X Ray Photoelectron Spectroscopy

Working with XPS in the lab

X-ray photoelectron spectroscopy sounds like something from a textbook, but the reality is much more mundane and occasionally frustrating. You stick a sample under an X-ray source, wait for electrons to come flying out, and then try to figure out what they mean. The instrument costs about as much as a small house and the maintenance bills don't help either. I have spent more years than I care to admit wrestling with this technique, mostly because it does not do what you want it to do until you convince it to. The core principle is the photoelectric effect, which most people learned about in high school physics without really understanding it. You shine X-rays on a material, electrons absorb that energy and escape from their atomic orbitals, and then you measure their kinetic energy. From there, you work backward to find binding energy using a simple equation. The binding energy is unique to each element and even to the chemical state of that element within the sample. That is why XPS is useful for surface analysis rather than bulk composition. The surface sensitivity comes from the fact that only electrons generated within the top few nanometers can actually make it out of the material before losing their energy through collisions. The typical escape depth is around three to ten nanometers depending on the electron's kinetic energy and the material density. This is both a blessing and a curse, because it means you are getting incredibly surface-specific data, but it also means contamination completely ruins your spectrum if you are not careful.

Getting a decent spectrum

Sample preparation is where most people fail. You need a flat, clean surface and that is harder than it sounds. If you are working with a bulk metal, you might get away with a quick polish and a solvent rinse. Powders are worse because you have to mount them without adding organic contaminants from the adhesive or tape. I usually use double-sided carbon tape and press the powder into it as hard as possible to improve electrical contact. Poor electrical contact causes charging, which shifts your peaks and makes interpretation a guessing game. Insulating samples are the real problem. When I first started, I blew through several samples trying to get reasonable spectra from ceramic materials. The charging was unpredictable and the peaks were all over the place. The standard workaround is to use a low-energy electron flood gun, but even that requires tuning. You need just enough electrons to neutralize the positive charge building up on the surface, but too much flooding broadens your peaks and degrades resolution. Finding the sweet spot usually takes twenty to thirty minutes of trial and error, and it shifts every time you change the X-ray beam size or power. Depth profiling adds another layer of complexity. Ion sputtering is the common approach for measuring composition versus depth, but it is inherently destructive and can create artifacts that look real. I once spent two weeks convinced that a titanium oxide sample had a genuine reduced titanium suboxide layer at depth. The spectrum looked perfect, the stoichiometry made sense, and everything fit my hypothesis. Then I ran a control sample with a known thickness of each oxide and realized the ion beam was reducing the TiO2 to Ti2O3 during sputtering. The artifact was coming from the analysis itself, not from the sample. The workaround was switching to a lower energy ion beam and keeping the total dose to a minimum, but even then the reduction problem was never fully gone.

Understanding what the peaks actually mean

Reading an XPS spectrum is not just about identifying elements. Chemical shifts tell you about oxidation state and local bonding environment, which is the whole point of using this technique instead of something simpler like energy-dispersive X-ray spectroscopy. A sulfur 2p peak for sulfide sits around 162 electron volts binding energy, while sulfate shifts it up to roughly 168 or 169 electron volts. Those are clear separations, but real samples rarely have only one species present. Overlapping peaks are a constant headache and they require curve fitting with multiple components. The Shirley background subtraction is the standard approach, but it is not always appropriate. For samples with strong plasmon loss features, you need a different background treatment entirely. Gold 4f peaks often show distinct plasmon satellites about one to two electron volts above the main peaks, and if you do not account for those during fitting, your quantification will be wrong. The atomic sensitivity factors used for quantification are also approximate. They assume a uniform infinite sample and ideal transmission function, neither of which is ever true in practice. The errors from these assumptions are usually in the ten to fifteen percent range for relative concentrations, which matters more than most people realize when they are trying to prove a subtle compositional difference.

Pitfalls that waste your time

Charging remains the single biggest practical problem. Even with a flood gun, conducting samples like metals can exhibit differential charging if the surface is not perfectly uniform. Oxide films on metals create insulating patches that charge differently from the bare metal areas, and the result is peak broadening or splitting that makes quantification unreliable. I have seen people report false evidence of multiple oxidation states on a supposedly pure metal surface when the real issue was uneven oxide coverage and charging artifacts. Radiation damage is another trap, especially for organic and polymeric samples. The X-ray beam itself can break chemical bonds and alter the surface composition during the time you are acquiring data. A survey scan might take one or two minutes, but a high-resolution scan on a narrow energy range can run for several minutes per pass. If your sample is sensitive, you will literally change what you are measuring while you are measuring it. The practical fix is to raster the X-ray beam across a larger area and keep the total dwell time per spot to a minimum. I typically limit high-resolution acquisitions to under five minutes per energy window for any organic-containing sample, and even that is sometimes too long. Contamination from the atmosphere is unavoidable, but it is manageable if you plan for it. A hydrocarbon contamination layer builds up on almost every sample within minutes of exposure to air. This adventitious carbon layer is actually useful as a charging reference, since the C 1s peak from sp3-bonded carbon is reliably found at 284.8 electron volts. Most labs calibrate their spectra to this peak rather than using an external standard. The problem arises when you need to analyze the actual carbon species on your sample, because the contaminant carbon overlaps with it and inflates the apparent carbon content. In those cases you have to either work under ultra-high vacuum conditions or accept that the carbon signal is contaminated.

When XPS is the wrong tool

This technique has hard limitations that people ignore at their own expense. XPS cannot detect hydrogen or helium, period. There is no way around this because the photoionization cross-section for those elements at X-ray energies is effectively zero. If you need to know whether your sample contains hydrogen, you need a different method. Neutron scattering, elastic recoil detection, or infrared spectroscopy would all serve better purposes. Light elements like lithium, beryllium, and boron are extremely difficult to quantify accurately due to low photoionization cross-sections and overlapping spectral features. Lateral resolution is another constraint. Conventional XPS instruments have a minimum probe diameter of roughly fifty to one hundred micrometers, which means you cannot analyze features smaller than that without specialized equipment. Micro-XPS systems exist but they are expensive, rare, and still struggle to go below twenty-five micrometers with good signal-to-noise. If you need sub-micron spatial resolution, you are better off with scanning electron microscopy combined with EDS or perhaps scanning probe techniques depending on what information you actually need. The vacuum requirement is non-negotiable and it eliminates whole classes of samples. liquids, wet gels, and anything that outgasses significantly under UHV conditions are basically impossible to analyze directly. You can do ambient pressure XPS on some instruments, but the resolution and sensitivity drop substantially because the gas molecules scatter the escaping electrons. If your sample is hydrated or requires a liquid environment to maintain its native state, XPS will give you a distorted picture at best.

Practical workflow for reliable results

I usually start with a full survey scan at moderate resolution to identify all detectable elements, then move to narrow scans on the specific regions of interest. The survey takes about two minutes, and I typically run two or three passes to improve the signal-to-noise ratio. For narrow scans, I use a pass energy of twenty electron volts, which gives roughly 0.05 eV per channel and sufficient resolution for most chemical state analysis. A single narrow scan on a well-behaved metallic sample takes about three minutes, but I usually acquire four scans and average them to reduce noise. Data processing should never be skipped or rushed. I always check the peak shapes before fitting, because asymmetric peaks on metals indicate either a genuine electronic effect or a charging problem. The Tougaard background is generally preferable to Shirley for most inorganic samples, though the difference is small and sometimes negligible. I fit the carbon contamination peak first to establish the charging reference, then adjust all other peaks accordingly. Quantification uses the standard relative sensitivity factors from the instrument software, but I always treat those numbers as starting points rather than absolute values. Reproducibility between instruments and laboratories is rarely better than fifteen percent for absolute concentrations. Keeping good records of acquisition parameters matters more than most people think. Beam energy, spot size, pass energy, integration time, charge compensation settings, and the number of scans all affect the final spectrum in ways that are difficult to reverse engineer later. I log everything in a spreadsheet along with the sample ID and date. When a colleague asks why their spectrum looks different from yours, the answer is almost always a different set of acquisition parameters, not a fundamental disagreement about the science.