Available online at www.sciencedirect.com

www.elsevier.com/locate/actamat

ScienceDirect

Acta Materialia 81 (2014) 173–183

In situ transmission electron microscopy and scanningtransmission electron microscopy studies of sintering

of Ag and Pt nanoparticles

M.A. Asoro a, P.J. Ferreira a,b, D. Kovar a,b,⇑

a Materials Science and Engineering Program, University of Texas at Austin, Austin, TX 78712, USAb Department of Mechanical Engineering, University of Texas at Austin, Austin, TX 78712, USA

Received 19 May 2014; received in revised form 11 August 2014; accepted 13 August 2014Available online 7 September 2014

Abstract

Transmission electron microscopy and scanning transmission electron microscopy studies were conducted in situ on 2–5 nm Pt and10–40 nm Ag nanoparticles to study mechanisms for sintering and to measure relevant sintering kinetics in nanoscale particles. Sinteringbetween two separated particles was observed to initiate by either (1) diffusion of the particles on the sample support or (2) diffusion ofatoms or small clusters of atoms to the neck region between the two particles. After particle contact, the rate of sintering was controlledby atomic surface diffusivity. The surface diffusivity was determined as a function of particle size and temperature from experimentalmeasurements of the rate of neck growth of the particles. The surface diffusivities did not show a strong size effect for the range of particlesizes that were studied. The surface diffusivity for Pt nanoparticles exhibited the expected Arrhenius temperature dependence and did notappear to be sensitive to the presence of surface contaminants. In contrast, the surface diffusivity for Ag nanoparticles was affected by thepresence of impurities such as carbon. The diffusivities for Ag nanoparticles were consistent with previous measurements of bulk surfacediffusivities for Ag in the presence of C, but were significantly slower than those obtained from pristine Ag.� 2014 Acta Materialia Inc. Published by Elsevier Ltd. All rights reserved.

Keywords: Microscopy; Diffusion; Kinetics; Coarsening; Densification

1. Introduction

Nanoparticles (NPs) possess unique properties stem-ming from their nanoscale dimensions and thus are of greatinterest in a wide variety of applications, such as textiles,renewable energy, the environment, health, electronicsand agriculture. However, due to their large surface-area-to-volume ratio and large curvature, nanoparticles have astrong tendency to coalesce and sinter during processingor usage over short time scales and at low temperatures,

http://dx.doi.org/10.1016/j.actamat.2014.08.028

1359-6454/� 2014 Acta Materialia Inc. Published by Elsevier Ltd. All rights r

⇑ Corresponding author. Tel.: +1 512 471 6271; fax: +1 512 471 7681.E-mail address: dekovar@austin.utexas.edu (D. Kovar).

which leads to significant changes in behavior andperformance. In some applications, such as in catalysis, sin-tering is detrimental because it decreases catalytic activity,but in other applications such as electrical interconnects,these changes are beneficial because they enhance conduc-tivity. In either case, understanding the mechanisms andmeasuring sintering kinetics at temperatures experiencedby NPs are critical for controlling material properties.

The influence of particle size and temperature on sinter-ing of micron-sized particles has been previously studiedand simple scaling models such as Herring’s law have beendeveloped to predict the influence of particle size on sinter-ing kinetics [1]. However, there are several complications inutilizing such predictions: (1) sintering models require

eserved.

174 M.A. Asoro et al. / Acta Materialia 81 (2014) 173–183

experimental data in the appropriate temperature andparticle size regimes. In the case of diffusion data, there isa paucity of data at low temperatures that are relevant toNPs. (2) Experimental studies on NPs often show discrep-ancies between these predictions made from data obtainedat high temperatures and for larger particle sizes [2,3]. (3)Theoretical models have traditionally been evaluated bycomparison with experimental observations of sinteringin powder ensembles containing a distribution of particlesizes that were obtained by postmortem characterization[4–10]. However, discrepancies between the experimentalobservations and the model predictions have caused debateabout the interpretation of indirect observations to studysintering processes [7,11].

The use of in situ transmission electron microscopy(TEM) techniques that are capable of monitoringdynamic processes in individual nanoparticles in realtime can resolve the uncertainties that arise from post-mortem experiments. Previous in situ TEM experiments[12–24] used a miniature furnace with a heating coil toindirectly heat the sample, while temperature was mea-sured with an attached thermocouple. Indirect heatingresults in substantial thermal drift, making it difficultto perform these experiments at high magnifications.Furthermore, it typically takes several minutes for thistype of furnace to ramp to the observation temperature,and thus sintering occurs during heating and before thetemperature has stabilized. Finally, some sintering exper-iments have been performed using a focused electronbeam in the transmission electron microscope to directlyheat the sample. For these experiments, the temperatureswere estimated from the beam current density and irradi-ation time. However, we have recently shown that tem-peratures cannot be accurately determined forindividual nanoparticles from these experiments unlessthe contact angle between a specific particle and sub-strate is known [25]. Together these effects make it verychallenging to perform quantitative sintering studies onsmall nanoparticles.

To address the aforementioned issues, we use a novelheating stage that is capable of rapidly heating the sam-ple with minimal thermal drift so that high resolutionimaging can be performed. In this study, we focus ontwo face-centered cubic (fcc) metals, Ag and Pt, andcompare their sintering behaviors. Two types of sampleswere studied: (1) commercial NPs that contain ubiqui-tous carbon on their surfaces; and (2) specially preparedNPs produced in our laboratory that did not haveorganics intentionally added on the particle surface.The influence of particle size, temperature and surfacecondition are studied by observing sintering in real timeusing TEM and scanning transmission electron micros-copy (STEM) and measuring the rates of sintering inthe two particle systems. From these data, the surfacediffusivities are determined and compared to previousmeasurements made on bulk materials.

2. Experimental procedure

2.1. Materials and sample preparation

Three types of samples prepared from commerciallyavailable nanoparticles were studied. Ag NPs with nominalmean sizes of 15 and 40 nm were obtained from Nanotech-nologies Inc. (Austin, TX). These NPs were synthesizedusing a pulsed-plasma, dry synthesis method that resultedin carbon deposited on the surfaces of the Ag NPs.Another set of samples consisted of Pt NPs (Johnson Mat-they Technology Centre, Oxfordshire, UK) with a nominalsize of either 2 nm or 6 nm that were deposited onto astrongly adhering particulate carbon support (Ketjen,Tokyo, Japan). A third set of samples consisted of PtNPs with a mean size of 2.8 nm (Tanaka KikinzokuKogyo, Tokyo, Japan) that were deposited onto weaklyadhering carbon (Vulcan, Cabot, Boston, MA) particles.TEM samples were prepared by first dispersing the NPsamples in ethanol and then ultrasonicating for 10 min toreduce particle agglomeration. A drop of the suspensionwas then deposited onto the TEM heater chip and theliquid was evaporated.

2.2. In situ TEM/STEM heating and characterization

In situ heating experiments were performed in both con-ventional bright field (BF) TEM mode using phase contrastimaging and in high angle annular dark field (HAADF)scanning transmission electron microscopy (STEM) modeusing Z-contrast imaging. The TEM experiments were per-formed in a JEOL 2010F TEM operated at 200 keV, whilethe STEM experiments were conducted using a JEOL2200FS STEM/TEM ora JEOL JEM-ARM 200F, bothoperated at 200 keV. Compared to conventional phasecontrast imaging in a TEM, the aberration-correctedSTEM offers superior resolution, allowing imaging of verysmall NPs, atomic clusters and even single atoms. TheHAADF images were obtained from a probe with a con-vergence angle of 25 mrad, and electrons scattered atangles ranging from 100 to 170 mrad. As a result theHAADF images contain nearly zero diffraction contrastand provide essentially mass-thickness contrast. Thus,HAADF imaging is a powerful technique for observingthe coalescence process, particularly in the neck regionbetween sintering particles, where there can be significantvariations in thickness.

The NPs were heated in situ using an Aduroe heatingstage (Protochips Inc., Raleigh, NC) [26]. This heatingstage uses a disposable micro-electro-mechanical system(MEMS) device that serves both as the specimen supportgrid and the heating element with an electrical feed-through that connects to an external power supply. Thedevice consists of a 150 nm thick, 500 lm � 500 lm, free-standing membrane made from a conductive ceramicmaterial that is suspended over a 4 mm � 6 mm silicon

M.A. Asoro et al. / Acta Materialia 81 (2014) 173–183 175

chip. For electron transparency, the ceramic membrane ispatterned with a series of 6 lm diameter holes, which aresubsequently overlaid with a holey carbon film which sup-ports the NPs. Joule heating occurs when current is forcedthrough the ceramic membrane and is used to control thetemperature. The current vs. temperature response of eachdevice is calibrated at the factory using an imaging pyrom-eter in a vacuum probe station at a pressure similar to whatis used in a TEM column. Unlike conventional heatingholders, this heating stage is capable of very fast heatingrates up to 106 �C s�1 with extremely low drift, even at hightemperatures. This allows isothermal sintering experimentsto be carried out in the TEM, since heating to the desiredsintering temperature can be achieved nearly instanta-neously. The in situ heating experiments were performedat temperatures from 25 to 500 �C in order to investigatethe effects of size and temperature on sintering of NPs.The size and temperatures for each experiment were chosensuch that the sintering process could be observed in situwithin a reasonable time frame.

The accuracy of the temperature measurements for theas-received heater chips was assessed using a knownsolid-to-vapor phase transformation. Details of this proce-dure are presented elsewhere [25]. Our experimentsrevealed temperature discrepancies of between 20 and150 �C between the apparent and actual temperatures thatwere attributed primarily to beam heating of the particles.The large range in the temperature discrepancies is believedto arise from variations in the contact area between indi-vidual particles and the TEM heater substrate, which leadsto point-to-point variations in heat transfer rates betweenthe NPs and the TEM substrate. The beam heating effectsand point-to-point variations in temperature results in sig-nificant but quantifiable error bars on our temperaturemeasurements during the in situ heating experiments.

2.3. Determination of sintering parameters

During the early stages of sintering, neck growth occursto reduce the large chemical potential at the particle con-tact points where the neck radius is very small. Possiblemass transport mechanisms for neck growth include grainboundary diffusion, surface diffusion and lattice diffusion,as well as plastic deformation. For NPs, it is unlikely thatdislocation-driven plastic flow would contribute signifi-cantly to neck growth in fcc NPs given the large stressesrequired for plastic flow [27]. Surface diffusion is the dom-inant mass transfer mechanism at lower temperatures, asshown by the experiments of Kuczynski [28], but this isespecially so for small particles that have larger surface-atom-to-volume atom ratios. The surface diffusivities forfcc metals at low temperatures are typically many ordersof magnitude larger than the other diffusivities [29].

For surface diffusion, the change in neck radius, x withtime, t for a given particle radius, a and temperature, T isgiven by [28,30]

X 7 ¼ 56Xa3csDsdstkT

ð1Þ

where X is the atomic volume, cs is the surface energy, Ds isthe surface diffusivity, ds is the surface diffusive width and k

is Boltzmann’s constant. From our sintering experimentsand using Eq. (1), Ds for NPs was determined by measuringx and a as a function of time at a given temperature, sinceall of the other parameters are known.

The measurements of the particle geometry relevant tosintering were performed using DigitalMicrographe soft-ware (version 1.71.38, Gatan Inc. Pleasanton, CA). NPswere selected, such that they were nearly spherical (i.e.the major and minor axes were within 5% of each other),so that a two-particle sintering model could be used todetermine the sintering parameters. To measure a, theedges of the NPs were first detected and a circle was fitto each NP using a Matlab� routine [31] that computed aleast-squares fit for a circle with a set of (x,y) coordinatesselected on the edges of the NP. The area of the fitted circlewas used to obtain an equivalent particle radius for aspherical NP. The neck radius was calculated from one-half the length of a line drawn perpendicular to the inter-section of the two particles. Examples of measurementsof a and x are shown in Fig. 1.

3. Results

3.1. Sintering between two NPs

Fig. 2 shows a representative sequence of TEM imagestaken at 400 �C that shows sintering between two 40 nmNPs that were produced by combustion synthesis. As thetwo NPs contact each other, there is rapid growth of theneck in the subsequent 3 min before the rate of neckgrowth slows over the subsequent 12 min. Note also thatthere is clearly some carbon present on the surface of theNPs that is most apparent in the first frame (indicated bythe arrows).

Contrast changes are also apparent in the NPs fromframe-to-frame. For example, the particle on the left,which is bright in the initial frame, darkens after 3 minand then brightens after 8 min. Contrast changes such asthese could be due to either rotations of the NPs or themovement of the carbon support on which the particlesare suspended, effectively rotating the particles with respectto the electron beam. In addition, we see unusually brightcontrast around the edges of the NP on the left. Such con-trast has been attributed either to the presence of a liquidphase on the particle surface or variations in thicknessaround the edge of the sample [32]. An analysis of the melt-ing temperature as a function of particle size suggests thatthe melting temperature for these particles is at least 940 �C[33]. Since the temperature at which this experiment wasconducted was well below the melting temperature of theAg NPs, the contrast changes must be due to thicknessvariations.

Fig. 1. TEM images of Ag NPs showing examples of measurements of (a) particle radius and (b) neck radius.

Fig. 2. A sequence of in situ TEM heating images showing sintering of two 40 nm Ag NPs at 400 �C. The arrows in the first frame show carbon present onthe surface of the NPs.

176 M.A. Asoro et al. / Acta Materialia 81 (2014) 173–183

Another sequence of in situ TEM images, presented inFig. 3, shows a second type of behavior that was observedin some of the sintering experiments where the onset of sin-tering was activated by a third particle. In this case, thesequence was taken at 200 �C and shows two 15 nm silverNPs that contain twins, which are common for Ag particlesin this size range. The particles are initially separated by asmall distance before contact between the NPs is initiatedby the motion of a third particle. At this point, sinteringoccurs by rapid neck growth. As the dihedral angleapproaches the equilibrium value of 161� [34] after�15 min, motion of the grain boundary between the

particles begins. In addition, the smaller NP at the bottomof the frame appears to move towards the two NPs sinter-ing in the middle of the frame and eventually coalesces withthe larger NPs after some time.

Contrast changes are also apparent in the NPs shown inFig. 3 from frame-to-frame. To further study the cause ofthese changes, fast Fourier transforms (FFTs) were takenfor each of the images shown in Fig. 3. The diffraction dataobtained from the FFTs are shown in Fig. 4. This sequenceshows that the FFT patterns change from frame-to-frame,confirming that particle rotation occurs during sintering.For example, the particle goes from a near two-beam

Fig. 3. A sequence of in situ TEM heating images showing sintering of two 15 nm Ag NPs at 200 �C.

M.A. Asoro et al. / Acta Materialia 81 (2014) 173–183 177

condition at 13 min to a near zone axis condition at 17 min.These rotations are on the order of a few degrees since theyare all in the same zone axis.

3.2. Sintering between many NPs

It is important to determine if the mechanisms for sin-tering that were observed in two particle systems are alsoactive for the more realistic case of sintering between manyNPs. Fig. 5 shows a sequence of in situ TEM images takenat 100 �C of multiple Pt NPs with a mean size of 6 nm. Theparticles have been labeled A–D. Particles B and C initiallyappear to be in contact with one another while particles Aand D are isolated from the other particles. After 5 min,particle A migrates towards particle B and particle Dmigrates towards particle C. The particles were monitoredfor 16 min and very little neck growth was observedbetween particles A and B after the initial contact. ParticlesC and D move towards each other but do not appear tocontact. After 9 min particles B and C appear to moveapart without sintering.

There are at least two possible explanations for theobserved behaviors: (1) the NPs may be mobile on the sup-

port and coalesce only when they make contact or, if theyare not in the same plane, they may simply move past eachother with their projections appearing to overlap; (2) theentire carbon support on which the particles are suspendedmay move due to differential heating. To distinguish whichof these phenomena are responsible for the observedbehavior, HAADF STEM imaging was employed. Fig. 6is a sequence of HAADF STEM images showing sinteringof 2 nm platinum NPs at 500 �C that are labeled 1–11.There is clearly relative motion between the particles, asseen from the changes in positions of the particles withtime. For example, particles 5 and 6 move toward eachother and sinter while particles 9 and 10 drift apart. It isnot likely that substrate motion is responsible for theapparent relative motion between particles since theirmotion appears to be random. Motion of the carbon sup-port would cause correlated motions between the particles,unless the magnitudes of the perturbations of the surfacewere of the order of the size of the NPs, which is not plau-sible given the size scales observed here.

It is also interesting that particles 9 and 11 in Fig. 6 dis-appear from the support with little or no motion on thesupport. Wen et al. have postulated that Ag NP mass

Fig. 4. A sequence of FFTs taken from the NP on the left in Fig. 3, showing that the diffraction spots change from frame to frame, indicating that particlerotations of a few degrees occur during sintering.

178 M.A. Asoro et al. / Acta Materialia 81 (2014) 173–183

transport can occur by evaporation–condensation, whereAg atoms evaporate into the vapor and the re-condenseon another nanoparticle [35,36]. The mechanism for thechange in particle size was investigated further by perform-ing additional higher resolution in situ HAADF STEMexperiments on 2 nm Pt NPs at 300 �C (Fig. 7). The NPsare initially completely separated from each other. How-ever, after �20 min they make contact and sinter via neckgrowth. We observe that the mass transport does not occurby evaporation, but rather by surface diffusion along thecarbon support. At this scale, single atoms and small clus-ters of Pt are clearly visible between the NPs and theyappear to be mobile on the substrate. This clustering ofatoms between the particles leads to the formation ofbridges between the NPs and, once contact is made, theneck begins to grow. The source of these atoms is likelysmall clusters or even individual atoms that are depositedonto the support surface and then diffuse due to thermalactivation from both the heating stage and the electronbeam. The atoms and clusters appear to be mobile on thesurface until they condense onto the surface of an existingparticle. This dissolution–condensation mechanism is anal-

ogous to an Ostwald ripening process that occurs in liq-uids. Note, however, that we do not see the expectedbehavior of larger NPs growing larger that is expected fromtraditional Ostwald ripening. Rather, sintering appears tobe favored for larger NPs that are mobile on the carbonsupport, while the Ostwald-ripening-like process appearsto be favored for smaller NPs that are immobile on the car-bon support. These experiments indicate that for manyparticle systems, ripening as well as sintering may occursimultaneously. Lastly, a comparison of the results at 100and 500 �C suggest that atomic/cluster surface diffusionand ripening are more likely to occur at higher tempera-tures. These in situ experiments reveal a mechanism for sin-tering of NPs where atoms and small clusters present onthe substrate can play a major role in initiating sinteringof NPs.

4. Analyses of sintering data

A number of in situ experiments similar to those shownin Figs. 1–7 were conducted with Ag and Pt particles with arange of particle sizes and at a range of temperatures. The

Fig. 5. A sequence of in situ TEM heating images of 6 nm Pt NPs on a carbon support at 100 �C, showing particle migration and/or motion of the carbonsupport.

Fig. 6. In situ STEM images of 2 nm Pt NPs at 500 �C, showing that sintering can occur by both particle migration and coalescence as well as Ostwaldripening.

M.A. Asoro et al. / Acta Materialia 81 (2014) 173–183 179

Fig. 7. In situ STEM images of 2 nm Pt NPs at 300 �C, showing that initial contact between NPs during sintering can be achieved by migration of singleatoms and/or small clusters on the substrate towards the neck region.

180 M.A. Asoro et al. / Acta Materialia 81 (2014) 173–183

measured values of neck size with time were measured fromthese experiments and used to calculate the surface diffusiv-ity using Eq. (1). The surface diffusion coefficients, Ds, forsilver NPs with size ranges from 12 to 40 nm and at temper-atures of 200–400 �C are shown in Fig. 8. The values of Ds

were found to be in the range 4.2 � 10�16–1.1 � 10�20

cm2 s�1. Given the error bars in measuring temperature,there does not appear to be a statistically significant NPsize effect on surface diffusivity for NPs in the size rangesthat were measured.

Fig. 8 also shows previously reported values of surfacediffusivities for bulk silver and thin films of silver obtainedunder different conditions and at different temperatures.The most comprehensive surface diffusivity data for Aghave been obtained from bulk measurements at tempera-tures above 0.5Tm [37] because the kinetics are too slowto observe experimentally using this technique at lowertemperatures. Thus, to compare bulk measurements tothe current results, it is necessary to extrapolate the bulkDs to temperatures below 0.5Tm using an Arrhenius rela-tionship (shown as a dashed line in Fig. 8). It is recognized

that such extrapolations may not be valid because the fun-damental mechanisms of diffusion can change with temper-ature (as occurs in bulk Ag at �500 �C). In addition, smalluncertainties in the high temperature data can lead to largeerrors when extrapolated to low temperatures. Althoughcaution is warranted when making quantitative compari-sons between data with large uncertainties, it is neverthe-less clear that the measured surface diffusivities from thisstudy are several orders of magnitude lower than previ-ously reported bulk values.

Surface diffusivities have also been measured previouslyfrom the motion of atomic clusters of Ag using scanningtunneling microscopy (STM) in ultra-high vacuum for clus-ters deposited on thin films of silver. These room-tempera-ture diffusivities for clusters range from 10�15 to10�18 cm2 s�1) [35,36,38]. Pai et al. have suggested thatthe diffusivities for clusters follow a power law with respectto the cluster size with an exponent of 2.28 for Ag [38].Using their data and extrapolating to a single atom ofAg, we obtain Ds = 8.8 � 10�20 cm2 s�1. This is signifi-cantly lower than the values obtained from bulk silver at

1.2 1.6 2.0 2.4 2.8 3.2 3.6 4.0 4.4

-20

-18

-16

-14

-12

-10

-8

-6

-4

log

D s (cm

2 /s)

Tm/T

12 nm15 nm17 nm40 nmPai-clusterWen-clusterPai-atomicGuy (Bulk Ag)Offermann (Ag/C)

900 700 500 400 300 200 100 25

Temperature (C)°

Fig. 8. Plot of surface diffusivity vs. temperature for Ag. The filledsymbols are for the current results. The open symbols are previouslyreported values obtained in ultra-high vacuum for clusters of Ag [35,36,38]and the value calculated from these data [38] for atomic diffusion of Ag[38]. The solid lines represents estimated values of surface diffusivities forbulk silver (red) [29] and Ag on carbon (blue) [44] at high temperaturesand the dashed lines are the corresponding extrapolations of these data tolow temperatures. (For interpretation of the references to color in thisfigure legend, the reader is referred to the web version of this article.)

M.A. Asoro et al. / Acta Materialia 81 (2014) 173–183 181

high temperatures and extrapolated to room temperature(10�13 cm2 s�1). Comparing the values of Ds obtained inthis study to those obtained previously from STM mea-surement, the values from this study are somewhat lowerthan the values from the STM experiments.

Because of the uncertainty in our temperature measure-ments due to electron beam heating, it is useful to put abound on this uncertainty in the values of Ds that were

0 20 40 60 80 100 120 140 1600.0

0.2

0.4

0.6

0.8

1.0

1.2

Δlo

g D s (

cm2 /s

)

ΔT (°C)

200 °C300 °C400 °C

Fig. 9. Plot of DlogDs vs. DT, showing the effect of a change intemperature on surface diffusion coefficient.

obtained. Assuming an Arrhenius dependence on tempera-ture, the sensitivity of Ds to temperature can be estimatedby plotting Dlog Ds vs. DT for the ranges of temperatureused in our experiments (Fig. 9). For these calculations,we used the previously reported values of the diffusivityprefactor, Do (1.6 � 10�3 cm2 s�1), and activation energyfor surface diffusion, ED (0.4 eV), obtained from STM mea-surements on silver {100} surfaces [39]. For a temperatureerror of 150 �C, the resulting error in diffusivity is less thanan order of magnitude, which is relatively minor comparedto the variations in diffusivity that have been reported inthe literature. Thus, the temperature uncertainty due toelectron beam heating cannot explain the large discrepan-cies in the measured values of surface diffusivity for Ag.

A more likely cause of the large scatter in the bulk dif-fusivity data for Ag and for the discrepancies betweenthe data can be related to differences in sample surface con-ditions. The activation energy for surface diffusion, ED, forthin films of silver obtained from STM [40] and spot-profileanalysis of low energy electron diffraction [41] experimentsperformed in ultra-high vacuum, are significantly lowerthan the ED values obtained from resistivity measurementson thin, {111} oriented films of silver in air [42]. This sug-gests that the environment plays a major role in the sinter-ing process and that the presence of oxides or otheradsorbed species on the surface of the NPs can retard thesintering of the NPs. We have shown that the presence ofcarbon on the surfaces of Ag NPs can significantly retardsintering [43] and our experiments with Ag NPs in whichwe reduced the amount of carbon present on the Ag sur-faces showed a significant increase in the diffusivities. Forexample, for a 40 nm NP at 400 �C, the surface diffusivitywas determined to be 1.2 � 10�14–1.8 � 10�14 cm2 s�1,which is two orders of magnitude greater than the valuewe obtained for the same size particles at the same temper-ature, but where there was significant carbon surface coat-ing [43]. In STEM studies of diffusion Ag clusters, Wenet al. also found that the diffusivity of Ag was significantlyretarded in the presence of contaminants [35]. Thus, diffu-sivities obtained in the presence of surface carbon and/orother adsorbed species on the surface of Ag NPs are notintrinsic surface diffusivities, but rather are effective diffu-sivities. This is a very important point, since most commer-cially available NPs usually contain some form ofhydrocarbon added to prevent particle agglomeration.These hydrocarbons pyrolyze during sintering, resultingin carbon residue on the particle surfaces that can affectsubsequent sintering if it is not fully oxidized. These resultsare also consistent with measurements obtained from silverdeposited on pyrolytic carbon [44] that show significantlylower values of Ds than for pure bulk silver. The valuesof diffusivity in this Ag/C system (10�16–10�12 cm2 s�1 inthe temperature range of 450–800 �C), when extrapolatedto low temperatures, agree well with our measurements(Fig. 8).

The surface diffusion coefficient, Ds, for 2 nm Pt NPs atroom temperature was determined to be in the range

1 2 3 4 5 6 7 8-25

-20

-15

-10

-5Current resultsLinderothKawasakiMarcosBlakely (Bulk Pt)

log

D s (cm

2 /s)

Tm/T

1200 800 600 400 300 200 100 25

Temperature (°C)

Fig. 10. Plot of surface diffusivity vs. temperature for Pt. Data from theliterature taken from Refs. [45–48]. The dashed line is an extrapolation ofhigh temperature bulk Pt measurements to lower temperatures.

182 M.A. Asoro et al. / Acta Materialia 81 (2014) 173–183

1.5 � 10�20–6.5 � 10�20 cm2 s�1 from our measurementsobtained from sintering. Fig. 10 shows a comparison ofthese values with measurements from other experimentsat or near ambient temperature on 300 nm thick Pt nano-sheets on graphite (10�15–10�16 cm2 s�1) [45], Pt singleatoms on Pt {110} surfaces (3.16 � 10�19 cm2 s�1) [46]and columnar structured Pt electrodes (1.13 � 10�19 cm2

s�1) [47]. Also shown on this plot are data obtained frombulk Pt at high temperature [48], with the correspondingexponential extrapolation to room temperature. This plotshows that diffusivity values obtained in the current studyare in general agreement with the diffusivities obtainedfrom previous work. It is notable that unlike Ag, whichshows wide range in the values of Ds due to differences insurface condition, there is little discrepancy in the valuesof Ds for the more noble Pt, even when it is obtained usingdifferent surface conditions and techniques.

5. Conclusions

The sintering of silver and platinum NPs was studiedusing a novel heating holder for in situ TEM/STEM heat-ing experiments. Two mechanisms were observed for sin-tering in NPs: (1) a surface Ostwald-ripening-like processin which atoms or clusters dissolve onto the substrate sur-face and then are transported to and deposit onto anotherparticle, or (2) particle migration along the surface, contactbetween the particles and coalescence. The Ostwald ripen-ing process was found to be more prevalent at higher tem-peratures. For coalescence, it was found that contactbetween NPs can be initiated by (1) diffusion of the parti-cles on the support, (2) migration of single atoms and small

clusters on the support towards the neck region or (3)motion of the support due to differential heating causedby the electron beam. Atomic or cluster migration aremechanisms for activation of sintering between two largerparticles that have not been previously reported. Theseresults demonstrate that the diffusion of small clusters ofatoms on the support may affect the sintering of largerNPs by forming a bridge in between the NPs, which thenleads to subsequent neck growth.

The values of Ds (4.2 � 10�16–1.1 � 10�20 cm2 s�1)thatwere obtained from in situ sintering experiments on AgNPs were significantly lower than values extrapolated fromhigher temperatures and were also lower than the valuesobtained from thin Ag films measured in ultra-high vac-uum at room temperature using STM. Consistent with pre-vious measurements of surface diffusivity, our resultssuggest that surface species are a likely cause for this dis-crepancy because the presence of carbon on the surfaceof NPs can significantly inhibit sintering in NPs. The effec-tive diffusivities that were measured for Ag NPs with car-bon surface coatings match well with reported values ofdiffusivity of silver in pyrolytically deposited carbon.

In contrast to the results obtained for silver, the valuesof Ds measured for Pt are in close agreement with reportedbulk values extrapolated from high temperatures and mea-surements performed at room temperature. This is believedto be due to the fact that platinum, being less reactive thansilver, is less susceptible to contamination and also appearsto be less susceptible to impurity effects on surfacediffusion.

Electron beam heating effects result in a 20–150 �Cincrease in temperature over the temperature measuredwithout the beam. The large range in temperatures causedby beam heating results from variations in beam current,NP size and, most importantly, the contact area betweenthe NP and the substrate upon which the NP rests. Thisrange results in some uncertainty in the temperature duringour experiments and a corresponding uncertainty in thesurface diffusivity, which was determined to be at mostone order of magnitude. While this level of uncertainty isgreat enough that small variations in diffusivity cannot bedetermined using this technique, it is reasonably accuratecompared to the large range in measured surface diffusivi-ties that have been reported in the literature for silver.Within the accuracy of the measurements, we did notobserve a statistically significant effect of particle size onthe surface diffusivity of silver, which suggests that suchan effect is relatively small over the range of NP sizes thatwere measured (for Ag, 12–40 nm).

Acknowledgments

The authors would like to thank Dr Manuj Nahar forpreparing the Ag NPs that were made using LAMA thatwere used in this study and the National Science Founda-tion for supporting this work under DMR 1006894.

M.A. Asoro et al. / Acta Materialia 81 (2014) 173–183 183

References

[1] Herring C. J Appl Phys 1950;26:1205.[2] Yeadon M, Yang JC, Averback RS, Bullard JW, Gibson JM.

Nanostruct Mater 1998;10:731.[3] Groza JR. Int J Powd Metall 1999;35:59.[4] Wynblatt P, Gjostein NA. Prog Solid State Chem 1975;9:21.[5] Flynn PC, Wanke SE. J Catal 1974;34:400.[6] Ruckenstein E, Pulvermacher B. J Catal 1973;29:224.[7] Wanke SE. J Catal 1977;46:234.[8] Harris PJF. Int Mater Rev 1995;40:97.[9] Campbell CT, Parker SC, Starr DE. Science 2002;298:811.

[10] Sehested J, Gelten JAP, Remediakis IN, Bengaard H, Nørskov JK. JCatal 2004;223:432.

[11] Granqvist BCG, Buhrman RA. J Catal 1976;42:477.[12] Theissmann R, Fendrich M, Zinetullin R, Guenther G, Schierning G,

Wolf DE. Phys Rev B 2008;78:205413.[13] Rankin J, Sheldon BW. Mater Sci Eng A 1995;204:48.[14] Stone P, Poulston S, Bennett RA, Bowker M. Chem Commun

1998(13):1369. http://dx.doi.org/10.1039/A802881H.[15] Yang F, Chen MS, Goodman DW. J Phys Chem C 2008;113:254.[16] Heinemann K, Poppa H. Thin Solid Films 1976;33:237.[17] Anton R, Kreutzer P. Phys Rev B 2000;61:16077.[18] Liu RJ, Crozier PA, Smith CM, Hucul DA, Blackson J, Salaita G.

Appl Catal A 2005;282:111.[19] Simonsen SB, Chorkendorff I, Dahl S, Skoglundh M, Sehested J,

Helveg S. J Am Chem Soc 2010;132:7968.[20] Jose-Yacaman M, Gutierrez-Wing C, Miki M, Yang DQ, Piyakis

KN, Sacher E. J Phys Chem B 2005;109:9703.[21] Baker RTK, Harris PS, Thomas RB. Surf Sci 1974;46:311.[22] Chen Y, Palmer RE, Wilcoxon JP. Langmuir 2006;22:2851.[23] Zhang WJ, Miser DE. J Nanopart Res 2006;8:1027.[24] Hajmrle K, Angers R. Metall Trans 1974;5:817.[25] Asoro MA, Kovar D, Ferreira PJ. ACS Nano 2013;9:7844.

[26] Allard LF, Bigelow WC, Nackashi D, Damiano J, Mick SE. MicroscMicroanal 2008;14:792.

[27] Shan ZW, Mishra RK, Syed Asif SA, Warren OL, Minor AM. NatMater 2008;7:115.

[28] Kuczynski GC. Trans AIME 1949;185:169.[29] Guy AG. Essentials of materials science. New York: McGraw-Hill; 1976.[30] Kingery WD, Berg M. J Appl Phys 1955;26:1205.[31] Gander W, Golub GH, Strebel R. Bit 1994;34:558.[32] Williams DB, Carter CB. Transmission electron microscopy: a

textbook for materials science. New York: Springer; 2009.[33] Couchman PR, Jesser WA. Nature 1977;269:481.[34] Murr LE. Interfacial phenomena in metals and alloys. London: Addi-

son-Wesley; 1975.[35] Wen JM, Evans JW, Bartelt MC, Burnett JW, Thiel PA. Phys Rev

Lett 1994;73:2591.[36] Wen JM, Evans JW, Bartelt MC, Burnett JW, Thiel PA. Phys Rev

Lett 1996;76:652.[37] Rhead GE. Acta Metall 1965;13:223.[38] Pai WW, Swan AK, Zhang Z, Wendelken JF. Phys Rev Lett

1997;79:3210.[39] Antczak G, Ehrlich G. Surface diffusion: metals, metal atoms, and

clusters. New York: Cambridge University Press; 2010.[40] Brune H, Bromann K, Roder H, Kern K, Jacobsen J, Stolze P, et al.

Phys Rev B 1995;52:14380.[41] Luo EZ, Wollschl

p§ger J, Wegner F, Henzler M. Appl Phys A

1995;60:19.[42] Stark D. Surf Sci 1987;189:1111.[43] Asoro MA, Kovar D, Ferreira PJ. Chem Commun 2014;50:4835.[44] Offermann PJ. Nucl Mater 1977;64:249.[45] Kawasaki H, Sakai G, Kijima T. Appl Surf Sci 2006;253:1512.[46] Linderoth TR, Horch S, Laegsgaard E, Stensgaard I, Besenbacher F.

Phys Rev Lett 1997;78:4978.[47] Marcos ML, Gonzalez Velasco J. Chem Phys Lett 1998;283:391.[48] Blakely JM, Mykura H. Acta Metall 1962;10:565.