Low-Energy Amorphization of Ti₁Sb₂Te₅ Phase Change Alloy Induced by TiTe₂ Nano-Lamellae

Abstract

Increasing SET operation speed and reducing RESET operation energy have always been the innovation direction of phase change memory (PCM) technology. Here, we demonstrate that ∼87% and ∼42% reductions of RESET operation energy can be achieved on PCM cell based on stoichiometric Ti₁Sb₂Te₅ alloy, compared with Ge₂Sb₂Te₅ and non-stoichiometric Ti_0.4Sb₂Te₃ based PCM cells at the same size, respectively. The Ti₁Sb₂Te₅ based PCM cell also shows one order of magnitude faster SET operation speed compared to that of the Ge₂Sb₂Te₅ based one. The enhancements may be caused by substantially increased concentration of TiTe₂ nano-lamellae in crystalline Ti₁Sb₂Te₅ phase. The highly electrical conduction and lowly thermal dissipation of the TiTe₂ nano-lamellae play a major role in enhancing the thermal efficiency of the amorphization, prompting the low-energy RESET operation. Our work may inspire the interests to more thorough understanding and tailoring of the nature of the (TiTe₂)_n(Sb₂Te₃)_m pseudobinary system which will be advantageous to realize high-speed and low-energy PCM applications.

Introduction

Non-volatile phase change memory (PCM), as one of the promising candidates, has great potential to serve as a storage class memory (SCM) to mitigate the widened performance mismatch between dynamic random access memory (DRAM) and non-volatile NAND Flash memory^1,2. In the PCM cell, a chalcogenide material, for example, Ge₂Sb₂Te₅, can be switched between the crystalline (c-) and amorphous (a-) phases, corresponding to the SET and RESET states^3,4, respectively. The big resistance contrast of such two states is utilized for storing “0” and “1” data states. The RESET operation refers to an amorphization procedure which melts the c-phase and subsequently quenches it into a-phase by applying a short intense electrical pulse on the PCM cell. Conversely, a longer pulse of lower intensity for SET operation can heat the a-phase to a temperature between crystallization temperature (T_c) and melting point (T_m) to obtain the c-phase. To achieve high density SCM application, scaling capability of the PCM cell is largely limited by its high RESET current or energy⁵ for which the premature degradation of the switching material or the thermal disturbance among nearest cells would happen and cause reliability and endurance issues⁶.

Since the slow SET speed and high RESET power remain to be important limitations for developing DRAM-like PCM, our previous work demonstrated at least one order of magnitude faster SET speed and as low as one-fifth of the RESET current and energy on a Ti_0.4Sb₂Te₃ based PCM cell compared to those of the Ge₂Sb₂Te₅ based cell with the same size⁷. The stable c-Ti_0.4Sb₂Te₃ alloy has the nano-scale phase separated morphology that hexagonal (HEX) Sb₂Te₃ and HEX-TiTe₂ crystals coexist⁸. The triple-layered TiTe₂ crystal lamellae locate adjacent to the quintuple-layered Sb₂Te₃ grains, while there are also some Ti atoms penetrate into Sb₂Te₃ lattice by occupying Sb sites and constructing Ti-centered octahedrons with surrounding Te atoms⁸. Such TiTe₂ lamellae along with the Ti-centered octahedrons can preserve their ordering configurations even after high temperature RESET operations, which is believed to be responsible for the performance boost of the Ti_0.4Sb₂Te₃ based PCM cell^7,8.

The influence of Ti doping content (x) on the phase change properties of Ti_xSb₂Te₃ materials was also studied^9,10. Although increasing Ti content can enhance 10-year data retention, reduce the volume change upon phase transition, improve interfacial adhesion ability, excessive Ti content (x ≈ 0.56, ∼10.1 at.%) causes Ti segregation which leads to poor endurance characteristic of the PCM cell^9,10. This phenomenon may correlate to a low solid solubility of Ti atoms in the HEX-Sb₂Te₃ lattice, where only limited number of Ti-centered octahedrons could dispersedly distribute inside the quintuple-layered building blocks so as to preserve the ordering configuration⁸. To inhibit the Ti precipitation after repeated RESET-SET operations, it would be worthwhile to try to concomitantly raise the Te content of the non-stoichiometric Ti_xSb₂Te₃ materials. Thus, in this paper, we show the better electrical phase change properties based on a stoichiometric Ti₁Sb₂Te₅ material. It is like a pseudobinary compound with 1 : 1 ratio of TiTe₂ and Sb₂Te₃ components. Even containing higher Ti content (12.5 at.%), the Ti₁Sb₂Te₅ based PCM cell has superior endurance characteristic over the Ti_0.56Sb₂Te₃ based one. In addition, without significantly sacrificing the SET speed, the RESET energy of the Ti₁Sb₂Te₅ based PCM cell is further lowered by 42∼47% compared to the Ti_0.4Sb₂Te₃ based one. We argue that the richer concentration of TiTe₂ lamellae in c-Ti₁Sb₂Te₅ plays a major role in decreasing the RESET energy, meanwhile the lack of Ti-centered octahedrons resided in the quintuple-layered Sb₂Te₃ lattice may slow down the nucleation rate, however the increasing TiTe₂ lamellae can act as structure-ordering template to enhance the crystal growth rate¹¹. Accordingly, we believe that Ti₁Sb₂Te₅ material is promising for realizing DRAM-like PCM application once advanced fabrication techniques being applied to further shrink the device dimension.

Figure 1 shows the temperature-dependent sheet resistance (R_s) curves of as deposited TiTe₂, Sb₂Te₃, Ti_0.4Sb₂Te₃ and Ti₁Sb₂Te₅ films upon in situ annealing with a heating rate of 10 °C/min. As the annealing temperature increases, a continuous decrease in R_s is observed for each film. Due to the partial crystallization during the sputtering process, compared to Ti_0.4Sb₂Te₃ and Ti₁Sb₂Te₅ films, both TiTe₂ and Sb₂Te₃ films have smaller initial R_s and present a smooth decrement in R_s. By comparison, one can observe the sudden drop in R_s occurs for Ti_0.4Sb₂Te₃ and Ti₁Sb₂Te₅ films when the temperature reaches T_c (both around 186 °C)⁹. The decrease in R_s with increasing temperature just before the onset of the crystallization indicates a semiconductor-like behavior. The temperature dependence for the R_s in a semiconductor can be expressed by R_s = R_s0exp(−E_σ/kT)¹², where R_s0 is a pre-exponential factor and E_σ is the activation energy for electrical conduction. The fitting results of E_σs of Ti_0.4Sb₂Te₃ and Ti₁Sb₂Te₅ are 0.11 eV and 0.13 eV, respectively. The activation energy of electrical transport is simply determined by half of the band gap E_σ = E_G/2 + ΔE, where E_G/2 is the distance from the Fermi level to the conduction band and ΔE is the depth of the trap states¹³. In the case of intrinsic conduction with equal amounts of electrons and holes, the Fermi level is situated at the middle of the band gap. Thus we can roughly estimate the optical band gaps (E_OPs) of a-Ti_0.4Sb₂Te₃ and a-Ti₁Sb₂Te₅ to be ∼0.22 eV and ∼0.26 eV, respectively, both of which are quite smaller than that of a-Ge₂Sb₂Te₅ (∼0.70 eV)¹⁴. Because the carrier density inside the semiconductor is proportional to exp(−E_G/2kT), where E_G is the electrical band gap which is roughly identical to the E_OP, a decrease in the band gap as the temperature approaching to the T_c will lead to the generation of a large number of carriers, which makes a major contribution to the quick drop in film resistivity¹⁴. On this view, one may roughly estimate that a-Ti_0.4Sb₂Te₃ and a-Ti₁Sb₂Te₅ could have quite faster crystallization (resistivity decrement) speed than that of a-Ge₂Sb₂Te₅ and a-Ti_0.4Sb₂Te₃ could be a little quicker than a-Ti₁Sb₂Te₅.

Figure 1

The sheet resistance (R_s) as a function of in situ annealing temperature for the TiTe₂, Sb₂Te₃, Ti_0.4Sb₂Te₃ and Ti₁Sb₂Te₅ films.

The 150 nm thick films deposited on the SiO₂/Si (1 0 0) substrate are measured with the in situ heating rate of 10 °C/min.

Figure 2a compares the SET speed of Ge₂Sb₂Te₅, Ti_0.4Sb₂Te₃ and Ti₁Sb₂Te₅ based PCM cells with the same size (BEC D = 190 nm), which has the same trend as aforementioned estimation. As the magnitude of applied voltage pulse reaches 1.3 V and 1.5 V, respectively, both the Ti_0.4Sb₂Te₃ and Ti₁Sb₂Te₅ cells show the SET speed of ∼6 ns. Apparently, under lower bias, the Ti_0.4Sb₂Te₃ cell can complete the SET operation more quickly than the Ti₁Sb₂Te₅ cell. In contrast, the SET operation of Ge₂Sb₂Te₅ cell requires ∼75 ns at 1.6 V and ∼35 ns even at 2.1 V. In other words, one order of magnitude faster SET speed can still be achieved even by using the Ti₁Sb₂Te₅ cell. Supplementary Information Figure S1 shows the cell resistance versus required time curves for SET operation of such three cells.

Figure 2

Comparisons of the SET operation speed and RESET energy.

(a) SET operation speeds for the Ge₂Sb₂Te₅, Ti_0.4Sb₂Te₃ and Ti₁Sb₂Te₅ based PCM cells with the same D = 190 nm W BEC. (b) RESET energy as a function of BEC diameter D for the Ge₂Sb₂Te₅, Ti_0.4Sb₂Te₃ and Ti₁Sb₂Te₅ based PCM cells.

In terms of the RESET operation, even the Ti_0.4Sb₂Te₃ (3.12 nJ) and Ti₁Sb₂Te₅ (1.65 nJ) cells with D = 190 nm BEC have noticeable lower energies compared to the Ge₂Sb₂Te₅ cells with smaller BEC (9.48 nJ for D = 130 nm BEC and 4.20 nJ for D = 80 nm BEC). More significant RESET energy reduction can be achieved on the Ti_0.4Sb₂Te₃ (0.95 nJ) and Ti₁Sb₂Te₅ (0.55 nJ) cells with D = 80 nm BEC. Namely, ∼78% and ∼87% of the energy have been saved via using the Ti_0.4Sb₂Te₃ and Ti₁Sb₂Te₅ cells (D = 80 nm BEC), respectively. Note that the Ti₁Sb₂Te₅ cell achieves a substantial (42∼47%) RESET energy reduction on the basis of the Ti_0.4Sb₂Te₃ cell. The shrinkage of BEC size also remarkably decreases the RESET current as shown in the Supplementary Information Figure S2. An ∼82% reduction of the RESET current is realized for both the Ti_0.4Sb₂Te₃ and Ti₁Sb₂Te₅ cells (∼0.5 mA) compared to that of Ge₂Sb₂Te₅ cell (∼2.8 mA) with the same D = 80 nm BEC. Moreover, the RESET current of the Ti₁Sb₂Te₅ cell (∼1.1 mA) is relatively smaller than that of the Ti_0.4Sb₂Te₃ cell (∼1.3 mA) with the same D = 190 nm BEC. In addition to the improvements in RESET energy and current, the endurance characteristics of the Ti₁Sb₂Te₅ cell (∼10⁷ cycles) is not inferior to that of the Ti_0.4Sb₂Te₃ cell⁷, which is obviously far more better than that of the Ti_0.56Sb₂Te₃ cell (<10⁶ cycles with severe fluctuation of the RESET state)⁹, as shown in the Supplementary Information Figure S3.

We used Raman spectroscopy (Fig. 3a) and in situ XRD (Fig. 3b) to characterize the c-Ti₁Sb₂Te₅ phase. As a CdI₂-like structure with space group Pm1 space symmetry, c-TiTe₂ has two Raman active modes produced entirely by the Te atoms for both the in-plane, E_g peak (∼122 cm⁻¹) and out of plane, A_1g peak (∼143 cm⁻¹), with the Ti atoms at rest¹⁵. Since there is no distinctive shoulder near ∼160 cm⁻¹, the c-TiTe₂ film can be considered as a stoichiometric compound without noticeable defects or impurities¹⁶. Because c-Sb₂Te₃ crystal belongs to the space group Rm, it has three Raman active modes, including two out of plane vibrations A_1g(1) peak (∼69 cm⁻¹) and A_1g(2) peak (∼165 cm⁻¹) and one in-plane vibration E_g(1) peak (∼112 cm⁻¹)¹⁷. Compared to the Raman curves of the c-TiTe₂ and c-Sb₂Te₃ films, the c-Ti₁Sb₂Te₅ film has five Raman active modes identically peaked corresponding to E_g and A_1g of TiTe₂ and A_1g(1), A_1g(2) and E_g(1) of Sb₂Te₃, respectively, as shown in Fig. 3a. The coexistence of c-TiTe₂ and c-Sb₂Te₃ Raman active modes in c-Ti₁Sb₂Te₅ no doubt shall originate from the two separated phases which is already observed in c-Ti_0.4Sb₂Te₃⁸. In fact, our in situ XRD result of Ti₁Sb₂Te₅ film clearly proves such phase separation phenomenon as shown in Fig. 3b, where both HEX-Sb₂Te₃ and HEX-TiTe₂ diffraction peaks can be identified. Nevertheless we did not observe such distinct phase separation in Ti_0.4Sb₂Te₃ through the same in situ XRD measurement⁷. The phase separation in Ti_0.4Sb₂Te₃ occurs in nano-scale dimension⁸, where the TiTe₂ lamellae segregate into no more than 10 nm-width belts and most of the TiTe₂ lamellae (<2 nm in width) inlay with the Sb₂Te₃ quintuple-layered blocks. Due to the similar HEX lattice structures of TiTe₂ and Sb₂Te₃ and also considering the lower doping content of Ti (∼7.4 at.%) in Ti_0.4Sb₂Te₃, less concentration of the TiTe₂ lamellae may result in undetected XRD signal. Note that the pure TiTe₂ alloy has quite high T_m (>1200 °C)¹⁸ and the HEX-phase of its thin film can be maintained even at 700 °C (higher than the T_m ∼618 °C of Sb₂Te₃)⁸, thus there is no segregated Te or Ti phase being observed in c-Ti₁Sb₂Te₅ as shown in Fig. 3b.

Figure 3

The Raman spectra of c-TiTe₂, c-Sb₂Te₃ and c-Ti₁Sb₂Te₅ and in situ XRD results of Ti₁Sb₂Te₅.

(a) Raman spectra of the c-TiTe₂, c-Sb₂Te₃ and c-Ti₁Sb₂Te₅ films. One can see two main E_g (○) and A_1g (□) phonon modes for c-TiTe₂ film and three main A_1g(1) (■), E_g(1) (●) and A_1g(2) (▲) phonon modes for c-Sb₂Te₃ film. The c-Ti₁Sb₂Te₅ film shows all the five phonon modes. (b) In situ XRD curves of Ti₁Sb₂Te₅ film at different temperatures, where HEX lattice planes of Sb₂Te₃ (♦) and TiTe₂ (◊) can be identified.

Since the Ti dopants in c-Ti_0.4Sb₂Te₃ either segregate in the form of TiTe₂ lamellae or construct Ti-centered octahedrons scattered in the quintuple-layered blocks of the Sb-Te lattice⁸, this non-stoichiometric composition can be chemically regarded as (TiTe₂)_xTi_0.4−xSb₂Te_3−2x (0 < x < 0.4), where (TiTe₂)_x part corresponds to the TiTe₂ lamellae and in Ti_0.4−xSb₂Te_3−2x part the Ti_0.4−xTe_0.8−2x accounts for the scattered Ti-centered octahedrons. The Ti_0.4-xSb₂Te_3-2x part behaves more like the Sb-rich Sb-Te compound contributing to the faster recrystallization⁸. Extending this analysis to the stoichiometric Ti₁Sb₂Te₅ = (TiTe₂)_yTi_1-ySb₂Te_5-2y (0 < y < 1), where (TiTe₂)_y part also stands for the segregated TiTe₂ lamellae, one can easily find that, in Ti_1−ySb₂Te_5−2y part (=Ti_1−yTe_2−2ySb₂Te₃), if Ti_1−yTe_2−2y is assigned as the Ti-centered octahedrons accommodated in Sb-Te quintuple-layered blocks, it will be contradictory to keep the rest Sb-Te part in a ratio of 2 : 3 without Te deficiency. In other words, since there is no Ti or Te phase separation, it is more appropriate to describe the c-Ti₁Sb₂Te₅ as (TiTe₂)₁(Sb₂Te₃)₁ in which all the Ti atoms are supposed to be contained in the separated TiTe₂ lamellae. Of course, this is a rough estimate, however, we may still get a reasonable inference that compared to the c-Ti_0.4Sb₂Te₃, c-Ti₁Sb₂Te₅ should have a higher concentration of the c-TiTe₂ lamellae but a lower concentration of the “solid-solute” Ti-centered octahedrons.

Note that the quasi-two-dimensionalc-TiTe₂ is semimetallic¹⁹ (see Fig. 1 also) with a quite low thermal conductivity (∼0.12 W/mK)²⁰. The TiTe₂ lamellae could act like the embedded nano-electrodes in c-Ti₁Sb₂Te₅ to conduct the electrical current to the adjacent Sb₂Te₃ grains to generate Joule heat. The heat dissipation could be concomitantly refrained by those low thermal-conductive TiTe₂ lamellae so as to effectively enhance the thermal efficiency of the RESET operation. Not surprisingly the c-Ti₁Sb₂Te₅ (∼1/2 = 50.0% concentration of TiTe₂ lamellae) based PCM cell can accomplish substantial reduction of the RESET energy compared to the c-Ti_0.4Sb₂Te₃ (< ∼0.4/1.4 ≈ 28.6% concentration of TiTe₂ lamellae) based PCM cell. On the contrary, the lack of survived Ti-centered octahedrons in Sb-Te rich amorphous matrix after RESET operation may slow down the nucleation rate for recrystallization process^8,9, leading to a relatively slower SET speed for the Ti₁Sb₂Te₅ based PCM cell as compared to that of the Ti_0.4Sb₂Te₃ based one.

We also used the two-dimensional finite element method (FEM) to simulate and compare the RESET operations of the Ge₂Sb₂Te₅, Ti_0.4Sb₂Te₃ and Ti₁Sb₂Te₅ based PCM cells, as shown in Fig. 4. The Joule heat is mainly generated in the phase change films. The thermal transfer obeys the standard heat conduction equation:²¹

Figure 4

The two-dimensional finite element method simulations for the RESET operation.

Simulated RESET temperature distributions in PCM cells with (a) c-Ge₂Sb₂Te₅, (b) c-Ti_0.4Sb₂Te₃ and (c) c-Ti₁Sb₂Te₅ layers. In (a) all the square grids represent the small Ge₂Sb₂Te₅ crystal grains, while in (b,c) the grids marked with slash lines stand for the TiTe₂ crystal grains and other non-marked ones are belonged to the Sb₂Te₃ crystal grains. The isothermal curves with corresponding temperatures in the PCM cells are also shown.

where κ, is the thermal conductivity, c, the specific heat, ρ, the density, t, the time, T, the temperature and Q, the Joule heat per unit volume and per unit time, which is called the heat density. The key material parameters for FEM simulations include κ of c-Ge₂Sb₂Te₅ (∼0.46 W/mK)²¹, c-Sb₂Te₃(∼0.78 W/mK)²² and c-TiTe₂ (∼0.12 W/mK)²⁰, c of c-Ge₂Sb₂Te₅ (∼1.20 J/cm³K)²³, c-Sb₂Te₃(∼1.02 J/cm³K)²⁴ and c-TiTe₂ (assumed to be ∼1.80 J/cm³K of c-TiSe₂)²⁵ and ρ of c-Ge₂Sb₂Te₅ (∼6.2 g/cm³)²¹, c-Sb₂Te₃ (∼6.5 g/cm³)²⁶ and c-TiTe₂ (∼6.3 g/cm³)²⁷. The phase change film layers of the models are divided into grid shape to represent the poly-crystalline morphology. Each grid denotes the small crystal grain. ∼29% and ∼50% of the grids in the c-Ti_0.4Sb₂Te₃ and c-Ti₁Sb₂Te₅ layers are randomly chosen to be the c-TiTe₂ grains, respectively, as shown in Fig. 4b,c. Constant voltage pulse is applied to the axis-symmetric mushroom-type (T-shaped) cells. It can be observed the highest peak temperature is achieved in the Ti₁Sb₂Te₅ based cell (Fig. 4c) while the Ge₂Sb₂Te₅ based cell (Fig. 4a) has the lowest peak temperature. Apparently, more heat can be generated and confined in the phase change film layer as c-TiTe₂ concentration increases, therefore lower energy is needed for the RESET operation.

In summary, the pseudobinary Ti₁Sb₂Te₅ phase change alloy shows drastically decreased RESET energy, while increasing the SET speed, of the PCM cell compared to the Ge₂Sb₂Te₅ based one. Without significantly reducing the SET speed as compared to the Ti_0.4Sb₂Te₃ based PCM cell, nearly half of the RESET power can be saved on the Ti₁Sb₂Te₅ based one. These improvements are achieved by introducing more nano-scale separated TiTe₂ lamellae. We believe with more thermally stable TiTe₂ lamellae the efficiency of electric conduction and heat inhibition could be greatly enhanced for the low-energy RESET operation. We expect the speed/power to be further increased/decreased significantly on thorough investigations of the (TiTe₂)_n(Sb₂Te₃)_m pseudobinary system and device dimension scaling techniques. In this regard, for example, a superlattice or multilayered structure constructed by alternate TiTe₂/Sb₂Te₃ stacking film instead of the co-sputtering one will be a great help. It may also be possible to search topological superconducting properties on a finely-tuned TiTe₂/Sb₂Te₃ superlattice sample.

Methods

Ti_0.4Sb₂Te₃ films were deposited by co-sputtering of pure Ti and Sb₂Te₃ targets. By adding an additional pure Te target, three-target co-sputtering technique was used to fabricate the Ti₁Sb₂Te₅ films. Similarly, TiTe₂ films were obtained by co-sputtering of pure Ti and Te targets. For Sb₂Te₃ and Ge₂Sb₂Te₅ films, respective pure alloy target was used for sputtering. The compositions of all films were measured by X-ray fluorescence spectroscopy using a Rigaku RIX 2100 system. The temperature-dependent sheet resistance changing trends of TiTe₂, Sb₂Te₃, Ti_0.4Sb₂Te₃ and Ti₁Sb₂Te₅ films with the same 150 nm thickness were studied by Linkam LMP 95 hot stage. For real-time observation of structure transition in Ti₁Sb₂Te₅ film, vacuum in situ X-ray diffraction (XRD) measurement with a 20 °C/min heating rate was performed on 300-nm-thick film (deposited on Si substrate at room temperature) using PANalytical X’Pert PRO diffractometer with a Cu Kα (λ = 0.15418 nm) radiation source. The diffraction data were collected in the 2θ range of 10°–60° with a scanning step of 0.02°. Raman spectroscopy (Thermo Fisher DXR) was performed on 300-nm-thick film samples at room temperature using an Ar⁺ laser (wavelength 532 nm) with ∼1 μm² beam spot.

T-shaped PCM cells with diameter (D) = 190 (80) nm tungsten plug bottom electrode contact (BEC) were fabricated using 0.13 μm complementary metal-oxide semiconductor technology. In all the PCM cells, the thickness of the switching material films is around 170 nm. The 15-nm-thick TiN and 300-nm-thick Al films were used as top electrode for all cells. All the electrical measurements were performed by using the Keithley 2600C source meter (measuring cell resistance), the Tektronix AWG5002B pulse generator (generating voltage pulse with a minimum width of ∼6 ns), the homemade constant current driver (generating current pulse with a maximum magnitude of ∼10 mA) and the Tektronix 7054 digital phosphor oscilloscope (measuring transient voltage drop across the cell when current pulse is applied).