Energy Dispersive X Ray Spectroscopy

Why Your EDX Maps Look Like Static

I spent three days trying to figure out why my copper map showed a ring artifact around every particle. Turned out the detector was slightly misaligned and the dead time was sitting at 38 percent. Most people would have just accepted the garbage data. I learned to watch that number like a hawk. Energy Dispersive X Ray Spectroscopy is one of those techniques everyone assumes they understand until their first real sample comes back looking like nonsense. The theory is straightforward enough. You hit a specimen with electrons, inner-shell ionizations happen, characteristic X-rays come out, and a silicon drift detector catches them. You get a spectrum. Peaks correspond to elements. Done. The reality involves a lot more frustration.

Getting Started With Energy Dispersive X Ray Spectroscopy

First thing nobody tells you: your sample needs to be conductive. If it is not, you charge up and the whole spectrum drifts across the detector. I have seen peaks move by forty electron-volts over a ten minute acquisition because someone mounted a polymer on carbon tape and forgot about it. Coat it. Carbon, gold, platinum — whatever your lab has. Five nanometers is enough for most things. Don't overdo it or you will start seeing your coating in the spectrum. The detector needs a stable beam current. If your SEM is drifting, your quant results are garbage no matter how fancy the ZAF correction is. Let the system warm up for at least thirty minutes if it has been cold. I once quantified a steel sample and got 15 percent chromium when the spec was 18 percent. Three hours of debugging later I found the column had been cycling between two currents because the power supply was old. The machine was still producing images that looked fine. That is the danger. Set your acceleration voltage based on what you are looking for. For light elements — boron, carbon, nitrogen, oxygen — you want low kV, something like five to ten kilovolts. Higher voltages push the interaction volume deeper and your X-rays get absorbed before they escape. For heavier elements, you can go higher, but there is diminishing return. Twenty kilovolts usually covers most metallurgical work without creating excessive background or exciting too many secondary peaks from the column.

What Actually Happens When You Collect Data

Your spectrometer software will ask for acquisition time and live time. Live time is what matters. It is the actual time the detector is counting. Real time includes dead time and pauses. If you set a ten minute real time acquisition and the dead time is 40 percent, you only collected six minutes of useful data. Most software will tell you the dead time percentage. Watch it. Keep it below 35 percent if you can. Above that, pulse pile-up starts corrupting your peaks and the software corrections become increasingly unreliable. Peak overlap is the single biggest source of error in routine EDX work. Here are the ones that bite people constantly: the sulfur K-alpha line at 2.307 keV overlaps with the molybdenum L-alpha at 2.293 keV. The lead M-line at 1.812 keV sits right under the sulfur K-alpha. The calcium K-beta at 3.691 keV overlaps the vanadium K-alpha at 3.656 keV. If you are doing environmental scanning electron microscopy of particulate matter, you will hit all three of these. Your software may deconvolute them, but deconvolution is only as good as your peak shapes, and peak shapes degrade when your detector is not properly cooled or when the dead time is high. I had a case where a ceramic insulation sample kept showing up with mysterious sodium and chlorine peaks. I was convinced I had contamination. Spent two days cleaning the stage, baking the samples, checking everything. Turns out the epoxy I was using to mount the specimens contained a plasticizer that outgassed chlorine under the beam. The beam heated the mount point and the epoxy released HCl. The X-ray detector picked it up right where the sample sat. Switched to a different mounting medium and the chlorine disappeared. Not the sample. Not the stage. The glue.

Quantification: What the Numbers Actually Mean

EDX quantification is semi-quantitative at best. The software applies matrix corrections — ZAF or phi-rho-z depending on your package — and spits out weight percentages. These are useful for identifying phases and estimating composition, but they are not certified reference material accuracy. You are typically looking at plus or minus five to ten percent relative error for major elements on homogeneous samples. For minor and trace elements, it gets worse fast. Detection limits sit around 0.1 weight percent for most elements under ideal conditions, though the actual limit depends heavily on your beam current, acquisition time, and the background under the peak you care about. Here is something most manuals skip: the standard you choose matters more than you think. If you are quantifying an iron-based alloy against a pure iron standard, the absorption correction will be different than if you quantify against an Fe2O3 standard. The matrix is completely different. Use standards that match your sample matrix as closely as possible. If you do not have a matching standard, use the raw intensities and let the software do an abstract standardless quantification. It will be less accurate but at least you know what you are getting. For thin samples or coatings, the approximation that X-rays come from an infinitesimal volume breaks down. Your interaction volume can be several micrometers deep depending on the acceleration voltage and the density of your material. If you are analyzing a two hundred nanometer gold layer on silicon, your spectrum contains signal from both the gold and the silicon substrate. The software will try to correct for this, but the correction factors blow up when the layer is thinner than the interaction volume. In those cases, drop your voltage. Ten kilovolts or lower for thin gold. Ten for thin copper. You reduce the interaction volume and your spectrum becomes more surface-sensitive.

Mapping and Why It Takes Forever

Elemental maps are useful but they are also a trap. People set up a twenty by twenty micrometer map with two hundred by two hundred pixels and then complain it takes three hours. Each pixel needs enough counts for a decent spectrum. If you are spreading your acquisition time across four hundred pixels and you only get three seconds per pixel, your detection limits are awful and your maps look grainy. The fix is simple: either increase the dwell time per pixel and reduce the resolution, or increase the pixel count and accept the longer acquisition. There is no free lunch. For quick surveys, I use low resolution maps at long dwell times. A fifty by fifty micrometer field at fifty by fifty pixels with twenty milliseconds per pixel gives you something in about two minutes and shows you where the action is. Then I go back and zoom in on the interesting regions with higher resolution. This approach cuts my total mapping time from around two hours down to roughly twenty minutes for most samples. One thing to watch with maps: beam drift during acquisition. If your stage is warm or yourSEM column is drifting, the map will blur. I mark the area, acquire a quick low-mag image, come back after the map finishes, and check that the features line up. If they do not, I let the chamber thermalize longer and try again. It sounds basic but I see people publish maps where the edges of particles are smeared across multiple pixels and they treat the data as solid.

When EDX Simply Cannot Help You

There are situations where you should not bother with EDX at all. Hydrogen, helium, and lithium are invisible to standard silicon drift detectors. Period. If you need to detect hydrogen in a polymer, you are looking at nuclear reaction analysis or elastic recoil detection, not EDX. Light elements from boron through fluorine are technically detectable but the quantitative reliability is poor without special thin-window detectors and careful calibration. EDX also cannot distinguish between different chemical states of the same element. Your spectrum will show a sulfur peak whether it is sulfide, sulfate, or elemental sulfur. If you need speciation, you need X-ray photoelectron spectroscopy or Raman, not EDX. I had a colleague trying to identify corrosion products on a copper pipe and she kept getting confused because the EDX could not tell cuprite from malachite. Both contain copper and oxygen. The sulfur in the water supply was showing up as a contaminant and throwing her off. She switched to XRD and identified the phases in twenty minutes. Should have done that first. Organic materials without staining or heavy metal tagging produce very weak EDX signals. The carbon and nitrogen peaks are there but the low atomic number means few X-rays are generated and most of what is generated gets absorbed in the sample itself. You get a spectrum dominated by carbon, some oxygen from contamination, and a lot of noise. For anything organic, you are better off combining EDX with other techniques or accepting that EDX will only tell you the bulk elemental composition at best.

Practical Habits That Save Time

Always acquire a quick survey spectrum before committing to a quantitative analysis or map. Thirty seconds at five thousand counts per second tells you whether your peaks are reasonable, whether you have contamination, and whether the dead time is acceptable. If the survey looks wrong, fix the problem before spending twenty minutes on a detailed analysis. I have walked away from supposedly good data sets after the survey showed an unexpected sodium peak that turned out to be finger print contamination from the person who prepared the sample. Keep your detector window clean. A contaminated or damaged detector window attenuates the low energy X-rays and makes light element analysis nearly impossible. Check your oxygen and carbon peaks regularly as a proxy for window condition. If the carbon peak grows mysteriously over weeks, the window is getting dirty or the detector is failing. Replace it before it ruins your light element work. Document your acquisition parameters. Acceleration voltage, beam current, live time, dead time, working distance, detector angle. These matter for reproducibility and for anyone else who might need to reanalyze your data. I once tried to reproduce a result from six months earlier and could not because I had forgotten the working distance. Changed it from twelve millimeters to eight and the absorption corrections shifted enough to change the quant by about three percent for the elements I was tracking. Small difference but enough to throw off a tight comparison.

The technique works well when you understand its limits. It is fast, it is relatively inexpensive, and it gives you usable qualitative and semi-quantitative data on the spot. It is not a replacement for WDS when you need trace level accuracy, and it is not going to tell you anything about bonding or crystal structure. Know what it can and cannot do, and you will get useful results instead of frustration.