Atomistic understanding of crystallization in solids is incomplete due to the lack of appropriate materials and direct experimental tools. Metallic glasses possess simple metallic bonds and slow crystallization kinetics, making them suitable to study crystallization. Here, we investigate crystallization of metallic glass-forming liquids by in-situ heating metallic glass nanorods inside a transmission electron microscope. We unveil that the crystallization kinetics is affected by the nanorod diameter. With decreasing diameters, crystallization temperature decreases initially, exhibiting a minimum at a certain diameter, and then rapidly increases below that. This unusual crystallization kinetics is a consequence of multiple competing factors: increase in apparent viscosity, reduced nucleation probability and enhanced heterogeneous nucleation. The first two are verified by slowed grain growth and scatter in crystallization temperature with decreasing diameters. Our findings provide insight into relevant length scales in crystallization of supercooled metallic glasses, thus offering accurate processing conditions for predictable metallic glass nanomolding.
Nanoscale metallic glasses (MGs) offer novel technological applications1 due to their superior mechanical properties such as high mechanical strength, high elasticity, high fracture toughness and high corrosion resistance2,3,4,5,6,7,8,9. In some cases, their unique chemical compositions are favourable for catalysis10, bioimplants11 and antibacterial applications12. Owing to the absence of an intrinsic length scale, hierarchical structures with multiscale surface features from centimeters down to nanometers are possible for MGs to engineer novel material properties10,13, which is challenging in crystalline metals. Patterning of MGs is typically achieved by thermoplastic forming14—pressing heated MGs into molds, where processing temperatures are in the supercooled liquid region bounded by the glass transition temperature and crystallization temperature. Most readily, molding is realized at the highest temperature where crystallization can still be avoided, thus retaining the attractive properties of the amorphous material15,16. For reliable thermoplastic processing of multiscale structuring of MGs, crystallization kinetics at all length scales must be understood. On the macroscale, crystallization kinetics have been determined for a wide range of alloys, enabling predictable molding15. By contrast, the crystallization behaviour of MG-forming liquids at nanoscale dimensions has not been carefully considered. This is in part due to the non-trivial experimental difficulty associated with detecting and accurately recording the heat signals from nanoscale samples as they undergo crystallization in the differential scanning calorimetry experiments that are typically used, to good effect, for macroscale samples. Despite experiments on a large ensemble of nanoscale samples indicating accelerated crystallization17, it is unclear whether crystallization kinetics remains unaltered at nanoscale. Up to now, macroscale processing parameters have been used for nanoscale molding of MGs with often-unpredictable crystallization behaviours18. Recently, nanocalorimetry has been used to study nanoscale samples19,20. One major challenge of nanocalorimetry is accuracy, and calorimetry alone cannot directly relate heat release with nanoscale MG morphology. In contrast, in-situ transmission electron microscope (TEM) techniques can simultaneously introduce and visualize structural changes to reveal phenomena that cannot be observed from ex-situ, ensemble techniques. In-situ TEM has been used to observe the dynamic structural changes of nanomaterials, such as growth of nanocrystals21 and nanowires22, and phase transformations in glassy materials23,24, providing a near atomistic understanding of their underlying mechanisms.
Here, we apply in-situ heating of MG nanorods inside a TEM and visualize their crystallization characteristics at unprecedentedly small length scales. We directly observe the structural changes of MG nanorods during crystallization and reveal unusual intrinsic size effects on their crystallization temperature, marking a departure from continuous flow description of MG kinetics.
In-situ TEM heating
Amorphous MG nanorods of Pt57.5Cu14.7Ni5.3P22.5 are prepared by a nanomolding technique14 (Supplementary Fig. 1). Figure 1a shows a representative TEM image of as-pressed MG nanorods on an in-situ TEM thermal grid that can be heated up to 1100 °C. The nanorods are heated from room temperature to above their liquidus temperature inside the TEM, and their structural changes during heating are observed using bright-field (BF) imaging, dark-field (DF) imaging and selective area electron diffraction (SAED) (Fig. 1b,c). The glass transition (Tg), crystallization (Tc) and melting (Tm) temperatures of bulk samples are indicated in Fig. 1b.
Size-dependent crystallization temperature
We characterize the crystallization behaviour of individual nanorods in the diameter range of ∼6–330 nm under in-situ heating. Tc is detected from SAED or DF TEM movies during heating (Fig. 2) as the onset of a rapid increase in the electron intensity in the back focal plane (SAED) or image plane (DF), respectively. SAED in-situ movies reveal that a thinner nanorod crystallizes at a lower Tc than a thicker rod (Fig. 2a,b, Supplementary Movies 1 and 2). The general trend is a monotonically decreasing Tc with decreasing nanorod diameter down to ∼25 nm. For nanorods below 25 nm, Tc shows a reverse trend, increasing rapidly with decreasing nanorod diameter. Figure 2c shows snapshot BF TEM images that compare the crystallization kinetics of two nanorods with diameters below ∼25 nm. Here, a thinner nanorod crystallizes at a higher Tc than a thicker one, which is the opposite behaviour of the ones observed in Fig. 2a,b. This reversed trend of Tc is best visible in tapered nanorods with narrowing diameters. Figure 2d shows snapshot DF TEM images of a tapered nanorod under in-situ heating (Supplementary Movie 3). Crystalline grains grow first in the thicker region and subsequently grow towards the thinner region. Remarkably, the thinner region, which is below 5 nm, never crystallizes. Crystallization of an ∼6-nm nanorod occurs at ∼340 °C (Fig. 2e), which is ∼100 °C higher than the Tc for a 150-nm nanorod (Fig. 2a). Figure 2f presents quantitatively measured Tc’s of nanorods of various diameters under in-situ heating (see Methods). It reveals the initial decrease of Tc with decreasing nanorod diameter and a rapid increase of Tc below 25 nm. No detectable changes in the chemical composition of nanorods were observed during in-situ heating by energy-dispersive X-ray spectroscopy (Supplementary Fig. 2) and cyclic heating experiments (Supplementary Fig. 3), ruling out chemical and electron beam effects as the origin for the observed Tc change.
A summary of Tc’s for all considered nanorods is given in Fig. 3a. Tc monotonically decreases with decreasing diameter down to ∼25 nm, below which it rapidly increases. This non-monotonic crystallization behaviour is discussed in the context of size-dependent apparent viscosity, nucleation probability and heterogeneous nucleation, all of which affect nucleation, growth rates and crystallization kinetics (Fig. 3b–e). For nanorods, heterogeneous nucleation on the surface of nanorods dominates over homogeneous nucleation, which would accelerate crystallization kinetics with decreasing nanorod diameter, thus lowering the Tc (Fig. 3b, black curve)25. Below a critical nanorod diameter, however, two additional factors affect the crystallization kinetics: reduced probability of nucleation and increase in the apparent viscosity.
We discuss the increase in the apparent viscosity first. When the sample size approaches the size scale of intrinsic flow units such that the shear transformation zone can no longer freely operate, the flow resistance increases, causing apparent viscosity, η, to increase26. Such departure from a continuum flow description, reflected in the increase of viscosity with decreasing confinement sizes, or equally, increasing Knudsen number, has been observed in other viscoelastic systems, including polymers or glassy materials26. The increase in viscosity would lead to decrease in nucleation rate and grain growth rate, as the jumping frequency and the diffusion constant are inversely proportional to the viscosity (Fig. 3b,c, red curves)27. We have experimentally investigated the growth kinetics in a nanorod with a locally narrowed region (Fig. 4). Indeed, we observe a suppressed grain growth rate in the narrower region during in-situ crystallization (Supplementary Movie 4). The growth front significantly slows down through the narrow region, as indicated by the snapshot DF TEM images (Fig. 4). The growth mode appears to remain the same throughout the entire growth time. Thus, this is a direct consequence of the increase in the apparent viscosity, which leads to the slowdown of crystallization kinetics and the increase of Tc, in agreement with our observation. Systematic size-dependent grain growth rate data are shown in Supplementary Fig. 4. Discussion on the growth mechanism can be found in Supplementary (Supplementary Figs 4,5).
In addition to the increasing apparent viscosity and heterogeneous nucleation, reduced probability of nucleation critically affects the crystallization kinetics, particularly for small nanorods. As the number of nuclei scales with the surface area of the nanorod, the probability of nucleation is anticipated to decrease if the nanorod size approaches the mean distance between nuclei (Fig. 3d, Supplementary Fig. 6). This ultimately slows down the crystallization kinetics, reflected as increase in Tc. Due to the stochastic nature of nucleation, the reduced probability of nucleation will result in a larger scatter in Tc for smaller rods. To confirm this, we performed in-situ crystallization experiments repeatedly and measured the scatter in Tc as a function of nanorod diameter. In agreement with our prediction, the scatter in Tc gradually increases for smaller nanorods (Fig. 5), pointing to the effect of the reduced probability of nucleation for small rods. The non-monotonic behaviour of Tc shown in Fig. 3a, the main finding of this work, still holds even in the presence of the scatter. We note that a decrease in melting temperature (Tm), expected from the Gibbs–Thomson effect, does not alter the observed diameter-dependent Tc (Supplementary Fig. 7).
Size-dependent Tc that we map out for nanoscale MGs down to ∼5 nm lays the foundation for their stable processing. The increase of Tc, owing to the reduced probability of nucleation, points to a larger processing window than previously assumed for feature sizes smaller than ∼30 nm to circumvent crystallization. By employing higher processing temperatures, current size limits for nanoscale MGs may be overcome. Fundamentally, our in-situ study of size effects on the crystallization behaviour of MGs reveals relevant length scales and a departure from a continuous flow description of MGs at the nanoscale.
Preparation of MG nanorods
The synthesis of Pt57.5Cu14.7Ni5.3P22.5 MG nanorods used in this study is described in detail in our previous paper14. Briefly, alloying of high-purity constituents with nominal compositions was carried out in a vacuum-sealed quartz tube using induction melting. Subsequently, the master alloys were fluxed with boron trioxide (B2O3) and vitrified by water quenching from a temperature of 1100 °C, which is 350 °C above the thermodynamic liquidus temperature. A commercially available porous anodized aluminium oxide (AAO) mold (Synkera Inc.) was placed between two custom-built heating plates, at a constant temperature of 260 °C (Supplementary Fig. 1a). A piece of bulk MG was placed and equilibrated on the AAO mold at 260 °C for 30 s (Supplementary Fig. 1b). Nanorods were thermoplastically formed by pressing the bulk MG into the selected AAO mold, under a load at a linear ramping from 0 to 100 kN (Supplementary Fig. 1c). The pressed sample was subsequently released from the mold by dissolving the AAO mold in a 20 wt.% potassium hydroxide (KOH) solution, heated at 80 °C for ∼10 h (Supplementary Fig. 1d). The released MG plate with nanorods attached was rinsed with distilled water and isopropyl alcohol (IPA). The rinsing process was repeated more than four times to minimize the residual salts and AAO mold fragments. Then, the nanorods were released and collected by sonication in IPA (Supplementary Fig. 1e). The minimum diameter of the AAO molds used was 13 nm. Nanorods of diameters <∼10 nm were stochastically produced in response to the dewetting process, by rapidly melting/quenching as-pressed nanorods inside the TEM column (∼10−5 Pa vacuum level). The melt/quench was carried out at 900 °C with a fast quenching rate of ∼106 K s−1. Energy-dispersive X-ray spectroscopy characterizations (Supplementary Fig. 2) show no detectable differences in the chemical composition of nanorods irrespective of diameters and in-situ heating processes.
In-situ TEM heating experiments
All TEM experiments were performed with a FEI Tecnai Osiris 200 kV TEM. As-pressed nanorods dispersed in IPA were transferred to in-situ TEM substrates, via a drop casting method. We used commercially available in-situ TEM heating grids (E-AHA) and a heating holder (Aduro 300DT System) manufactured by Protochips, Inc. The heating substrate contains a resistive ceramic membrane contacted with metal electrodes, which introduce Joule heating to the membrane. The ceramic membrane contains an array of holes, over which a thin carbon film is overlaid. Nanorods sitting on the amorphous carbon film were inspected for Tc identification, so there is no contribution of the ceramic on diffraction spots or TEM imaging. The temperature profile of each substrate was pre-calibrated by the manufacturer. For the identification of Tc, nanorods were first rapidly heated to 170 °C (below Tg) within 0.1 s. Subsequently, a slower heating with a ramp rate of 0.67 °C s−1 was applied from 170 to 350 °C. During the heating, nanorods were monitored in real time under active DF/BF TEM imaging modes or SAED modes, which were simultaneously video-recorded (Snagit software).
Quantification of T c based on the intensity change in TEM imaging contrast
Quantification of Tc for nanorods with varying diameters was carried out by a series of intensity profile measurements made over an image contrast variation. Image frames of the BF or DF TEM movies recorded during in-situ heating were aligned using an automated pixel-based plug-in application, TurboReg (Biomedical Imaging Group, Swiss Federal Institute of Technology, Lausanne, Switzerland)28 in ImageJ software (National Institutes of Health, Bethesda, MD)29. After alignment, two regions of interest (ROIs) were selected: one on the bright region of the sample (for BF/DF image movies) or on the first diffraction ring (for SAED pattern movies) and the other on the background. Intensity counts from the two ROIs were extracted from all image frames as a function of imaging time. The relative intensity profile for each nanorod was calculated as,
where Isample(background) is the intensity of ROI from the sample (background). At the onset of crystallization, the relative intensity increases rapidly (Fig. 2f). We define Tc at the intercept of the constant background and the linear fit to the increasing relative intensity.
How to cite this article: Sohn, S. et al. Nanoscale Size Effects in Crystallization of Metallic Glass Nanorods. Nat. Commun. 6:8157 doi: 10.1038/ncomms9157 (2015).
This work was supported by National Science Foundation MRSEC DMR 1119826. Microscopy facilities used in this work were supported by the Yale Institute for Nanoscience and Quantum Engineering (YINQE) and MRSEC DMR 1119826.
In-situ electron diffraction movie of a 150 nm-diameter nanorod.
In-situ electron diffraction movie of a 38 nm-diameter nanorod.
In-situ dark field image movie of a tapered nanorod (3-20 nm).
In-situ dark field image movie of a 17 nm-diameter nanorod with a narrow region (~9 nm).
This work is licensed under a Creative Commons Attribution 4.0 International License. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in the credit line; if the material is not included under the Creative Commons license, users will need to obtain permission from the license holder to reproduce the material. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/