Size-dependent and tunable crystallization of GeSbTe phase-change nanoparticles

Chalcogenide-based nanostructured phase-change materials (PCMs) are considered promising building blocks for non-volatile memory due to their high write and read speeds, high data-storage density, and low power consumption. Top-down fabrication of PCM nanoparticles (NPs), however, often results in damage and deterioration of their useful properties. Gas-phase condensation based on magnetron sputtering offers an attractive and straightforward solution to continuously down-scale the PCMs into sub-lithographic sizes. Here we unprecedentedly present the size dependence of crystallization for Ge2Sb2Te5 (GST) NPs, whose production is currently highly challenging for chemical synthesis or top-down fabrication. Both amorphous and crystalline NPs have been produced with excellent size and composition control with average diameters varying between 8 and 17 nm. The size-dependent crystallization of these NPs was carefully analyzed through in-situ heating in a transmission electron microscope, where the crystallization temperatures (Tc) decrease when the NPs become smaller. Moreover, methane incorporation has been observed as an effective method to enhance the amorphous phase stability of the NPs. This work therefore elucidates that GST NPs synthesized by gas-phase condensation with tailored properties are promising alternatives in designing phase-change memories constrained by optical lithography limitations.


Figures
. Energy dispersive X-ray spectrum, providing information on the stoichiometry of the NPs. The ratio of Ge:Sb:Te derived from the spectrum is 20:23:57 (±1), showing good agreement with the nominal composition of Ge 2 Sb 2 Te 5 .           Figure 4 in the main text. This measurement shows that the crystallization temperature dose not depend detectably on the concentration of NPs .   Table S1. The size dispersion and full-width half maximum (FWHM) of the temperature peaks displayed in figure S4. The FWHM because of the slow continuous heating can also be used as an indication of the crystallization speed.  Figure 3a-c in main text shows an example of selected area electron diffraction (SAED) patterns recorded for the NPs shown in Figure 2 in main text during their transformation from the amorphous phase to the crystalline one with increasing temperature. At room temperature, the diffraction pattern shows a broad band without any sharp diffraction spots, confirming the amorphous nature of the NPs. When this sample is heated to 140 °C, some faint, discrete diffraction spots emerge, suggesting that crystallization has started. Continuous heating triggers the increase of both the numbers and the intensity of these diffraction spots, demonstrating the progressing of the phase transformation. Finally, the diffraction rings, consisting out of discrete diffraction spots, are formed at 175 °C. The increase of temperature is halted when no obvious increase of diffraction intensity can be observed during a significant increase of the temperature (5 ℃). Too high temperatures are avoided in order to suppress evaporation of the NPs, which is detrimental to the TEM.

Crystallinity determination through in-situ heating TEM
Azimuthal integration has been applied to the diffraction patterns in order to derive in a straightforward manner the intensities for different diffraction rings as function of the distance in reciprocal space to the centre point. 1 The interplanar spacings calculated from SAED patterns after heating are d 200 =0.3058 nm, d 220 =0.2114 nm, indicating that the lattice parameter obtained by SAED is in good agreement with the one derived from the HRTEM images (see Figure 1b in the main text).
Repeating the azimuthal integration for all the SAED patterns recorded at different temperatures, the evolution of diffraction intensities with temperature is sequentially obtained, as shown in Figure 3d in the main text. Note that in this figure the background has been subtracted after the azimuthal integration. A broad band at the position corresponding to the {200} planes, which can be explained by the existence of short-range order (local structure of Ge/Sb-Te bondings) in these materials, is generally present in the X-ray diffraction and SAED measurements. 2,3 From Figure 3d, it is evident that the crystallization starts already at 130 °C, indicated by a visible diffraction intensity for {220} planes. Further heating leads to a continuous increase of the diffraction intensities, until the temperature increasing is halted at 175 °C, because no obvious increase of diffraction intensities is observed anymore in-between 170 °C and 175 °C.
Because of the fact that all the diffraction patterns were recorded at the same area, the evolution of the diffraction intensities with temperature is directly indicative for the progress of crystallization (fraction transformed). In order to quantitatively investigate the crystallization process and subsequently obtain the crystallization temperature (T c ) of these NPs, the diffraction intensities of {220} planes were selected as indicative of the fraction of NPs crystallized. The {200} diffraction intensities are less suitable, because of the interference with the initially present broad amorphous halo around the {200} reflections. The crystallization fractions as a function of temperature for NPs produced with hydrogen are depicted by open symbols in Figure 3e of the main text. The large, medium and small sized NPs (13.2±1.4 nm, 10.7±1.7 nm and 8.4±1.7 nm) are represented by black, red and blue colours, respectively. The Boltzmann function was used to fit the data in order to obtain the T c , which is denoted as temperature where the maximum transformation speed occurs. Good fits have been obtained for all three curves, with adjusted R 2 as 0.9959, 0.9957 and 0.9937 for the big, medium and small NPs respectively. Figure S8 depicts the spectra measured by energy dispersive X-ray spectroscopy (EDXS) attached to the transmission electron microscope (JEOL 2010F). A significantly higher carbon concentration (22 ± 11 at%) is observed for the NPs produced with high amount of CH 4 compared to the ones (8 ± 2 at%) produced with H 2 , while the carbon concentration on the SiN membrane (as reference) is negligible. We also observed that the carbon concentration varies remarkably between different measurements. A possible reason for this is that the NPs can also absorb carbon from the environment during transportation and/or inside the TEM because of their large surface to volume ratio, leading to the difference in carbon concentration. The detectable carbon in the nanoparticles produced with H 2 also indicates that some carbon is absorbed by the NPs, because no carbon is introduced during their deposition process. The low sensitivity of EDXS detectors to carbon strongly attributes to the extreme difficulty to accurately quantify the carbon concentration for NPs. It should be noted that the carbon concentration for the NPs is high (for both types of NPs), we are not able to quantify the relative amounts of carbon that are inside or surrounding the NPs. Nevertheless, the significant difference in the carbon concentration measured for the NPs produced with CH 4 compared to the ones with H 2 strongly supports that the difference in crystallization temperature observed for both particle types must be attributed to the carbon incorporation.

Numerical calculation for size-dependence of crystallization temperatures
Analogous to a previous work, 4  with d the diameter of the NPs, r a E the activation energy for the NPs with a diameter of r, k the Boltzmann constant and T the temperature. The symbol with a star represent the values for the bulk materials (here the nanowires with 190 nm in width). 5 The values for activation energies at different sizes are taken from reference. 5 Because of the relatively low T c observed here, an Arrhenius relation was used for crystal growth rate: with g E the activation energy (2.4 eV). 6 Using JMAK theory, 7-11 we can derived the peak temperatures of phase transformation (identical as T c defined in the present work). Adjusting the value of 0 u in Equation S2, the T c of bulk GST is set to 150 °C, close to the value for bulk GST PCMs from literature. The modeled T c as a function of diameter is displayed as the purple data points, in Figure 4 in main text.