barrier (E a ) between the conduction band mobility edgy E C and the quasi-Fermi level E F , and d is the thickness of the material. With the assumption of τ 0 = 10 −15 s, the results of the analytical model well explain the mechanism of the current inhibition of the material at low voltage as shown in Figs 4 and 5.
According to the results, E a and ∆z determines the off state conduction of the material. The magnitude of I off is determined by E a , showing the magnitude of I off reduced by 10 times for every 0.05 eV increase in E a (Fig. 5a). Therefore, higher E a is preferable for efficient leakage current inhibition. The dependency of I off to the external electric field (in other words, the slope of the I-V curve) is determined by ∆z (Fig. 5b). Larger ∆z causes a steeper slope, and consequently higher I off at 1/2 V th ; while too small ∆z (<3 nm) causes higher I off in the lower voltage region. Hence, a moderate value of ∆z (3 nm < ∆z < 7 nm) is preferable for selector application. E a and ∆z according to the composition are shown in Fig. 5c and d. Correlation between the energy barrier E a extracted from the analytical model and the band gap energy E g measured by spectroscopic ellipsometry (Fig. 2b) shows the validity of the model (Fig. 5c). Both E a and E g have a maximum at a specific composition of x = 0.65, and gradually decrease when the composition reaches either end. ∆z showed similar dependency on the material composition and reached a maximum at x = 0.65 (Fig. 5d). The performance-determining parameters, E a and ∆z, well explain the dependency of electrical properties on the material composition. In pure Te (x = 1), even though OTS phenomena occur, E a and ∆z are both small, such that leakage current cannot be suppressed efficiently. Thus, pure Te is not sufficient to use as a selector device due to its high leakage current. As mixing Zn into Te, E a and ∆z increases with Zn content until reaching their maximum at x = 0.65 composition. The leakage current was therefore minimized at x = 0.65, providing the best performance of the material for selector application. As the  Table 1. Various analytical models and corresponding expressions on conduction of OTS materials from previous literatures. The OTS phenomenon has been widely studied theoretically; however, its conduction mechanism and corresponding analytical model is still a highly controversial topic. Zn content further increases excessively (0 < x < 0.5), E a and ∆z decrease and thus the leakage current at low bias (<1 V) increases. Material composition optimization (Zn 0.35 Te 0.65 ) makes the material suitable for selector application by modifying material parameters to suppress leakage current effectively. In other words, the key material parameters E a and ∆z extracted from the model well explains how composition can maximize the selector device performances, by correlating the device performance to the material properties.

Discussion
In summary, we have investigated the material characteristics and the device performance of Zn 1−x Te x binary OTS devices according to compositional change, using optical and electrical analysis methods. By comparing various analytical models, the electrical characteristics were best explained by an analytical model based on thermally assisted hopping conduction 14,19,20 , showing good agreement with the measured data. Consequently, the performance-determining material parameters were successfully extracted. Those parameters well explained the strong correlation between the material properties and the electrical device performance of Zn 1−x Te x in wide range of composition. The origin of OTS phenomenon and the mechanism of its conduction was explained experimentally and theoretically. Finally, the selector performance of the OTS device was successfully improved by optimizing the key parameters; the concentration of the sub-gap trap states and the bandgap energy of the material. This study provides better understanding of the sub-threshold conduction mechanism of the OTS device and gives clue for process optimization to maximize device performance of OTS selector devices.

Methods
Zn 1−x Te x films of various compositions, including pure zinc (x = 0) and pure tellurium (x = 1), were deposited at room temperature using an RF magnetron sputtering system, in order to analyse their material characteristics and electrical properties. The material composition was measured using energy dispersive spectroscopy (EDS) analysis. For material property characterization, ZnTe films were deposited on flat wafers and were analysed by XRD, XPS and spectroscopic ellipsometry. For electrical property characterization, selector devices were fabricated with a W/ZnTe/W structure with 10 nm thick ZnTe layer. Tungsten was chosen as the electrode material, because of its advantages such as high electrical conductivity, high stability, and back-end-of-line (BEOL) compatibility. Device size was controlled by the size of W plug bottom electrode isolated by electrically insulating oxide sidewalls. The I-V response of the devices was measured using the following two measurement methods with a semiconductor device analyser using the circuit scheme shown in the inset of Fig. 1. Most of the electrical characteristics including the high resistance off state was measured in a longer time scale (>100 ms) with current compliance to prevent device damage due to the high current flow through the low on state resistance of the device. However, the on state conduction (shown in Fig. 1) was measured in a shorter time scale (<1 ms) in order to extract the intrinsic I on of the device.