通过液相透射电子显微镜研究温度对纳米颗粒成核与生长的影响

摘要

在液相电子显微镜实验中进行温度控制，为研究纳米颗粒在模拟其形成或应用环境的液体条件下的动态行为提供了新的视角。利用近期开发的加热型液体池，我们直接观察了温度对金纳米颗粒在水中成核与生长过程的影响。

摘要

温度控制是近年来的一项新进展，为利用液体池透射电子显微镜研究纳米化学提供了额外的自由度。本文中，我们描述了如何制备原位加热实验，以研究温度对水中辐射还原驱动的金纳米颗粒形成过程的影响。该实验方案相对简单，涉及一种可均匀加热至100 °C的特殊液体池、一种具备液体流动功能的液体池透射电镜样品杆，以及一个集成化的温度控制系统接口。我们发现，金纳米颗粒的成核与生长机制在液体池中受到温度的显著影响。通过扫描透射电子显微成像（STEM）和纳米衍射技术，实时揭示了生长中纳米颗粒的密度、尺寸、形貌及原子结构的演化过程。利用自动化图像处理算法，从视频序列中提取出有用的定量数据，例如纳米颗粒的成核速率和生长速率。该方法为理解纳米材料液相合成过程中复杂的物理化学过程提供了新的依据。

引言

金属纳米颗粒（NPs）具有优异的物理化学性质，可广泛应用于光学传感¹、医学²和能源³等领域。湿化学合成法是一种高度通用的方法，可用于制备尺寸和形貌均一的金属纳米颗粒。在过去几十年中，研究人员已发展出多种策略以实现对纳米颗粒合成过程的调控，包括种子介导生长法⁴、晶面阻断法⁵、动力学控制合成法⁶、选择性刻蚀法⁷以及温度控制合成法⁸。然而，尽管驱动合成的化学反应本身相对简单，成核与生长机制却十分复杂，因为在纳米材料形成过程中有诸多参数共同作用，而仅通过在特定时间点从反应体系中取出样品进行离原位表征，难以准确解析各个参数的独立影响。为了真正理解成核与生长过程并建立有效的调控方法，我们必须采用原位表征技术，以实现在精确控制的液相环境中对这些过程进行实时观测。

在这一方面，液相透射电子显微镜（LCTEM）已成为揭示金属纳米颗粒合成过程的强有力方法^{9,10,11,12,13}。通过在液体反应环境中直接成像单个纳米结构的动态演化过程，该技术加深了人们对成核与生长机制的理解，特别是晶体缺陷、晶种形貌以及有机配体在调控定向生长或刻蚀过程中的作用，从而获得具有特定形貌的纳米材料（如纳米棒、纳米星、纳米片、纳米壳）^{10,11,12,13,14,15,16,17,18,19}。当透射电子显微镜的电子束与液体相互作用时，辐射分解过程会产生强还原性和氧化性物质，改变辐照区域内的溶液化学环境，进而可用于驱动生长或刻蚀过程。值得注意的是，辐射产物的浓度已知随电子剂量率的增加而升高，而电子显微镜中的该参数可实现精确调控²⁰。因此，研究人员已利用辐射分解对剂量率的依赖性来控制反应速率，并揭示其对纳米结构形成过程及最终形貌的动力学影响^11,15,20。

尽管温度是纳米材料合成中的一个关键参数，但由于具备可靠温度控制的商用液体池最近才得以开发，此前利用液相透射电子显微镜（LCTEM）对其效应的研究尚未得到充分开展。然而，此类原位研究对于揭示温度变化所引发的复杂动力学和热力学效应至关重要。事实上，一方面，升高温度会显著影响生长过程中的晶面形成，加快原子和分子在液体中的扩散速率，并改变反应速率；另一方面，纳米结构的纳米相图对温度也极为敏感。本文中，我们利用近期开发的加热型液体池，在室温至100 °C的温度控制范围内，实时追踪金纳米颗粒在水中的辐射还原生长过程。该方法结合扫描透射电子显微成像与衍射技术，在日益接近真实合成条件的环境中进行观察，从而缩小了原位透射电镜观测与常规实验规模合成之间的差距。

方案

1. Align the transmission electron microscope for STEM HAADF imaging

1. Follow manufacturer instructions for microscope alignment.
2. Use a conventional dried sample to align the microscope. Do not use a liquid sample.
3. To minimize the growth speed, minimize the electron dose rate (see discussion section) which implies using small condenser aperture and small spot size to reduce the beam current.

2. E-chip handling

NOTE: Commercial liquid-holders fit on almost all TEM but use the holder that is specifically design for the microscope brand and pole piece. A liquid cell is made of two MEMS-based silicon chips called E-chips, both of which are silicon substrates with a 500 x 50 µm window covered by a 50 nm thick amorphous silicon nitride (SiN) film that is electron transparent (Figure 1A). These two e-chips have different sizes. The small one is 2 x 2 mm with gold spacers that fix the distance between the two E-chips (150 nm here) and the liquid thickness. The large one is 4 x 6 mm and it has a resistance embedded inside the silicon substrate that allows a uniform heating of the liquid sample (Figure 1B). Because of the way they are fabricated in clean rooms, the E-chips have two different sides: one where the window looks small (here after called the front side) and the other where the window is large with a sink shape (here after called the back side).

1. When handling the E-chips, never touch the window with tweezers and grip the chips by the sides. To avoid scratching the surface of the silicon substrate use carbon-tipped tweezers.
2. If placing an E-chip on a surface, make sure that the back side is in contact with the surface because the SiN film is fragile and it is deposited on the front side.

3. Cleaning of the liquid cell holder (before the experiment)

1. Remove the lid covering the tip of the holder. Remove the dummy liquid cells using tweezers. Remove the gasket used with the dummy liquid cells.
   NOTE: Dummy liquid cells are liquid cells without the SiN window and only silicon. They are used for storing the liquid cell holder in the vacuum pump. Pay attention to the state of brass screws because they crumble easily over time. Particularly, if the screw heads are damaged, the screws must be changed. Otherwise, it may be difficult to unscrew them after the experiment and small debris may also disrupt sample loading.
2. Manually inject 2 mL of distilled water inside the holder using syringes and the external PEEK tubing to connect to the back of the holder.
   NOTE: There are 3 microfluidic tunnels inside the holder. All three must be cleaned up with water. Pay attention to the water coming out of front of the holder: if the water is colored because of a previous experiment, continue to put water inside the holder until the liquid is uncolored.
3. If injecting a solution in the liquid cell during the experiment (1 mM of HAuCl₄ in water in our case), fill the tubing of the sample holder with this solution.
4. Dry up the tip of the liquid-cell holder using an air pistol.

4. Preparation of the liquid cell (E-chips)

1. Cleaning of the liquid cells.
   1. Fill a glass Petri dish with acetone.
   2. Fill a glass Petri dish with methanol.
      ​CAUTION: Due to the toxicity of methanol, the Petri dish with methanol must be put under a fume hood. Methanol should be handled with the adequate protective gear (gloves).
   3. Put one small and one large E-chip in the Petri dish with acetone and wait for 2 minutes.
      ​NOTE: The E-chips are coated with a protective layer that needs to be removed before the experiment. Acetone will remove the photoresist and clean up the E-chips of debris. To enhance the cleaning, the solution can be gently agitated.
   4. Put both E-chips in the Petri dish with methanol and wait for 2 minutes. The methanol will clean up the E-chips from the acetone and the rest of the debris.
      CAUTION: The transfer of the E-chips between the acetone and the methanol must be done as fast as possible in order to not let the E-chips dry up in the air.
   5. Dry up the liquid cells using an air pistol. Hold the E-chip using tweezers while using the air pistol. Be careful to not press too much on the air pistol trigger otherwise the E-chip can drop out of the tweezers. If they drop out, restart the cleaning with acetone and methanol.
   6. Verify the integrity of the silicon nitride window using a binocular magnifier or an optical microscope (Figure 2).
      ​NOTE: Make sure that the windows of both E-chips are clean and not broken. If the E-chips do not seem clean, try to put them back in acetone and methanol again. If the dirt is still on the window or if the window is broken, the E-chips must be changed with new ones.
   7. Plasma clean the E-chips with a mixture of argon and oxygen gas for 2 minutes. Plasma cleaning the E-chips allows them to be hydrophilic. Here are the details of plasma cleaning set up: argon gas flow = 35 sccm, oxygen gas flow = 11.5 sccm, gas flow timeout = 20 s, Forward RF target = 50 W, Forward RF Range = 5 W, Maximum reflected RF = 5 W.
2. Loading the liquid cells in the TEM holder (Figure 3).
   1. Load the gasket O-rings inside the liquid cell holder (Figure 3B). Verify that the gasket used is clean. If not, clean it rapidly with distilled water. Dry it up using a clean filter paper. To remove debris and fibers on the gasket, press it between two sheets of parafilm multiple times.
   2. Put the small E-chip inside liquid cell holder (Figure 3C). To reduce the bowing of the SiN films towards the vacuum of the microscope, place the windows of the liquid cell in a crossed configuration. Therefore, the window of the small E-chip must be parallel to the length of the holder and the front face up. Make sure that the small E-chip is well inserted inside the gasket.
   3. Prepare the liquid sample (here, 1 mM of HAuCl₄ in water).
   4. Drop ≈2 µL of the liquid sample on the small E-chip using a micropipette (Figure 3D). If the small E-chip has been properly plasma cleaned, the aqueous liquid sample will spread evenly across the surface of the chip.
   5. Remove the extra liquid with a filter paper. With a sharply cut piece of filter paper, reduce the thickness of the liquid layer on the small E-chip until it forms a flat dome.
   6. Put the big E-chip inside the liquid cell holder (Figure 3E). Place the large E-chip on the small one with its front face down (the front sides of the two chips must face each other). The electrodes on the large E-chip must be in contact with the electrode pad on the holder.
   7. Slide the lid back on the liquid cell holder. Gradually tighten each screw (Figure 3F).
   8. Dry up the eventual liquid coming out of the E-chips using small cut-out filter paper. Verify that there is no liquid coming out on both sides of the liquid cells by rotating the liquid cell holder around its axis.
3. Test the vacuum sealing of the liquid cell in a pumping station. If the vacuum level of the pump reaches 5 x 10^-2 Pa then continue the protocol. If not, check the integrity of the window (it is most likely broken) and start the protocol from the beginning with a new set of E-chips.
4. Verify one last time the integrity of the silicon nitride window using a binocular magnifier or an optical microscope. Sometimes, the liquid cell will be able to sustain the vacuum of the pumping station even if the window is broken. This is because when the window breaks and the liquid spills out, it can form aggregates of salt on the broken part of the window thus covering the hole. If it happens, prepare a new set of E-chips.
5. Load the liquid cell holder in the TEM and check the vacuum level. Even if the liquid cell sustained the vacuum of pumping station and there is no visible problem with the window, micro-leak of the liquid-cell can prevent to reach the vacuum level required to operate the TEM. If the microscope cannot reach the required vacuum level to operate (2-5 x 10^-5 Pa), remove the sample holder and prepare a new set of E-chips.

5. Use the liquid holder in flow mode

1. Fill up 2 syringes with a few milliliters of the solution to be injected (1 mM of HAuCl₄ in water in our case).
2. Connect 2 external PEEK tubes to the syringes. Place the 2 syringes on the syringe pumps. Insert the external PEEK tubes in the 2 entries of the liquid cell holder. Insert one additional external PEEK tube for the output of the liquid cell holder.
3. Inject the solution with a flow rate of 5 µL/min in each inlet.

6. Heating of the liquid environment

1. Connect the power supply to the holder. Connect the power supply to the computer on which the heating software is installed.
2. Power up the computer and open the Heating software. Power up the power supply.
3. Click on the device check button. If the software indicates "passed" then the experiment can continue. Otherwise, the large E-chip might have a problem (incorrect loading of the E-chip, broken electrodes…).
4. Click on the Experiment tab. Click on Manual to activate the manual mode of heating.
5. Select the targeted temperature and change accordingly the temperature rate. Press apply to heat up the E-chips to the targeted temperature (Figure 4).
   NOTE: The E-chips can be heated up to 100 °C. If an aqueous solution is used for the experiment (as in our case), avoid heating the E-chips above 90 °C. Otherwise, the liquid sample can dry up. When heating up the liquid, the temperature can temporarily rise above the targeted temperature and then fall back to the desired temperature. Use low heating rate to minimize such overshoots (1 °C/s is fine).
6. Click on Ambient to go back to the ambient temperature (25 °C). Click on Stop to stop the heating abruptly. Click on the End session tab to end the heating experiment.

7. STEM imaging of nanoparticles growth

1. Use the microscope in STEM mode using the HAADF detector. Go to a pristine area of the sample, near a corner of the observation window where the liquid thickness is minimum. Acquire videos of nanoparticle growth for different temperatures of the liquid (Figure 5).
   NOTE: Gold nanoparticles immediately appear and grow in the scanned area. Video recording with a frame rate of one image per second is a good compromise to observe the growth processes with good signal to noise ratio and a good time resolution.

8. STEM nanodiffraction of single nanoparticles

1. Acquire a STEM HAADF image of several nano-objects. Acquire the diffraction pattern of individual nanoparticles selected on the image using the STEMx software (Figure 6).
   NOTE: STEM nanodiffraction is a technique that allows acquiring the diffraction pattern of single nanoparticles in liquid during growth experiments²².
2. After the acquisition of a STEM HAADF image, select several nano-objects on the image and the STEMx software automatically synchronizes the position of the probe and the CCD camera to acquire the diffraction pattern at each position of the probe. To avoid the overlapping of the diffraction spots, use a small convergence angle of the STEM probe (7.4 mrad in our case) by using small condenser aperture (10 µm in our case).

9. Cleaning of the liquid cell holder (after the experiment)

NOTE: Here we describe a standard cleaning procedure for the liquid cell holder. If this cleaning is not efficient enough, it is possible to use diluted nitric acid and methanol to flush out the eventual nanoparticle aggregates in the liquid cell holder. The chemical compatibility documentation of the liquid cell holder should be consulted before. In any case, always finish the cleaning with the injection of distilled water.

1. Remove the lid. Remove the used E-chips. Remove the internal gasket.
   NOTE: The used E-chips can be stored in an adapted box. It is then possible to perform ex situ TEM or SEM analyses of the nano-objects that remained attached to the SiN windows after unsealing the liquid-cell¹⁵. It is not advisable to reuse the E-chips for another in situ experiment, but it is still possible if the SiN film has not been broken during the unsealing of the liquid cell. Overgrowth experiments in a different solvent can then be performed.²³
2. Inject 5 mL of distilled water in the inlet and outlet tubing of the liquid cell holder.
3. Clean the tip of the liquid cell holder using an ultrasonic bath for 20 minutes. The contact pad can be immersed in the bath. Only immerse the part that is covered with the lid. Do not immerse the vent holes in liquid.
4. Dry up the liquid cell holder using an air pistol.
5. Put back the gasket used with the dummy liquid cells. Put back the dummy liquid cells and the lid.
6. Store the sample holder in a vacuum station.

10. Post-experiment analysis using Fiji (ImageJ)

NOTE: It is recommended to split each frame of the video taken into single images. The purpose of this post-experiment analysis step is to transform the original videos of the nanoparticles into binary videos that can be analyzed by Fiji. A median filter is used in order to enhance the contrast of the nanoparticles on the background (Figures 7B & 7E). This is essential to facilitate the binarization of the video.

1. Open the file directory containing the images of the video on Fiji by clicking on File | Import | Image sequence. The sequence options window will pop-up. Select the appropriate starting image (if the beginning of the video is to be discarded). Enter the increment number for the image sequence (it corresponds to the number of frames it takes for the STEM scan to reach the bottom of the image). Check the box for Convert to 8-bit grayscale. Save the image sequence in tiff format.
2. Crop out all the undesired artifacts from the video (for example the scale bar or the edge of the liquid cell window).
3. Click on Process | Filters | Median to apply a median filter on all the images. Save the processed image sequence in tiff format.
   NOTE: A window will pop-up asking the radius used for the median filter. We used a radius of 2 pixels but feel free to use different parameters. Other filters are also available in Fiji that could be used to enhance the image processing. Particularly, the subtract background algorithm can be used to make the background intensity flat if it is non-uniform. For this, click on Process | Substract background. In our case, this process creates small white patches in the background of the first images that can be interpreted as false nanoparticles. Thus, we did not use this process but it should be tried on other datasets, because the increasing liquid thickness from the corner to the center of the liquid cell usually induces non-uniform background intensity on low magnification LCTEM images.
4. Click on Image | Adjust | Threshold. Move manually for a better precision of the threshold of the binarization until only the nanoparticles are colored in red. Press the apply button. The Convert stack to binary window will pop-up. Uncheck Calculate threshold for each image. Save the binary image sequence in tiff format (Figures 7C & 7F).
   NOTE: It is recommended to check if the threshold is satisfying on each frame of the video.
5. Do this step only if there is a contrast inversion of the nanoparticles during the video. Click on Process | Binary | Dilate. Do it one more time if necessary. Click on Process | Binary | Fill Holes (Figure 7G, see the discussion section).
6. Click on Analyze | Analyze particles. Define the size range of the analyzed nanoparticles observed during the experiment. Check Summarize.
   NOTE: It is very important to at least define the minimum size of the nanoparticles observed. Without it, the small black dots (noise) that appear in the binary image sequence will be considered as nanoparticles. As a first try, choose the size of the smallest nanoparticles identified by the eye, but then a trial and error process is necessary to understand the effect of this parameter and optimize the automated data analysis (see discussion section). Before this step, to retrieve the area of the nanoparticles, click on Analyze | Set measurements and check Area. Other measurements are available.
7. Save the Results and Summary windows. The number of nanoparticles for each frame is in the Summary data window.

结果

图5 展示了在25 °C和85 °C下，历时80秒采集的两组金纳米颗粒形成的STEM HAADF图像序列。在所有这些实验中，纳米颗粒的成核与生长均由水的辐射分解驱动。在电子束诱导产生的各种化学物种中，强还原剂（即水合电子和氢自由基）可还原四氯金酸，从而在SiN窗口与液体的界面处形成金纳米晶体。这两项在相同电子剂量率下进行的原位观察证实，本方法能够直观展示温度对液相中纳米颗粒形成过程的显著影响。在低温条件下，观察到大量小尺寸纳米颗粒形成的致密聚集体；而在高温条件下，则生成少数尺寸较大且晶面清晰的纳米结构。由于STEM HAADF图像的衬度与金纳米颗粒的厚度成正比，因此可以看出，在这些生长实验过程中形成了两类不同结构：一类是衬度较高的三维纳米颗粒，另一类是衬度较低、呈三角形或六边形的大尺寸二维纳米结构（如图5中红色箭头所示）。

本方案中描述的视频分析方法可通过随时间测量观察区域内纳米颗粒的数量及其平均表面积，从而对成核和生长过程进行量化。如图8所示，在低温条件下，观察的数十秒内可形成超过800个纳米颗粒，而在高温条件下仅形成30个纳米颗粒。除两个三角形和六边形纳米片外，所有纳米颗粒在高温跟踪过程的第一张图像中就已全部存在。图9显示，在85 °C下纳米颗粒的平均表面积增长速度比25 °C下快40倍。

图6 展示了两个金纳米颗粒的典型扫描透射电子显微镜（STEM）图像及其衍射图样，这两个纳米颗粒直接在图像上被选中（由图6A中的红色箭头指示）。在此，我们可以识别出金的面心立方（FCC）结构，其分别沿[001]（图6B）和[112]（图6C）晶带轴取向。

图1：电子芯片（E-chips）及液体池支架尖端的示意图。 (A) 用于加热液体池的大尺寸电子芯片（顶部）及其电阻结构，以及小尺寸电子芯片（底部）。(B) 两个电子芯片均装入液体池支架中。大尺寸电子芯片的电极与液体池支架上的电极焊盘接触，其电阻可对液体池进行加热。请点击此处查看该图的放大版本。

图 2：E芯片的光学显微镜图像，显示：（A）实验所需的完整SiN窗口。（B） E芯片边缘处破损的硅晶圆。如果破损区域位于液体池密封后形成的湿润区域之外（即破损区域位于O形圈所界定区域之外），则此类E芯片仍可使用。（C） E芯片表面的残留物。若此类残留物在重复清洗步骤后仍无法去除（参见第4.1节），则不应使用该E芯片。（D至F） SiN窗口破损的E芯片（不可使用）。请点击此处查看此图的高清版本。

图3：液体池装入透射电镜样品杆的分步操作照片。（A） 单独的样品杆。（B） 将垫圈O型圈放入腔体内。（C） 将小型E芯片插入垫圈O型圈中。（D） 在小型E芯片上滴加一滴溶液。（E） 将大型E芯片覆盖在小型E芯片之上。（F） 通过拧紧盖子密封整个液体池。 请点击此处查看该图的放大版本。

图 4：控制液体池温度的加热软件界面截图。 请点击此处查看此图的放大版本。

图 5：金纳米颗粒生长过程的低倍STEM HAADF图像序列。(A) 在25 °C下。(B) 在85 °C下。每幅图像左下角标注了相应的时间。红色箭头指示二维纳米结构。所有图像均采用相同的电子剂量率3.4 electron·s^-1·nm^-2采集。请点击此处查看此图的放大版本。

图6：单个纳米颗粒的扫描透射电子显微纳米衍射。(A) 用于选择发生衍射的纳米颗粒的STEM图像（衍射采集过程中探针位置以红色箭头标示）。(B,C) 两个选定纳米颗粒的衍射图样。请点击此处查看该图的放大版本。

图7：使用Fiji对STEM HAADF图像进行数据处理与分析。 图像采集于生长开始后40秒。（A至C） 在25 °C下采集的图像。（D至G） 在85 °C下采集的图像。（A, D） 原始STEM图像。（B, E） 处理后的图像（中值滤波）。（C, F） 二值图像。（G） 对像素进行两次膨胀操作，随后应用“填充孔洞”处理。请点击此处查看该图的放大版本。

图8：在25 °C和85 °C下金纳米颗粒数量随时间变化的曲线图。 25°C下的两条曲线是通过自动测量获得的，其最小检测尺寸（S_min）分别为20（红色）和50（蓝色）像素²。绿色圆点表示在25 °C下采集视频后，通过人工计数得到的12秒和60秒时的纳米颗粒数量。 请点击此处查看该图的放大版本。

图9：金纳米颗粒平均表面积随时间变化的图表，分别表示在 25 °C 和 85 °C 条件下的结果。绿色圆点代表在 85 °C 下采集的视频中，特定时间点金纳米颗粒平均面积的手动测量值。请点击此处查看该图的放大版本。

图 10：85 °C 下单个金纳米立方体生长过程的高倍 STEM HAADF 图像序列。 该图像序列的电子束剂量率为 83.6 electron.s^-1.nm^-2。 请点击此处查看此图的放大版本。

讨论

本方案描述了在温度可控的液体介质中，通过辐射分解驱动金纳米颗粒成核与生长过程的实验方法。结合自动化的视频处理技术，可测量温度对纳米颗粒合成关键参数的影响，如纳米颗粒的密度、尺寸、形貌及原子结构。这些重要数据可用于评估温度对成核速率和生长速率的作用，检测可能发生的相变过程，并可视化决定胶体溶液最终形态的晶面形成过程。结合对反应介质成分的调控能力，温度可控的液体池透射电子显微镜（TEM）技术为进一步在接近真实合成条件的情况下直接观察各类纳米结构的成核与生长过程提供了新的途径。本文结果的解读及其与成核-生长模型的比较将在他处讨论。本文旨在强调开展有意义的原位TEM实验时必须考虑的若干方法学要点。

首先，必须明确反应介质中的电子束效应，因为这些效应可能极大地影响实验结果。在此，由于水的辐射分解是纳米颗粒形成的驱动力，因此随着电子剂量率的增加，生长速度迅速加快，从而影响最终纳米结构的形貌^11,15。因此，为了研究温度对纳米颗粒成核与生长的影响，有必要在相同电子剂量率下进行生长实验的比较。在STEM模式下，电子剂量率等于束流（单位：电子/秒）除以成像区域面积（单位：nm²）。因此，保持恒定的电子剂量率意味着每次实验需维持相同的束流（即相同的聚光镜光阑和束斑尺寸）以及相同的放大倍数。使用CCD相机或法拉第杯对成像条件下的束流进行量化，对于数据的解释和可重复性至关重要。放大倍数及相应的剂量率应根据研究目的进行选择：若希望观察大量纳米颗粒的生长过程，以获得具有统计意义的生长动力学数据（图5）；或在单个纳米颗粒尺度上研究生长机制，以识别纳米颗粒表面的优先吸附位点（图10）。如果成核和生长过程过快，尤其是在高倍率下，应选择较小的聚光镜光阑和较小的束斑尺寸，以尽量降低剂量率。此外，通过降低分析溶液中金属前驱体的浓度也可减缓纳米颗粒的成核与生长，但需注意，辐射产物的浓度会随温度升高而增加。一般而言，还必须考虑整个样品所经历的电子辐照历史。例如，若在相邻区域快速连续进行多次生长实验，由于研究区域内金前驱体浓度逐渐降低，纳米颗粒的密度将随时间减少。该效应可通过在空间和时间上分隔生长实验，并采用流动模式的液体样品台来最小化。

界面追踪算法对于自动分析视频并定量提取大尺寸纳米颗粒组装体的成核与生长结果极为有益。然而需要注意的是，图像二值化步骤始终依赖于具体的数据，这意味着为优化纳米颗粒/液体界面的检测而需施加在图像上的滤波方法和数据处理流程会因实验不同而有所差异。此外，将这些自动分析的结果与在少量图像上进行的手动测量结果进行对比至关重要，以便优化图像处理流程并明确其局限性。例如，在本研究中，随着高温下形成的三维纳米颗粒逐渐增厚，多重散射事件导致其核心在观察30秒后出现对比度反转，这是因为散射电子的角度展宽使得环形探测器角度范围内的信号收集减少。为了持续准确测量这些纳米颗粒的真实表面积，我们在图像二值化后采用了"填充孔洞"的数据处理方法，以填充环状对比结构的内部圆形区域（图7F，G）。然而，我们还需对物体进行轻微膨胀处理，以确保这些环状对比结构始终保持完全连接。后一步骤导致自动测量中纳米颗粒的平均表面积略有高估（图9）。类似地，在纳米颗粒检测过程中，我们必须设定一个被检测物体的最小尺寸（S_mim），以避免将噪声误判为信号，但该参数会影响测得的成核速率。如图8所示，实验初期检测到的纳米颗粒数量逐渐增加，最终趋于平稳。当S_min较大时（50像素²，相当于1543 nm²），自动测量与手动测量在平台期数值上一致（60秒时为835个纳米颗粒），但在自动分析中，纳米颗粒的检测出现延迟：手动计数在12秒时已达到835个，而自动检测则需更长时间才能达到相同数量。这种检测延迟导致成核速率被低估。将S_min降低至20像素²（即617 nm²）可减小纳米颗粒组装成核时间的误差，但会导致纳米颗粒密度被高估，尤其是在实验早期阶段（图8），这也会影响成核速率的准确性。在液相透射电镜中，对具有高度动态行为且信噪比较低的纳米物体进行检测及其尺寸与形状测量是一项普遍挑战，未来可通过其他分割与去噪方法²⁴或机器学习方法²⁵进一步改善。

最后但同样重要的是，液体池的制备以及液体样品 holder 的清洗必须非常仔细地进行，以避免反应介质受到污染。

通常，在液相透射电子显微镜（LCTEM）分析过程中控制样品温度，能够为研究固-液界面化学反应的热效应提供机会。因此，我们希望本方法能够为其他原位透射电镜实验开辟道路，以揭示在温度可控的液体环境中硬质材料、软质材料或生物材料的动态行为。

材料

本文使用的材料清单

姓名	公司	目录编号	评论
2100 Plus 电子显微镜	Jeol
丙酮	Merck
气动喷枪
ARM 200F 电子显微镜	Jeol
双目镜或光学显微镜
碳尖镊子
配备加热软件的计算机	Protochips 公司软件
蒸馏水
假 e-chip	Protochips
垫圈/O型圈	Protochips
金水溶液	Merck		1 mM HAuCl4 —— 需预先配制
大型液体加热 E-chip	Protochips
甲醇	Merck
One View 相机	Gatan
培养皿			数量：2
等离子清洗仪	Gatan
Poseidon Select	Protochips		液体池样品杆
电源 Keithley 2450
防护手套
红色 PEEK 管			数量：3
带扭矩的螺丝刀
小型液体 E-chip	Protochips		150 nm 间隔层
STEM HAADF 探测器	Jeol
STEMx 软件	Gatan
注射器			数量：2
注射泵	Harvard Apparatus		数量：2
真空泵	Gatan