A map from the question you are actually asking to the technique that answers it, what each one can and cannot tell you, what it demands from the sample, and the order in which to run them so the specimen survives.
A materials group writes a proposal, lists a scanning electron microscope and an X-ray diffractometer because those are the instruments everyone lists, and wins the grant. Eighteen months later the research question has settled into something about carrier concentration in a doped oxide, which neither instrument measures. The equipment is good. The match between the question and the instrument was never made, and it is now a five year problem.
Characterisation money is spent well when the order runs question first, technique second, instrument third. That order is easy to state and hard to keep, because instruments are exciting and questions are not. What follows is the map, arranged by the question rather than by the manufacturer.
Before comparing techniques, fix three properties of each one. They explain most of the disagreements between two honest measurements of the same sample.
| The question | Techniques that answer it | What the sample has to be |
|---|---|---|
| Which elements, and roughly how much | EDS on an electron microscope, X-ray fluorescence, wet digestion followed by solution analysis | Flat and conductive for EDS; a few grams if you are digesting it |
| Which compound or phase | X-ray diffraction, Raman, infrared spectroscopy | Powder, a flat face, or a clean area a few micrometres across for Raman |
| Which iron site, valence and magnetic order | Mossbauer spectroscopy | Iron bearing powder of controlled areal density, or a foil |
| Surface topography and local mechanics | Atomic force microscopy | Flat, clean, firmly mounted, in a room that is not drifting |
| Morphology and interfaces in cross section | SEM or TEM on a polished or ion milled section | A section prepared without smearing the soft phase |
| Layer thickness and optical constants | Spectroscopic ellipsometry, reflectometry, masked step on a profilometer | Smooth, with a stack you can write down |
| Sheet resistance and resistivity | Four point probe, van der Pauw | Uniform, contactable, with an independently known thickness |
| Carrier type, density and mobility | Hall effect measurement | Defined geometry with ohmic contacts |
| Magnetic moment, coercivity, hysteresis | Vibrating sample magnetometry | A small accurately weighed piece in a low signal holder |
| Transitions, heat flow and mass loss | DSC, TGA, simultaneous thermal analysis | Milligrams in a crucible that will not react with it |
| Thermal diffusivity and conductivity | Laser flash analysis, hot disk, guarded heat flow | A disc of known thickness and density, often coated with graphite |
| Expansion, softening and sintering shrinkage | Dilatometry, thermomechanical analysis | A rod or pellet with flat parallel ends |
| Emission spectrum and excited state lifetime | Photoluminescence spectroscopy, time correlated single photon counting | Anything that emits, mounted so it does not move or bleach |
| Redox behaviour and electrode kinetics | Potentiostat with a three electrode cell | A working electrode, an electrolyte and a reference you trust |
Energy dispersive X-ray spectroscopy on an electron microscope is the default answer because the detector is already on the instrument. It is also the most over-interpreted result in materials science. EDS reports which elements are present in the interaction volume and, with care, their approximate proportions. It does not report chemistry. Finding oxygen and titanium together is not evidence of titanium dioxide, and a reviewer will say so.
Three limits are worth writing on the wall next to the microscope. Light elements are difficult and lithium is out of reach. The interaction volume is of the order of a micrometre, so a hundred nanometre film on silicon gives you a spectrum dominated by silicon. And quantification without standards is a calculation based on assumptions about the sample, not a measurement, so quote it as approximate or measure standards.
When you need composition as a number rather than an indication, the route is chemical: digest the material and analyse the solution, or use X-ray fluorescence against matched standards. When you need chemistry rather than elements, move to a vibrational technique. Raman identifies compounds and polymorphs, distinguishes anatase from rutile, and reads strain and defect density in carbon materials. Infrared spectroscopy covers the functional groups that Raman is weak on. For iron bearing materials specifically, Mossbauer spectroscopy resolves oxidation state and site occupancy in a way no other bench technique manages.
Structure splits into three questions that people run together: which crystalline phase, how the grains are arranged, and what the surface looks like. Diffraction answers the first, electron microscopy the second, scanning probe the third.
The common trap is treating diffraction as a complete structural answer. A diffractometer sees crystalline material. An amorphous fraction contributes a broad hump and is easy to dismiss, so a sample that is half glass can be reported as a single phase. Raman sees the amorphous fraction clearly, which is why the two techniques appear to disagree. They do not. Run both on a new material and treat the combination as the answer.
For grain size and morphology, a scanning electron microscope answers quickly on a fracture surface or a polished section. Below the tens of nanometres range, and for anything requiring lattice imaging or local diffraction, a transmission electron microscope is the only honest route, and it brings a sample preparation burden that is larger than the microscope itself. Focused ion beam lift out for a site specific lamella is a skilled job with a long training tail; budget the operator time, not just the instrument.
A crystal monitor inside a deposition chamber is a rate controller, not metrology. Thickness has to be measured on the finished sample, and the choice is between a direct geometric measurement and an optical model.
If the sample is a deposited film, the thickness question sits inside a longer sequence covering roughness, composition, crystallinity and adhesion, and running that sequence in order saves repeated preparation. The order matters more than the instrument list.
A four point probe gives sheet resistance. That is a per square quantity, and converting it to resistivity requires the thickness, so an error in thickness propagates straight into the resistivity you publish. The four contact arrangement removes lead and contact resistance from the voltage reading, which is why it beats a two probe multimeter, but it removes nothing else: geometry, edge proximity, probe spacing against sample size, temperature and illumination all remain.
A four point probe cannot tell you whether the carriers are electrons or holes, how many there are, or how mobile they are. That is the Hall measurement, which needs a defined geometry, ohmic contacts and a magnetic field. The combination of resistivity and Hall coefficient separates carrier density from mobility, and that separation is usually the point of the experiment: two samples with the same resistivity can differ by an order of magnitude in mobility, and only one of them is a good device.
Temperature dependent transport is where the mechanism appears. Activated behaviour, hopping and metallic conduction look different as the sample cools, and a room temperature number alone cannot distinguish them. That argues for a cryogenic probe station rather than a bench probe if the group works on semiconductors or oxides. For local rather than average conduction, conductive atomic force microscopy maps current at the tip and scanning spreading resistance microscopy profiles dopant across a polished cross section.
A vibrating sample magnetometer gives the total moment of the specimen against applied field: saturation magnetisation, remanence, coercivity, and the shape of the loop. It is a bulk average, so it tells you what the sample does and not where the moment lives. Two practical points decide whether the loop is publishable. The sample mass has to be known accurately, because magnetisation per gram inherits every weighing error. And the holder and the substrate contribute a signal of their own, usually diamagnetic and linear, which has to be measured and subtracted rather than assumed to be negligible on a thin film.
When you need to know which phase carries the moment, magnetometry alone will not answer. In an iron bearing sample, Mossbauer spectroscopy separates the contributions by site and can identify a magnetically ordered phase that magnetometry has smeared into the average. For surface domain structure, magnetic force microscopy maps the stray field gradient above the sample, which is a qualitative picture rather than a moment. Field measurement itself, with gaussmeters, teslameters and fluxmeters, is a different job again and belongs to whoever is characterising magnets rather than materials.
Thermal analysis is the part of the characterisation stack most often bought last and used most. Four instruments cover most questions.
Evolved gas analysis, where the exhaust of the thermal instrument goes to a mass spectrometer or an infrared cell, converts a mass loss step into an identified species. It is the difference between reporting that the sample lost mass at some temperature and reporting what came off. If the group works on hydrated materials, polymers or anything with a binder, specify the coupling at purchase rather than retrofitting it.
Cooling a sample is not an exotic requirement. It is how you freeze out phonon scattering, resolve transitions that are washed out at room temperature, measure activation energies from the slope of a conductivity curve, and separate a superparamagnetic particle from a genuinely non-magnetic one. Cryogenic sensors, controllers and probe stations exist for exactly this.
For a laboratory in Bangladesh the decision is dominated by the cryogen supply. Liquid helium has to be imported, arrives with losses, and cannot be relied on for a scheduled experiment. A closed cycle refrigerator removes that dependency and replaces it with a dependency on cooling water and stable power, both of which are local problems you can solve. Plan for a closed cycle system unless you have a proven helium route, and specify the cooling water circuit and the power conditioning in the same purchase order as the cryostat, not afterwards.
Most wasted characterisation effort comes from running the destructive measurement first. Fix the order at the start of the campaign.
Write the plan down before the first measurement, including which piece of the sample goes to which instrument. A single deposition run that yields four coupons rather than one wafer removes most of the conflict.
A department starting a characterisation capability should buy against its own research question rather than against a generic list. Three rules hold up in practice.
Buy sample preparation before the second large instrument. A microscope with no polishing, cutting, coating or ion milling capability produces poor images from good optics, and the preparation equipment costs a fraction of the microscope. The same applies to thermal analysis: crucibles, a good balance and a reference set are what make the instrument quantitative.
Buy the consumable chain with the instrument. Cantilevers, probe tips, crucibles, sealed sources, standards and reference samples are all imported, and a five week lead time on a cantilever stops an atomic force microscope as effectively as a broken controller. Ask for a first year consumable schedule at quotation stage and hold a working stock.
Write the tender so that more than one supplier can bid. Specifications lifted from a single brochure produce a single compliant bid on e-GP, which delays the award and invites a challenge. State the measurement requirement, the sample types, the environment and the service response you need, then let the technical committee compare. If a requirement genuinely rules out every instrument but one, say why in the justification rather than hiding it in a parameter.