I completely agree with Dominik. Clean working at all times, also during sample preparation in another lab, for instance, is extremely important.

A year ago we got two in-situ plasma cleaners from ibss for our SEM lab. One is permanently mounted onto a SEM/FIB with EDX, EBSD, CL and we activate it once a day for a few minutes to get rid of all carbon contaminations inside the chamber or even on the inserted samples. We saw great improvements by this cleaning, although at very high beam currents and long time scans, there can still be a generation of contamination visible, but it was heavily decreased in comparison to the earlier state.

If your sample has a low conductivity and you want to perform EBSD, I found it helpful to completely wrap the sample in copper tape of silver paste, even on the front side, and only leave the area of interest free. This can really help you to reduce charging which leads to drift.

Regarding contamination of different materials: I suppose it is a question of secondary electron and backscattered electron yield. For example, a gold surface is contaminated extremely fast. The explanation is easy: The more electrons are scattered back from the substrate, the more gaseous hydrocarbons are cracked and deposited on the surface

Avoid all contaminations as good as possible, e.g. make sure that all users wear gloves whenever they handle samples or open the SEM. Make sure that samples are as clean as possible before you insert them; samples cleaned with organic solvents (use high purity ones) need to be completely dry before they are transferred into the SEM. Check the pumps of your SEM system: oil-free ones are standard with most (but maybe not with all) suppliers (e.g. membrane rough pump instead of oil pump + turbopump with magnetic bearing).

If you still have contamination problems, consider using a plasma cleaner which removes carbon by an oxygen plasma, see e.g. http://www.evactron.com. You can add this to your systems if you have a flange. Plasma cleaning can sometimes give great improvements of the vacuum and sample cleanliness conditions.

What about the conductivity of your sample? If this is limited, you'll also get drift problems. Make sure that there is a highly conducting connection to the sample stage. If you use carbon pads, make sure that the connection area to your sample is as large as possible - the conductivity of carbon pads is OK, but not very excellent for small contact area, low attachement forces and large sample currents. You might consider of adding small amounts of inert gases like argon into your vacuum, this can help to minimize charges.

I completely agree with Dominik. Clean working at all times, also during sample preparation in another lab, for instance, is extremely important.

A year ago we got two in-situ plasma cleaners from ibss for our SEM lab. One is permanently mounted onto a SEM/FIB with EDX, EBSD, CL and we activate it once a day for a few minutes to get rid of all carbon contaminations inside the chamber or even on the inserted samples. We saw great improvements by this cleaning, although at very high beam currents and long time scans, there can still be a generation of contamination visible, but it was heavily decreased in comparison to the earlier state.

If your sample has a low conductivity and you want to perform EBSD, I found it helpful to completely wrap the sample in copper tape of silver paste, even on the front side, and only leave the area of interest free. This can really help you to reduce charging which leads to drift.

Regarding contamination of different materials: I suppose it is a question of secondary electron and backscattered electron yield. For example, a gold surface is contaminated extremely fast. The explanation is easy: The more electrons are scattered back from the substrate, the more gaseous hydrocarbons are cracked and deposited on the surface

Do you know for sure that your sample isn't slipping/moving during analysis rather than 'contamination' causing the shift? Carbon 'stickies' tend to swell and 'droop' in the vacuum. How do you assess beam positioning/stability during a long scan?   I can't remember, but I recolllect the NPL have an system/program for registering the image and correcting for sample movement during long scans. If your sample is of low conductivity, I'd be more concerend that surface charging is causing the drift -  carbon coating (~20angstroms - a nice foxy brown on brass) is typically used in EPMA, although other evaporative coatings have been used.  Carbon coating, however, means plasma cleaning in-situ is not exactly ideal.  

Sample and substrate outgassing is a real potential issue and the world of 'preffered' epoxys and substrates is  fraught, especially in the UHV ion probe community.  Personally, users that insist on whacking the beam all over the substrate are more of an issue than what epoxy or acrylic has been used.  grr. 

Low vaccuums (gas injection) are used in SEMs to disapate surface charge but will, inevitably, generate a larger spot due to increased beam scattering (and  sample charging) in comparison to coated and high-vac samples. In quant- and semi-qaunt SEM work this isn't really an issue as the e- interaction volume is the limiting factor, but not so in imaging or EBSD.    Finally,cracking of gas hydrocarbons is more likely a function of temperature rise  under the spot, which will in turn be a function of material conductivity, stopping power (i.e. average Z) and operating conditions (kV) . Off the top of my head, I'm not sure why backscatter electrons (as opposed to secondary or incident) would promote cracking?  The typical equations used for estimating thermal rise in insulating samples (such as those in Reed's book) are an underestimate given the propensity for, say, alkali element migration during analysis.  This is especially apparent with the advent of FEG-EPMA systems where carbon build up is a real issue.

Dear Jon, slipping/moving (of sample or stage) is certainly a potential reason and has been discussed many times, but for my case this I can exclude. I have in mind that e.g. the same (also metallic) sample, not changed in position, tilt or perhaps intermediately removed, shows drift phenomena. Sometimes more, sometimes less. It is slightly but not exclusively related to the EBSD detector. Moved in a certain image shift becomes visible for all EBSD detectors I operate. I guess that non-conductive parts like FSD detector systems, or parts of the detector screen, or dust ob the screen are responsible for these effects. So we usually keep the system scanning so that we generate something like an equilibrium.

Of course, the visible "drift" is virtually increased by the tilt compensation during scan, i.e. we mainly observe a vertical component which is in general about 3 times more  sensitive than in x, i.e.  the usual drift we observe is along y, i.e. perpendicular to the screen position, but then not always in the same direction. Sometimes maps - or parts of the maps - are compressed (too fast scan along y) or elongated (scanning on the "same" position).

We know that the vacuum control (in VP mode) and the cyclic cooling also influences the beam position, but then mainly at high magnifications. This seems to be a general problem if you look through high resolution maps published over the last years. In these cases vertical boundaries often show an undulate character. The main problem from my point of view is that we usually know these effects but don't "realize" them (because of missing vertical boundaries :-) ) and neclect them during analysis. We are usually already happy if the final scan looks similar to the image we collected before measurement. If you compare an image before and after scan I would bet a bet a wager that there is no scan which does not show any drift over magnification of 5.000. The major question is, how big is the drift? This is also one reason why the image is never superimposed by the map results since then the drift becomes obvious if clear boundaries are visible.

What I realized is that some materials - especially steel - sometimes tend to contaminate dramatically. During discussions with colleagues around the world I realized that the carbon content is assumed to be the major reason. Well, I also observe  contamination at iron meteorites which usually do not have any carbon inside. However, I have to admit that we have to use our equipment for failure analysis so that we cannot clean possibly contaminated samples in order to find the real reason for the respective failure.

Anyway, the main question remains unanswered: How accurate a microstructure description can be if we don't know anything about irregular beam shifts during long term operation. And I am not yet talking about loose of focus (or astigmatism correction) during scan which becomes especially visible by a more or less continuous increase of grain boundary thickness in pattern quality or band contrast maps.

Nevertheless, many thanks for the very useful discussion regarding coating, LV usage, sample prep and contamination reasons.

The software I use for my EBSD data collection, takes an SE image at the start and at the end of a EBSD mapping project. Comparison between the two SE images gives you a fairly good idea of the amount and direction of drift during the mapping. However, this does not give information on the drift rate. But often a linear approximation is a reasonable assumption.

Did you ever checked the linearity, or is this only your assumption?

I've seen many maps where the the drift is very non-linear, i.e. drifting at the beginning and stabilising after a certain time, or the opposite: the charging starts to increase because of a locally different apperance of phases. 

It's my intuition, I seldom have problematic amount of drift in my samples. As Dominik said earlier clean sample handling is half the battle.

Well, usually the sample preparation for EBSD requires more attention than usual and we are quite satisfied and don't have much trouble. And we do the same reagarding cleaning :-). Unfortunately there seems to be no clear systematic.

We apply EBSD for practically all materials (from iron meteorites to rocks, embedded and not embedded). What we observed is a quite strong contamination especially on steels. Despite the fact we use "hot drying" at the end. The problem might be that we really work with totally different materials so that we don't have costant conditions over a long time. At the Geman EBSD meeting  a few years ago we had a presentation where the use of deionized water (instead of only destilled) has been recommended but this could not totally solve all problems.  

In my experience of EBSD on Al and IF steels I almost always found non uniform drift at the beginning of high resolution (sub 75 nm step) maps. I can't say quantitativley but I managed to reduce this by imaging the sample quite a bit before to stabilise the heat load of the beam. I cannot recal though if I saw a difference in the two materials with respect to the thermal conductivity and co-efficient of expansion. For fine grain materials one does not need a very large temperature increase for expansion to be on the order of the grain size and therefore be significant.

Contamination is never really a problem for relatively clean samples as long as the step size is greater than than the contamination spot size. If the step size is smaller then the beam merely acts a s broom an sweeps an almost always increasing ammount of contamination ahead of the beam - therefore the beam never sees a clean surface.

I have also always used the pre and post SEM images to make a qualitative assesment of the degree of drift and there fore the reliability of each EBSD map and to establish if any parts of the map sould be cropped. With the modern systems that now have automatic drift correction it should be relatively easy to extract the linearity and the rates of drift quantitatively.

I have not tried it myself but I know of others who have given their samples a quick shower in the FIB to clean off the last bits of contaminants on the sample surface. This would be worth trying if you don't have access to a plasma cleaner. Many years ago in Manchester we used to have a system with a cold finger which worked well to reduce system contamination.

For the final cleaning of samples after polishing I have always used either ethanol or methanol after cleaning in (soapy) deminerlaised/distilled water. I always ensure that the sample is completely covered with a meniscus of water which I then remove with the alcohol. Again before drying make sure that the sample has a meniscus of alcohol covering it. Final drying is best by rapidly blasting the meniscus off with a jet of compressed air from one direction only. This acts as an air knife and pushes all remaining suspended contaminants off one side of the sample which should obviously be avoided in the SEM. Note: if you are using OPS as the last polishing step I have always polished in hot water to remove the last polishing residues out of the polishing cloth followed by rinsing in hot water prior to demineralised water and alcohol washing.

Hi Jacob, thanks for the comments. We are doing a similar way to clean the samples, except of the hot water trick! This we will try in future since we also observe from time to time crystallized silica particles on the surface.

Regarding the automatic drift correction I am not that optimistic since it does not prevent drift but only "corrects" it in order to make it "invisible". And as far as I know, only Aztec (Oxford Instr.) offers this feature.

Nevertheless, it is worth to be mentioned, that it does not work for a "bigger" drift (in maximum the half of the image), i.e. at higher magnification and drift perpendicular to the tilt axis it is quite limited.  Although better than a totally distorted image of a microstructure, the multiple scanning of the area is not always helpful since one generates there additional contamination so that the cross correlation applied will more and more errornous. Moreover, for sample with very low contrast it does not help at all (except you have access to a FIB and you are generating some remarkable features at the surface - what I and probably the majority of EBSD users do not have). This might be one reason why i havn't seen many investigations where it has been applied.  

Again, many thanks for the numerous conventional tricks.

I haven't performed any EBSD analysis myself. But I did work on contamination and charge control issue for semiconductor SEMs for many years. The drift can be caused by many root causes: 

1) Sampling charging. Apparently, if the sample is not conductive, it is an obvious problem. Even if the sample is conductive, hydrocarbon deposition may still cause some level of charging on the samples. 

2) Charging inside electron optics column or parts that are close to electron beam. Electron optics column (including apertures and detectors) are constantly under the irradiation from some portions of primary electrons, SE or BSE. If your system is heavily contaminated with hydrocarbons, hydrocarbon will also contaminate those electron optics components that can be irradiated by electrons. So when you perform long imaging, charging will gradually build up on those contaminated surface as well. Contamination inside electron optics column is rarely symmetric due to the existence of deflectors. Therefore, those asymmetric charging act as bad deflectors and defect primary electrons gradually and results in beam drift. Charging can saturate. You may see the drift gradually settles after a long imaging process. You can move your imaging FOV to a location that you don't care, and image at the same optics conditions with really high beam current, it will help you to stabilize the charging build-up process faster before you collect final analysis data. Venting to atmospheric pressure can effectively neutralize those charging. So if you vent your system, you may need to repeat those process again. 

Mitigation method:

1) I have developed an sub-frame alignment algorithm for my previous job in an electron microscope company. It corrects image drift during long imaging process. The method is to collect many frames at the same location at faster rate. Then the algorithm can figure out the drift between each frame and within each frame. It will generate an improved images by summing all the drift-corrected frames. If you have access to those raw frames, I can help you process them to see whether it improves the image quality. 

2) Always keep your system clean from hydrocarbon contamination. My current company PIE Scientific (www.piescientific.com) has developed a remote plasma cleaner that help operators keep their electron microscopes free from hydrocarbon contamination automatically. You can contact our sales to arrange a loan unit to see whether it helps you reduce the hydrocarbon contamination issue in your electron microscope. 

Ethanol or methanol leave a invisible layer on the surface of sample and this layer is carbon rich. I've heard that isopropanol is better.

So grateful to read so many professional suggestions here. But as a less-experienced person, I have another question, that how to set parameters when using plasma cleaning? Such as time, power and so on. Thank you all.

Hi, if you want to see what a plasma cleaner can achieve, read my student's work on the reduction of carbon contamination on TEM samples by argon/oxygen plasma cleaning. The link to the Open Access file is here:

https://iopscience.iop.org/article/10.1088/1742-6596/241/1/012017/meta

Just as 2-cents follow-up to @ Jacob Ross Bowen answer - using "Liquid Chromatography" or "Molecular Biology" grade of alcohol with solid residue level of 1ppm or below, dispensed from a very clean bottle makes quite a difference in residual contamination. Isopropanol at such grade is quite reasonably priced as a rare and pleasant surprise.