After 12 years at CAMECA, one of my favorite parts of working in atom probe tomography is still the same: the questions. I always say that the questions are my favorite part of every webinar and it’s true: I love connecting with our users.
To celebrate my 12th anniversary of starting work at CAMECA on August 29th, I pulled together 12 practical questions that come up again and again in conversations with atom probe users, many of these were inspired by what folks asked during our recent webinar. The questions varied from beginner to advanced, just like our users. So to celebrate with me and check out some of my top 12 questions, some that came in recently and some of my personal best practices.
What are the top five tips for improving APT yield?
Let’s first define yield. What we mean is survivability of the specimen through the region of interest. It’s difficult to predict when/how an acquisition might end, many times in specimen failure or fracture. However, that doesn’t mean that every aspect of getting large datasets is out of our control. Here are some things to try as you begin:
- Prepare sharp, clean specimens. An upside-down cone is better than a pencil shaped specimen.
- Use low-kV cleanup steps, such as 5 kV and sometimes 1 kV, to reduce FIB damage.
- Avoid contamination: wear gloves, minimize unnecessary imaging, and get samples under vacuum promptly.
- Start conservatively with lower detection rates when survivability is uncertain. Lower pulse rates can also help, which seems a bit counterintuitive.
- Look carefully at the mass spectrum, including subvolumes and ROIs, before making strong conclusions.
The short version: investing time in your specimen prep often pays dividends.
How do I range my data?
Ranges determine which ions contribute to composition, reconstruction depth, feature size, and quantitative analysis. If the ranges are sloppy or incorrect, any analyses that you are doing will follow from there. A couple of tips:
- Although you can range ions after the fact in the IVAS window, take caution here as those post-recon ranges don’t get assigned a volume. If you are making big changes, it’s worth doing another reconstruction
- Think about your goals. If it’s composition, perhaps a wider range to grab as many ions that could be a part of that peak is the right choice. If it’s feature size/shape, perhaps narrow ranges are the best choice so no background is captured as part of the feature and smears it out
- Consider using Decomposition of Peaks in some cases if overlaps are suspected. This isn’t the best choice in all cases, but it can improve accuracy in others.
- Range consistently. Choose a rule and stick with it. Maybe you range every peak that is 2x the background level. Maybe you use a criterion like ranging down to full width at hundredth max, multiplied by a factor, or widen the ranges vs. mass according to a power law. Sticking to a rule can help promote consistency in your data analysis.
Peak ID tools are helpful, but they do not replace critical thinking. Most datasets benefit from more than one reconstruction pass as well.
How do I get better yield for tungsten and other high-field materials?
Tungsten itself is not the enemy. We can run tungsten wire all day long in the atom probe. The harder problem is tungsten next to materials with very different evaporation fields, such as dielectrics or complex semiconductor stacks. That is where field evaporation behavior, specimen shape evolution, and reconstruction assumptions can all start arguing with each other.
For heterogeneous high-field/low-field systems, I usually start with specimen preparation: remove unnecessary tungsten where possible, keep the region of interest close to the apex, and keep the tip small at difficult interfaces. During acquisition, lower detection rates and pulse rates may help. During analysis, fixed sample count profiles can be more reliable than fixed bin width profiles in low-density regions. Density relaxation or similar corrections may be useful in some cases, but I would treat them as tools to use carefully, not magic buttons. Density relaxation can be used for good or evil, choose wisely!
How can TKD help target grain boundaries for APT?
Transmission Kikuchi diffraction, or TKD, can be extremely useful for targeting grain boundaries before final annular milling. It’s also near and dear to my heart! It gives crystallographic context when the feature of interest is not obvious in standard SEM imaging. For grain-boundary segregation studies, that can be the difference between analyzing the boundary and analyzing a very nice piece of matrix that happened to be nearby.
TKD is also sensitive to geometry and surface conditions. The patterns originate from the exit surface of the specimen, so overlapping grains through the beam direction can make patterns difficult to interpret. A wire-based substrate can help because the specimen can be rotated to avoid stacking grains in the beam direction. Surface damage also matters, so a 5 kV or 1 kV cleanup can improve pattern quality. And yes, sometimes trying both 20 kV and 30 kV is worth it.
Fig.a-c) unique grain t-EBSD maps during the milling process taken at 30kV. Scale bars represent 200nm. d) grain misorientation profile accross the grain boundary identified by the yellow line, showing the misorientation to be approximately 50°.
How do I handle thermal tails in the mass spectrum?
Thermal tails are a reminder that acquisition and analysis are connected. If possible, optimize the acquisition first by reducing laser energy, lowering specimen temperature, and improving specimen preparation. Once the data are collected, the ranging strategy should depend on the analysis goal.
For reconstruction, broader ranges may be appropriate because ranged ions contribute to reconstruction volume and depth scaling. For a focused compositional question or cluster analysis, narrower ranges may improve signal-to-noise and peak specificity. For quantification, a practical approach is to capture a consistent fraction of comparable peaks and then report the impact of reasonable ranging choices. The most important part is consistency and transparency. If the result is sensitive to where the range ends, that is information the reader deserves to know.
If you are a
LEAP 6000 XR user, simultaneous voltage and laser pulsing can help too (V+L mode). This is still a relatively poorly understood or little-used feature, but adjusting the pulse shape by the fixed offset can help reduce thermal tails and help pull peaks out of the tail of a neighbor.
Is that feature real?
Sanity checking is the foundation of good data analysis. A couple of things to try.
- Does it follow a pole or zone line? If so, it could be an artifact
- Does the feature persist with different ranging/reconstruction parameters?
- Any density aberrations to be suspicious of?
- I always recommend preparing the sample in backside/upside down orientation to be sure, that way you can confirm it’s not a result of how the sample field-evaporated.
Just because it’s beautiful or interesting, doesn’t mean it’s automatically a real feature.
What is the resolution and/or detection limit of the atom probe?
I get this question a lot, and frankly it drives me a bit nuts each time. Atom probe is a bit unlike many of the materials analysis techniques in that you can’t predict a priori what volume you will analyze. So when we get this question, I usually ask one in return: “how many ions did you collect?” For example, if we are after ppm sensitivity, we need to collect well more than one million ions, considering detection efficiency, background, etc.
I like to start with some back-of-the envelope calculations on how many total ions are contained within a feature (a finFET or gate-all-around, for example), then work backwards to see how many ions need to be collected to observe a particular feature. Binomial statistics/Bernoulli counting can help here. We can essentially model or calculate for a given concentration and number of observations what we might expect. For example, if we are looking for a concentration of Boron at a 0.01% level, and sampling 1000-atom subvolumes
- No boron at all: 36.7% of the time
- 1 boron in the volume: 36.8% of the time
- 2 boron in the volume: 18.4% of the time
- 3 boron in the volume 6.1% of the time, and so on…
What I mean to say is it’s difficult to predict what you might be able to see without some prior knowledge of what to expect, and a determination of what volume you are analyzing. Similarly with spatial resolution, in highly ordered metal samples, we can do very well even seeing lattice planes. This is of course depending on the acquisition conditions chosen as well, using a laser will provide enough thermal energy to move the atoms around and degrade the spatial resolution a bit. So “sub-nanometer” is a good ballpark to start with and then start asking questions about the sample and acquisition conditions from there.
How to select acquisition parameters?
There are a lot of knobs to turn on the acquisition side: laser pulse energy, base temperature, pulse rate, etc. Typically, we always start with getting enough data to work with, so optimizing first for specimen survivability, then data quality is usually what I recommend. So how to do that? Lower field (i.e. more laser) is usually helpful to increase dataset sizes, but can lead to more thermal tailing. If your sample can withstand voltage pulsing mode acquisitions, this is the best way to collect data! However, not all samples can.
So start with higher laser energy to promote yield, then slowly work that down if you are getting good size datasets. Surprisingly, pulse rate and detection rate both have an effect on yield. The slower you can “dismantle” a specimen, the better off you are. A combination of slow pulse rates and low detection rates can help preserve the sample long enough to get yield.
First rule of atom probing: don’t force anything.
We usually mean hardware, you shouldn’t have to force the horizontal transfer rod, carousels, pucks, etc. Ever! But I am going to extend that to data analysis and interpretation as well. Sanity-check those results!
What cluster analysis parameters should I choose?
Cluster analysis is powerful, but it is also parameter dependent. The nearest-neighbor or friends-of-friends approach links solute atoms based on a user-defined distance, often dmax, for a selected nearest-neighbor order. The order parameter, dmax, Nmin, and derosion all influence the resulting cluster count, size, composition, and also what is determined to be the matrix.
As we often say, cluster analysis is a method to be applied and not a universal solution. This means that there is no single right answer to this problem. If the conclusion disappears when the parameters move slightly, it’s time to rethink those inputs. Going through the process from radial distribution function step-by-step to the cluster analysis can help guide parameter selection so you can make defensible decisions. A couple “pro tips”:
- using higher order values can help make you more confident that you are seeing real clustering. It’s a higher bar to clear for a cluster to be found, so you are less likely to pick up random noise as clusters
- use the cluster index .apt file to visually inspect your clusters before doing serious analysis on the results. Are clusters stuck together/broken apart? It may be time to redo the cluster analysis with different dmax values or other. It’s easier to visualize this with the cluster index file than to try to interpret the cluster results table.
Cluster index tool can help you visualize parameters to ensure you are making good cluster search choices.
What are emerging applications or areas for APT?
Good question! I will choose two of my favorites
First, automation. I am an automation fiend (junkie?)! From starting at CAMECA in 2014, we have come a long way in what we can do. Chain acquisition runs sample to sample automatically, and we can also script acquisitions to either run through sets of conditions to investigate acquisition relationships to concentration data; or improve yield. I believe we are crossing the threshold into having too much atom probe data from a set of specimens, rather than not enough, if automation methods are used well. That can really help us understand the fundamentals, if we can link acquisition conditions to outputs.
Secondly, the rise of cryo. Many groups around the world are employing cryo specimen preparation and cryo transfer with the VCTM to study hydrogen, liquid phase materials, or environmentally sensitive samples. This has gone from “hero work” to being almost routine with current technology like the Ferrovac-CAMECA vacuum cryo transfer module and Ferroloader. I am hoping to do another webinar on this topic at some point, the ease of use continues to improve. For hydrogen, atom probe + cryo transfer is a marriage made in heaven.
While sensible people can have mixed opinions on artificial intelligence of course, it may bring some interesting developments to atom probe data analysis. I am thinking of ranging and cluster searching in particular, but who knows what the future holds.
What keeps me excited about atom probe, 12 years in?
Work in the demo lab is never boring. We see new materials every day, and each material comes with a specific goal or task. Sometimes there are questions that can be answered only with atom probe. That is what keeps this field fun.
I am deeply dedicated to our users, to helping grow the field of atom probe tomography and to help newcomers to the field learn how to do atom probe, and to do it well. Here’s to the next dozen years of APT, I hope to be right there with you.
👉 Learn more about the APT technique and CAMECA's instruments by visiting CAMECA's APT Overview.
Authors: Katherine RICE (Applications and Market Development Manager)