Introduction

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.

Results

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.

Figure 1: Phase transition of MG nanorods under in-situ TEM.
figure 1

(a) TEM image of MG nanorods pressed into a nanomold with 55-nm-diameter holes. They are drop casts on a thin ceramic film, which gets heated inside TEM. (b) Schematics for the morphological change of MG nanorods at different stages of in-situ heating. As-pressed nanorods exhibit uniform TEM image contrast (first panel). As temperature increases above crystallization temperature, Tc, crystalline grains start to grow in MG nanorods (dark brown in the second panel). The grains are identified by TEM imaging or electron diffraction. As temperature increases above the melting temperature, Tm, the crystallized nanorods turn to liquid phase (third panel). (c) TEM characterizations of a nanorod under in-situ heating. At 30 °C, the nanorod is amorphous, confirmed by the uniform TEM image contrast (top left: BF image, bottom left: DF image) and a diffusive ring pattern in SAED (right panel). Above Tc at 350 °C the nanorod is largely crystallized, reflected in contrast changes in the TEM images. The crystalline phase appears dark for BF (top left) and bright for DF (bottom left) TEM images, respectively. The corresponding SAED displays diffraction spots, indicating the presence of crystalline grains (right panel). At 700 °C, which is above Tm, uniform TEM image contrast and diffusive SAED ring similar to those at 30 °C are observed to indicate that the nanorod is back to disordered state. As heating progresses with longer time, the nanorod becomes significantly deformed. At 900 °C, the nanorod becomes greatly shortened and balled-up.

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.

Figure 2: In-situ characterization of the Tc of nanorods with varying diameters.
figure 2

(ae) Individual nanorods were heated inside the TEM at a constant ramping rate of 0.67 °C s−1 and their crystallization kinetics were monitored in real time. SAED and TEM imaging are used to identify the crystallization kinetics and Tc. For thicker nanorods (diameter >35 nm), Tc was determined by identifying the onset of the first SAED spot. For smaller nanorods (diameter <35 nm), Tc was determined by identifying the TEM image contrast change in BF and DF modes. (a,b) Snapshot SAED to compare the crystallization kinetics of nanorods of 150 and 38 nm diameters, respectively. The red circles indicate the onset of the first diffraction spot, reflecting the onset of crystallization. (c) Snapshot BF TEM images to compare the crystallization kinetics of nanorods of diameters <25 nm. The dotted yellow lines indicate crystalline grains, which form at a lower temperature for the thicker nanorod. (d) Snapshot DF TEM images to show the crystallization kinetics of a tapered nanorod, revealing that the formation of crystalline grains (bright contrast) initiates on the thicker region. (e) Snapshot DF TEM images to show the crystallization kinetics of an extremely thin, 6-nm-diameter nanorod, showing that crystalline grains form at 340 °C, which is over 50 °C higher than the Tc of bulk MGs. (f) Quantification of Tc’s for nanorods of various diameters based on the intensity change of TEM image contrasts. The abrupt increase of the relative intensity in each trace (denoted by red arrows) reflects the advent of first crystalline grains, which is defined as Tc (see Methods). The decrease of the intensity after Tc is attributed to the dynamic motion of grains under heating, so initial diffraction spots do not last.

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.

Figure 3: Diameter-dependent crystallization kinetics of MG nanorods.
figure 3

(a) Correlation of Tc vs. nanorod diameter. All data were collected from non-tapered nanorods of uniform diameters. The uncertainty in Tc is calculated from the standard deviation of the temperature fluctuations during heating. (b-d) Schematics to illustrate the effects of surface-induced heterogeneous nucleation (b), apparent viscosity (c) and reduced nucleation probability (d) on crystallization kinetics. Red curves show the combined effect of the size-dependent surface-induced heterogeneous nucleation and apparent viscosity. The combined effect results in a size-dependent transition of the heterogeneous nucleation rate, while the growth rate monotonically decreases with decreasing diameter. This is due to the fact that diffusivity and the jumping frequency are inversely proportional to the viscosity. Black curves in b and c consider only the surface-enhanced nucleation. (e) The combined effect of size-dependent nucleation and growth rates results in a minimum of Tc at a certain diameter. The shaded area in red indicates the scatter in Tc’s, reflecting the stochastic nature of nucleation at the reduced nucleation probability.

Discussion

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).

Figure 4: Direct verification of growth rate difference in a nanorod.
figure 4

(a) Snapshot DF TEM images to observe the growth kinetics in a nanorod of 17 nm with a narrow region (9 nm, denoted by orange arrows) in diameter. The dotted line in the top image outlines the nanorod morphology. The grain growth rate is estimated by measuring the lateral expansion of the crystalline grain front, reflected as the change of imaging contrast (yellow dashed lines). Note that the growth front passes through the neck at 370 °C. (b) Quantification of the crystalline grain growth rate. The grain growth rate is suppressed at the narrow region of the nanorod during crystallization. The growth rate for the thicker region is 65 and 180 nm s−1 at 330 and 394 °C, respectively, and the rate for the thinner region is 4.5, 0.3 and 11 nm s−1, at 360, 370 and 380 °C, respectively. The detailed experimental condition is described in Supplementary Note 2.

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).

Figure 5: Statistical studies on Tc’s as a function of nanorod diameter.
figure 5

(a) The measured temperature scatter (ΔT=TTaverage), which is the difference between the crystallization temperatures (Tc) and the average Tc (Taverage), and (b) the corrected sample standard deviation of nanorods with a diameter of 10, 20, 50 and 120 nm, obtained from N number of nanorods. The scatter increases with decreasing nanorod diameter, suggesting that the reduced probability of nucleation plays a critical role.

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.

Methods

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 Tc 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.

Additional information

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).