Scandium doping brings speed improvement in Sb2Te alloy for phase change random access memory application

Phase change random access memory (PCRAM) has gained much attention as a candidate for nonvolatile memory application. To develop PCRAM materials with better properties, especially to draw closer to dynamic random access memory (DRAM), the key challenge is to research new high-speed phase change materials. Here, Scandium (Sc) has been found it is helpful to get high-speed and good stability after doping in Sb2Te alloy. Sc0.1Sb2Te based PCRAM cell can achieve reversible switching by applying even 6 ns voltage pulse experimentally. And, Sc doping not only promotes amorphous stability but also improves the endurance ability comparing with pure Sb2Te alloy. Moreover, according to DFT calculations, strong Sc-Te bonds lead to the rigidity of Sc centered octahedrons, which may act as crystallization precursors in recrystallization process to boost the set speed.

dopants match well with parent Sb 2 Te 3 structure and increase the stability of Sb 2 Te 3 in the phase change process 20 . Furthermore, a recent report from Science points out that Sc-doped Sb 2 Te 3 phase change material without phase separation has a very rapid SET speed reaching up to 700 picoseconds 21 Among the equilibrium phases of Sb-Te system, Sb 2 Te alloy has >50 °C higher crystallization temperature than Sb 2 Te 3 one. Hence, in this paper, Sc element was also been chosen as a dopant into Sb 2 Te alloy. We hope that through the adjustment of substrate material, the new one could have better thermal stability, and also useful for speed improvement in Sb 2 Te alloy.

Results
The sheet resistance as a function of temperature (R-T) for Sb 2 Te and Sc 0.1 Sb 2 Te films (~100 nm) was measured to clarify the influence of doping on the thermal stability as depicted in Fig. 1(a). The sharp drop of the resistance happens around the crystallization temperature (T c ), which significantly shifts to higher temperature after doping Sc. According to the derivative of logarithmic sheet resistance with respect to temperature (dlgR/dT), the T c of Sb 2 Te and Sc 0.1 Sb 2 Te films are estimated to be 156.1 °C and 174.9 °C, respectively, indicating better amorphous stability after Sc doping. Figure 1(b) shows the 10-year data retention characteristics for Sb 2 Te and Sc 0 . 1 Sb 2 Te films. Based on the Arrhenius equation: t = τ·exp(E a /k B T), where t is the 50% criterion failure time, τ is the proportional time constant, E a is the crystallization activation energy, k B is the Boltzmann constant, T is the absolute temperature. After Sc doping, the activation energy E a increases from 2.44 eV to 3.00 eV. By extrapolating the data retention time to 10 years, the data retention temperature for Sc 0.1 Sb 2 Te film is estimated to be 92.7 °C, demonstrating a better thermal stability than that of Sb 2 Te (63.8 °C) and conventional GST (85 °C) alloy. Apart from better thermal stability, four orders of magnitude resistance difference leave enough margins for identifying the high resistance and low resistance states.
From microstructural side, thermally-induced phase transition processes were investigated by in-situ transmission electron microscope (TEM) technique for Sb 2 Te and Sc 0.1 Sb 2 Te films (~15 nm) with a heating rate of 10 °C/min. Figure 2 shows TEM bright-field (BF) images and the corresponding selected area electron diffraction (SAED) patterns for Sb 2 Te (a-c) and Sc 0.1 Sb 2 Te (d-f) films at different temperature. Pure Sb 2 Te film starts to crystallize with an explosive crystal growth at 140 °C (Fig. 2b), and the grain size is in about several hundred nanometers scale. After the temperature increases to 200 °C (Fig. 2c), the grain size of Sb 2 Te film is almost the same comparing with Fig. 2b, because its crystallization process has already finished around 140 °C. As for Sc 0.1 Sb 2 Te film, it starts to crystallize at 160 °C with numerous nanocrystals (<~10 nm) as shown in Fig. 2d. Though these nanocrystals grow a little (<~15 nm) as temperature rises up to 200 °C (Fig. 2e), the grain size of Sc 0.1 Sb 2 Te is still much smaller than that of Sb 2 Te after crystallization. In addition, both of the SAED patterns in Fig. 2c,f can be indexed as hexagonal (h-) Sb 2 Te structure (JCPDS No. 80-1722). No extra diffraction rings appear in Fig. 2f, which demonstrates that Sc 0.1 Sb 2 Te film is a single h-phase without phase separation. The XRD result of crystallized Sc 0.1 Sb 2 Te film, as shown in Fig. S1, further confirms that Sc 0.1 Sb 2 Te has the same structure as Sb 2 Te. That is, Sc doping significantly affects the crystallization behavior of Sb 2 Te film without forming any new phase or new structure. Beyond that, a crystallized film with three times more of Sc doping level, as much as 11% (Sc 0.4 Sb 2 Te), was investigated to inspect the distribution of Sc atoms by using of STEM-EDS mapping in TEM as shown in Fig. S2. Even at a higher doping level, the crystalline structures of Sc 0.4 Sb 2 Te and Sc 0.1 Sb 2 Te remain the same from SAED pattern side. This EDS results give more direction on uniform distributed Sc, Sb and Te elements without obvious phase separation appearing in nanometer scale.
In order to understand the interplay between Sc atoms and Sb 2 Te lattice, XPS experiment was applied to investigate the bonding state of crystallized Sb 2 Te and Sc 0.1 Sb 2 Te films. Figure 3 shows the binding energy of Sb 3d and Te 3d core levels for Sb 2 Te and Sc 0.1 Sb 2 Te films. The C 1 s peak at 284.8 eV is used as a reference. After Sc doping, both peaks of Sb 3d shift to lower energies (~0.2 eV for Sb 3d 5/2 , ~0.25 eV for Sb 3d 3/2 ). Similar results are observed in the binding energies for Te 3d (~0.25 eV for both Te 3d 5/2 and Te 3d 3/2 ). Usually, binding energy will decrease when an atom bonds to another one with a lower electro-negativity. Since the electro-negativity of Sc (1.36) is smaller than that of Sb (2.05) and Te (2.12), Sc atoms is very likely to bond with Sb and Te atoms in Sc 0.1 Sb 2 Te film after crystallization, resulting in the decrease of binding energy for Sb and Te elements. Considering that the electro-negativity difference (ΔS) of Sc-Te and Sc-Sb is much bigger than Sb-Te, and a large ΔS between two atoms would increase nucleation probability 22,23 . More nuclei are likely to generate after Sc doping, and the intergrowth of nuclei produces more grain boundaries, which will suppress the subsequent crystal growth significantly. This may be contributing to explain the much smaller grain size distribution after Sc doping as shown in Fig. 2f.
Anyway, good device performances are the key to application. Figure 4a shows the resistance-voltage curves of Sc 0.1 Sb 2 Te alloy based PCRAM cell with different pulse widths (the falling edge of the voltage pulse is 3 ns). Both set and reset voltages slightly shift to a higher value when the pulse width decreases. But most of all, even 6 ns electrical voltage pulse can still induce reversible phase transformation in this PCRAM device. Comparing to conventional GST (crystallization speed of ~ 50 ns) 10 and Sb 2 Te (crystallization speed of ~ 20 ns) 13 , Sc 0.1 Sb 2 Te based PCRAM cell exhibits faster operation speed. Besides, endurance up to 3.3 × 10 5 cycles without failure (Fig. 4b) also demonstrates that Sc 0.1 Sb 2 Te alloy has great potential for PCRAM application 24 .

Discussion
To further verify the location of doped Sc atoms, Ab initial method was carried out to theoretically predict the most probable site by calculating the formation energy (E f ) in each site. Rather than simulating the exact experiment composition, we introduce a single Sc atom at various lattice sites in a Sb 2 Te supercell with lattice parameter of 12.816◊12.816◊ 17.633(Å 3 ) to evaluate the effects of doped Sc atom. In the Sb 2 Te supercell, there are seven possible dopant sites for Sc atoms, Sb 1 , Sb 2 , Sb 3 , Te 1 , Te 2 and In 1 , In 2 (shown in Fig. S3a). Sb and Te with subscript  mean the substitution doping in which the Sc atom replaces Sb or Te, whereas In1 and In2 mean the Sc atom enters the interstitial site. The formation energy of each structure after relaxed was calculated and shown in Fig. S3b. The E f was obtained according to the following equation: Here -E un doped and -E Sc doped denote the total energies of the relaxed structure before and after Sc doping. E Sc denotes the chemical potential of doped Sc, E Sb Te / denotes the chemical potential of Sb or Te being replaced, while it is zero for interstitial doping. As shown in Fig. S3b, the E f for Sb 1 is −2.482/2.583 eV for Sb/Te rich, which is much lower than all of the other conditions. Thus, Sb 1 is the most energetically favorable position for the doped Sc atom. To identify the bonding information of Sc 0.1 Sb 2 Te when Sc substitutes Sb 1 , the charge density difference (CDD) of relaxed structure was illustrated in Fig. 5. Their corresponding 2D charge density plot was shown in Fig. S4, which shows that Sc and Sb are bound with Te through a bond point 25 . In order to show the chemical environment of Sc, we present the nine layers that exist in the Sb 2 Te h-structure along C axis. As shown in Fig. 5b,c, there is only tenuous charge present in three of the Sb-Te bonds which is in distinct contrast with the noticeable charge accumulation at the bond center between Sc and Te. The original Sb-centered octahedron  shows three strong (3.045 Å) bonds and three weak bonds (3.159 Å). However, the Sc-centered octahedron shows six identical strong bonds (2.98 Å), the strong Sc-Te bonds lead to the rigidity of Sc-centered octahedrons. Even they may not necessarily be intact in the melt-quenched amorphous phase, yet the Sc-centered octahedrons can still be the subcritical embryos owe to their lowest formation energy. So the reconstruction of Sc-centered octahedrons is more advantageous than that of Sb-centered motifs in the recrystallization process. The existence of large amounts of precursors will refine the crystalline size, thus increase the grain boundaries which will accommodate more stress produced in phase change process of PCRAM device. Moreover, smaller grain size will increase the interface-area-to-volume ratios to facilitate the hetero-crystallization at the grain boundaries, further accelerating the crystallization speed. This may explain why the SET speed of Sc 0.1 Sb 2 Te based PCRAM device (6 ns) is faster than Sb 2 Te based one (20 ns). However, after 3.3 × 10 5 cycles, the non-uniform electronic fields in the active mushroom shape area might lead to a reallocation of elements, thus appear Sb 2 Te large grains that can result in device failure.
Comparing with Sc 0.2 Sb 2 Te 3 material, Sb 2 Te alloy in this study was choosing as the parent material instead of Sb 2 Te 3 alloy, considering the Sb 2 Te's thermal stability is better. Sc 0.2 Sb 2 Te 3 material can change from amorphous state to face centered cubic (f-) phase, and then to the stable h-phase with the increase of temperature. The 700 picoseconds set speed 21 only involves phase change process between amorphous and f-phase, which is a metastable state like f-GST phase. Avoiding f-to-h phase transition is very important in PCRAM device, with an emphasis on preventing grain growth 26 . Hence, in this paper, Sc 0.1 Sb 2 Te material can reach 6 nanoseconds set speed without intermediate phase, benefiting from its strong Sc-centered cluster. It has high-speed and good thermal stability together, which may point out some direction for the future design of PCRAM devices.

Conclusion
In this paper, Sc doped Sb 2 Te film was investigated to verify its application in PCRAM. After Sc doping, the thermal stability of Sb 2 Te alloy was improved, and the 10-year data retention time was increased. The crystalline Sc 0.1 Sb 2 Te film exhibits a single phase without phase separation. Sc 0.1 Sb 2 Te based PCRAM cell can still realize stable reversible switching behaviors even at 6 ns. Two orders of resistance difference between set and reset state makes it easy to distinguish "0" and "1". Furthermore, endurance up to 3.3 × 10 5 cycles makes Sc 0.1 Sb 2 Te a promising material for PCRAM application.

Methods
Sb 2 Te and Sc doped Sb 2 Te films was deposited on SiO 2 /Si (100) substrates and carbon coated TEM grid by co-sputtering Sc and Sb 2 Te targets using RF sputtering system at room temperature. The composition of the deposited films, BF images and SAED patterns were characterized by JEOL-2100F TEM with energy dispersive spectroscopy (EDS). The bonding situation of Sb 2 Te and Sc 0.1 Sb 2 Te alloy was evaluated by X-ray photoelectron spectroscopy (XPS) with Al Kα radiation. T-shaped PCRAM cells with a tungsten bottom electrode (190 nm in diameter) were fabricated using 130 nm CMOS technology. Afterwards, Sc 0.1 Sb 2 Te film (about 55 nm), TiN film (10 nm) and Al top electrode (300 nm) were sequentially deposited. Resistance-voltage curves and programming cycles were monitored with Keithley 2400 and Tektronix AWG 5002B. Calculations in this work was investigate by using density functional theory (DFT) 27 . The Vienna Ab-initio Simulations Package (VASP) 28 was used for calculations. The projector augmented wave (PAW) 29 pseudopotentials were used to describe electron-ion interactions. For the exchange-correlation energies between electrons, the Perdew-Burke-Ernzerhof (PBE) 30 function was employed. The energy cut offs were chosen to be 450 eV and 350 eV for relaxation and static calculation. A supercell containing 3◊3◊1 unit cells of Sb 2 Te was constructed for relaxation. The 5◊5◊3 K point mesh with Gamma centered was used. The relaxation was performed until the total energy converged to within 1 meV.