In-situ STEM imaging of growth and phase change of individual CuAlX precipitates in Al alloy

Age-hardening in Al alloys has been used for over a century to improve its mechanical properties. However, the lack of direct observation limits our understanding of the dynamic nature of the evolution of nanoprecipitates during age-hardening. Using in-situ (scanning) transmission electron microscopy (S/TEM) while heating an Al-Cu alloy, we were able to follow the growth of individual nanoprecipitates at atomic scale. The heat treatments carried out at 140, 160, 180 and 200 °C reveal a temperature dependence on the kinetics of precipitation and three kinds of interactions of nano-precipitates. These are precipitate-matrix, precipitate-dislocation, and precipitate-precipitate interactions. The diffusion of Cu and Al during these interactions, results in diffusion-controlled individual precipitate growth, an accelerated growth when interactions with dislocations occur and a size dependent precipitate-precipitate interaction: growth and shrinkage. Precipitates can grow and shrink at opposite ends at the same time resulting in an effective displacement. Furthermore, the evolution of the crystal structure within an individual nanoprecipiate, specifically the mechanism of formation of the strengthening phase, θ′, during heat-treatment is elucidated by following the same precipitate through its intermediate stages for the first time using in-situ S/TEM studies.


Alloy material and age hardening treatment on bulk samples:
An Al-Cu 5.7 wt. % alloy was used in this study. The as-received materials were cold-rolled sheets with a thickness of 1 mm. A sheet was then cut into pieces of 10×10×1 mm 3 . These pieces were solution treated at 520 °C for 1 h in an oil bath and were water-quenched to room temperature and next annealed at 120-200 °C.
Vickers micro-hardness tests were performed with a digital hardness tester at a load of 4.9 N with a dwell time of 10 s. Fig. S1 shows the hardness-time curves of the samples aged at 160 °C and 180 °C. The alloy shows a typical strong age hardening. Though the ageing kinetics are accelerated significantly at higher temperatures, similar trend at various temperatures are observed: the hardness increases rapidly at the beginning, it continues to increase gradually after a slight softening until the peak hardness is reached, and then a plateau is formed. The rapid hardness and softening normally correspond to the formation and dissolution of clusters or GP zones 1,2 , respectively. The gradual hardness increase represents the nucleation and growth of the effective strengthening phase, θ', in this case 3 , while the plateau indicates the microstructure has reached the steady state. Figure S1:The evolution of hardness as a function of time during artificial ageing. The plot indicates the peak hardness, 135~140 HV, is much higher than the as-quenched state (~75 HV). This considerable strength increment results from the precipitates formed during artificial ageing.

More details on in-situ STEM studies and Image Processing:
The in-situ TEM studies were carried out using a DENSsolutions wildfire D6 double-tilt TEM heating holder. The specimen preparation has been carried out using the standard protocols and scripts followed while using an FEI Strata Dual-Beam 235 SEM-FIB. As this instrument is not equipped with an in-situ micromanipulator, the as-prepared thin TEM lamellae were transferred onto the heating chip using a pneumatic micromanipulator with glass-capillary tips under an optical microscope. The dimensions of the specimen were 15 µm × 6 µm × 200 nm. The specimen thickness was intentionally 200 nm since it should be more than the mean-diffusion length, which is around 160 nm for Cu in Al matrix at 200 °C for a heating time of 24 hours. Fig. S2 shows the picture of Pt heating spiral, embedded in SiN membrane and the lamellar TEM specimens suspended over holes in the SiN membrane.

STEM) imaging
In HAADF-STEM imaging mode, the intensity scales approximately with Z 24,5 . As Cu (Z=27) has a higher atomic number than Al (Z=13), the precipitates containing a lot of Cu solutes appear with higher intensity in HAADF-STEM images in comparison with the Al matrix. The raw atomic-resolution images shown in the paper (Fig. S3a) were processed by applying a mask in the Fourier transform of the original images. This method reduces the noise without introducing artificial information as shown in Fig. S3b.  Fig. 4c; actually, this image is acquired by conducting inverse FFT on the masked FFT pattern in (a). Scale bar, 1 nm.

Data analysis and statistics using Image J
The process used to acquire the length of one of the precipitates (marked P1 in Fig. 1e) from in-situ STEM movie based on the method used in reference 21 of the main text is described in this section. Fig. S4a shows an ADF-STEM frame taken from the end of the movie S2. In order to enhance the contrast of the edges of the precipitate, the image stack is first duplicated and processed using a Gaussian blur (sigma radius 4.00). Then the original image stack is divided by the processed image stack to produce a stack of images with sharp precipitate edges. Subsequently, an orthogonal image (through the image series) with the time series on the X-axis and the precipitate length on the Y-axis from each of the precipitates is procured.    Fig. 1h were acquired in this way. To determine the length of the precipitate, first a bare outline map, Fig. S4d, is extracted from the 'Analyse Particles' plug-in of ImageJ using an appropriate threshold level. Then, all pixels outside the particle growth trajectory are set to 0 and the pixels inside are set to 1 to generate a bitmap as shown in Fig. S4e. The length of the precipitates is then calculated by adding the pixels along the vertical axis and multiplying with the pixel size of the ADF-STEM images.

Growth kinetics of the precipitates at different temperatures
From movies S1-S3, the changes in lengths of a few precipitates with ageing time were analysed, as shown in Fig. S5. The data of precipitates formed during ageing at 200 °Care not given here because their relatively large sizes could imply intersection with the surface. From the 3-D reconstruction (see below) based on STEM-tomography, the thickness of the TEM sample was determined to be about 220 nm, which was smaller than most of the precipitates formed in the sample aged at 200 °C.

Mechanism controlling precipitate growth
There exist several theories on the precipitate coarsening and most notable are the ones based on diffusion of the alloying elements 6 . Recently, models based on a combination of mesoscale phase-field method with atomistic approaches have been developed too, predicting the morphology of θ' precipitates in Al-Cu alloys 7,8 . In this study, we do not validate the model for morphology based on phase-field approach, however, we observe that the precipitate growth is diffusion-controlled.
The growth trajectories for twenty random precipitates in the field of view were extracted from Movies S1-S3, as shown in Fig. 1 and Fig. S5. For each of the precipitates, by synchronizing time of nucleation as t = 0, the length of a precipitate has been plotted as a function of t ½ . Seventeen precipitates show a linear relationship between precipitate length and t ½ as shown in Fig. S6a and Fig. S6b.These follow the volume-diffusion controlled growth of precipitates in metallic alloys 9,10 described by the relation: where the β is, a parameter related with the super-saturation of solutes, D is diffusion coefficient.
The slope values (A) for length change curves of different precipitates at various temperatures were 7 determined by fitting the data from the previous section. The value ranges are shown in Table S1.
Obviously, the precipitates grow faster with increasing ageing temperature.
A small fraction of precipitates (3 out of 20) show a length change linear with t ( Fig. S6c) following a relationship as: From the morphology of these precipitates and from the location determined from tomography studies, we conclude that these precipitates grow faster as they are located near the surface, also as observed by Ferrante & Doherty 9 . This kind of growth was observed to occur at higher ageing temperatures (typically beyond 180°C).The exception is when a precipitate interacts with a dislocation as shown by precipitate P3 in Fig. 1, the precipitate growth is assisted by volume-diffusion, till the precipitate interacts with the dislocation after which its growth rate is accelerated as shown in Fig. 1h

Shrinking kinetics and process
A remarkable observation in this study is the relatively fast shrinkage of one of the precipitates when it comes in contact with a larger growing precipitate in the vicinity, as shown in Fig. S7 a-e. Fig. S7 a-e shows the as-acquired HR-STEM images revealing the shrinkage of plate-shaped precipitates II in Fig. 3.
It is evident that the thickness remains unchanged during the entire process of shrinking while the length reduces significantly. The area of intersection is brighter than either of the two precipitates, indicating an overlap in the projection direction. A proposed position of precipitates I and II is illustrated in Fig. S7f The initial structure of the precipitate is composed of alternating Cu and Al planes, with the intensity of the former brighter than the latter in HR-STEM image, as proved by the grey level intensity analysis of the lattice planes in the precipitate shown in (T-T p ) · M + T p · P = I p (4) where T p is the length of the edge-on precipitate along the thickness direction.
Combining equation (3) and (4)   alternative Cu and Al planes, whose intensity ratio is larger than one. Therefore, the high ratio value thus indicates the precipitate keeps its original structure before disappearance.

Density functional theory calculation of precipitate structures in Al-Cu alloys
First-principles calculations were performed to investigate phase stability of precipitate structures formed at various stages. Supercells shown in Fig. S10 were used to model the precipitate embedded in the Al matrix. The calculations were performed using the plane-wave based Vienna ab initio simulation package (VASP) [11][12][13] . The interaction between ion and core electrons was described by the projector augmented wave (PAW) method, and plane waves with an energy cut-off of 250 eV are used to expand the Kohn-Sham(KS) wave functions. The generalized gradient approximation (GGA) for the exchange and correlation functional is employed with PW91 scheme. We used 15×15×10 k-points mesh generated by the Monkhorst-Pack scheme. With these input parameters (e.g., energy cut-off, k-point sampling and supercell size), the formation energies converge to better than 10 meV.
The formation enthalpy, ΔH, is defined as the difference between energies of a Cu-enriched precipitate structure and the isolated impurities. In the present work, we define the formation enthalpy with respect to the bulk Al and Cu atom dissolved in Al matrix as The enthalpy of a substitutional Cu in a relaxed Al supercell is given by (7) The reliability of this method was established by confirming from some of the existing precipitate structures. Low enthalpy and low mismatch with matrix indicate favourable structure. All the optimized crystal lattice parameters except a of transient stage (shift) are smaller than corresponding Al lattice parameters, as shown in Table S2. This indicates a lattice contraction when Cu is enriched at a specific site to form precipitate.

Three-dimensional reconstruction of the precipitate structure
STEM tomography was performed to obtain three-dimensional (3-D) information of the precipitates formed in the FIB lamella. The specimen was tilted over a range of ±45ºand ADF-STEM images (2-D projections) were acquired in steps of 1º. In this study, the discrete algebraic reconstruction technique (DART) 14 was employed to reconstruct the 3-D map of the precipitates from the tilt-series.
The main purpose of this reconstruction was not only obtaining a 3-D distribution of the precipitates, but also to eliminate the influence of the surface such that the growth kinetics of the plate-like precipitates can be extended to that of a bulk specimen.

Compositional analysis of the precipitates:
Although a high-purity alloy was used for this study, we carried out EDX mapping to analyse the composition of the precipitates. The STEM images and the corresponding EDX maps were acquired on an FEI TITAN operated at 300 kV equipped with an Oxford Instruments X-Max detector. The specimen was heat-treated in-situ at 160 °C for 8 h to grow the plate-type precipitates. Owing to the thick specimens (200 nm) used in this study and the limitation of the EDX detector, it was only possible to obtain low-resolution EDX maps and the STEM images and corresponding maps are shown in Fig. S14.
As can be seen from the Fig. S14, all of the plate-like precipitates are enriched in Cu, indicating q' type precipitates.

Electron-beam heating
Energy transfer from the electron beam to the specimen due to inelastic scattering results for an important part into heat generation. This heat production is balanced in a steady state by heat conductivity to the specimen surrounded by the irradiated area and heat radiation to the environment from both specimen surfaces. The latter term can be neglected according to Egerton in reference 18. The upper limit of the temperature increase can be estimated from the energy input by the electron beam and the heat loss due to conductivity. In reference 18, the temperature increase in a carbon foil exposed to a with a heat conductivity more than 100 times that in carbon is expected to generate a much lower temperature increase. In addition, the electron beam is not stationary, resulting in even less heat production. The temperature increase under irradiation was also calculated from the similar Fisher's model 15 where q is the deflected angle of the electron in the field of the atom nucleus, E max is the maximum energy transferred, E 0 is the incident electron energy (in eV), and A is the atomic mass number.
According to the equations, displacement energy and the corresponding threshold value of the incident energy for Al are 17 eV and 180 Kev, while 20 eV and 420 keV for Cu 19 .
Therefore, knock-on damage of Al could occur in our case. However, at high temperatures the atoms in the Al sample have a high recovery rate. Thus, the knock-on damage could be significantly reduced when the ADF-STEM images are obtained on the TEM specimen heated at the ageing temperatures.

Experimental results regarding the effect of the electron-irradiation on the precipitation
To reduce the effect of electron beam, we either lower the electron current density (dose rate) or shorten the dwell time until there is no appreciable difference between the precipitates formed under the electron beam and those formed without electron beam effect (the only stimulus is heating), as shown in Fig. S15. We made a comparison of the precipitates formed in the irradiation area with those formed 18 in other places. No appreciable difference was found for the samples aged in the temperature range we studied. For recording the time series of the individual precipitate in Fig. 4 at atomic-resolution, the electron beam was therefore blocked during the intervals (several hours) between recording each image to eliminate the electron radiation effect. For the low-magnification movies showing the morphological change, the electron beam effect was not observed in our in-situ experiment. And the difference of precipitation behaviours revealed in our results should be caused by the temperature difference, as the imaging conditions were the same. with that free of electron beam effect (the region outside the red box area). The sample was heated for 5 h at 160 ℃with the area (Movie S5), marked by the red box, under continuous scanning by electron beam.

Supplementary movie captions:
• Supplementary Movie S1: Movie showing the nucleation and growth of precipitates in a FIB specimen of Al-Cu alloy heated at 140°C. Each frame is an ADF-STEM image recorded with the electron beam parallel to <001> Al orientation. The time (in hours/minutes/seconds) for which the specimen is held at 140 °C immediately after quenching, is shown on the upper-left corner.