Thin film characterisation: which technique answers which question, and in what order

A deposition run finishes and nobody in the building can say whether the film is 80 nanometres or 140. Here is how to build a thin film measurement workflow, technique by technique, instead of buying instruments one grant at a time.

A sputtering run finishes at eleven at night, the student writes 150 nm titanium dioxide in the logbook because that is what the quartz crystal monitor said, and nine months later a reviewer asks how the thickness was measured. Nobody in the building can answer. This happens in well-funded laboratories too, because deposition equipment is bought as a capability while measurement is bought as an afterthought. The fix is not a bigger budget. It is deciding, before the first run, which instrument answers which question.

Five questions every film has to answer

Whatever the deposition route, the film has to survive the same interrogation before anyone can build a device or publish a claim.

1. How thick is it, and how uniform is that thickness across the substrate?
2. How rough is it, and what does the surface and cross-section morphology look like?
3. What is it made of, both which elements and in what chemical state?
4. Is it crystalline, and if so in which phase, with what grain size and texture?
5. Will it survive, meaning residual stress, adhesion, hardness and the functional property you actually care about.

No single instrument answers more than two of these. That is the whole reason a characterisation plan matters more than any one purchase.

Thickness: the number everyone quotes and few people measure

A quartz crystal monitor inside the chamber is process control, not metrology. It reports mass loading converted through a tooling factor and an assumed density, both of which were set once and probably not since. Treat it as a rate controller and verify it against a real measurement at the start of every material campaign and after any chamber clean or target change.

• Masked step with a stylus profilometer or AFM. The most direct answer. Mask a corner of the substrate with tape or a shadow mask on every run, then measure the step. The AFM gives a better number on thin, smooth films; the profilometer covers thicker and rougher ones and is far quicker.
• Spectroscopic ellipsometry. Fast, non-contact and excellent on smooth dielectric films. It fits an optical model, so it returns thickness and refractive index together, and for very thin or strongly absorbing films those two can trade off against each other. Anchor the model against a step measurement once per material.
• X-ray reflectivity. Gives thickness, density and interface roughness at the same time, which is more information than anything else on this list. It needs a smooth film and a diffractometer configured with the right optics, so it is not a casual measurement.
• Cross-section SEM or FIB lamella. Direct, visual, convincing to reviewers, and destructive. Also the only route that shows you a columnar microstructure or a buried void.
• Optical reflectometry. Cheap and quick for transparent films on a known substrate, and completely unreliable when you do not know the optical constants.

One habit fixes most of this: cut a witness coupon from the same substrate material, put it in the chamber on every run, mask part of it, and store it. When a question arises two years later you still have the sample.

Roughness and morphology

Roughness is scale-dependent, which means a roughness value without a scan size attached to it is meaningless. An RMS roughness of 1.2 nm over a 1 by 1 micrometre AFM scan and the same figure over a 50 by 50 micrometre scan describe two completely different surfaces. Fix a standard scan size for each material class in your laboratory, write it into the standard operating procedure, and quote it in every table and figure caption.

• AFM for quantitative RMS and Ra, grain height distribution and step heights.
• SEM for morphology: grain shape, pinholes, cracks, particulates, and in cross-section the columnar or dense structure that explains your electrical results.
• White light or confocal profilometry for larger areas and rougher coatings where the AFM z range runs out.
• Optical microscope with differential interference contrast before any of the above, because it costs nothing and catches gross defects in thirty seconds.

Composition and chemical state

This is where thin film work goes wrong most often. EDS on a scanning electron microscope is the default because it is already attached to the instrument, but the electron interaction volume at a normal accelerating voltage extends a micrometre or more into the sample. On a film of a hundred nanometres you are mostly analysing the substrate and reporting it as the film. Lower the accelerating voltage until the excitation volume sits inside the layer, accept the loss of the higher energy lines, and state the voltage in the paper. If you cannot get the volume inside the film, EDS is a qualitative check, not a composition measurement.

• XPS for the top few nanometres and, more usefully, for oxidation state: whether your titanium is metal, suboxide or dioxide, whether your nitride is nitrogen-rich.
• Raman for bonding, phase and stress. A Renishaw inVia system distinguishes anatase from rutile, amorphous from nanocrystalline silicon, and reads graphene layer number and defect density from the D, G and 2D bands. It works through transparent substrates and needs no vacuum.
• FTIR for hydrogen content and bonding configuration in PECVD silicon nitride and oxide, which is the parameter that decides whether the film passivates or blisters after annealing.
• SIMS or RBS for a depth profile or absolute stoichiometry. Realistically these are sent out. Build a working relationship with a laboratory that runs them rather than waiting for a grant that will not come.

Crystallinity, phase and texture

A standard theta and two-theta X-ray scan on a thin film mostly measures the substrate, because the beam penetrates through the layer. Grazing incidence geometry keeps the beam in the film and is the correct configuration for anything below about a micrometre. If your diffractometer cannot do grazing incidence, that is a capability gap worth listing on the next tender, not something to work around.

Raman is the fast partner to XRD: it identifies phase in seconds, needs no special geometry, and picks up amorphous content that diffraction does not see. Where the two disagree, the film is usually mixed phase, which is itself the result.

The properties that decide whether the device works

• Residual stress from substrate curvature before and after deposition. High compressive stress buckles; high tensile stress cracks. Either will pass every other measurement in this article and still fail on the bench.
• Adhesion by scratch or tape testing. Crude, standardised and far more predictive of device lifetime than anything else you will measure in a morning.
• Hardness and modulus by nanoindentation, with the usual warning that the indent must stay within roughly a tenth of the film thickness or you are measuring the substrate again.
• Sheet resistance and carrier concentration by four point probe and Hall measurement.
• Transmittance and band gap by UV-Vis spectrophotometry for optical and photovoltaic films.

A map from question to technique

Question	First choice	Cross-check with	What it will not tell you
Thickness	Masked step on AFM or profilometer	Ellipsometry or X-ray reflectivity	Whether the film is uniform away from the step
Uniformity across the substrate	Ellipsometry mapping	Multiple masked steps	Anything about composition
Surface roughness	AFM at a fixed scan size	White light profilometry	Buried interface roughness
Which elements	EDS at reduced accelerating voltage	XPS or SIMS	Chemical state or bonding
Oxidation state and bonding	XPS	Raman, FTIR	Anything deeper than the top few nanometres
Crystalline phase	Grazing incidence XRD	Raman	Amorphous fraction, in most cases
Residual stress	Substrate curvature	XRD peak shift	Adhesion strength
Electrical quality	Four point probe and Hall	Cross-section SEM	Why the number is what it is

Measure in this order

Sequence matters because measurements contaminate each other. Carbon coating a sample for SEM makes it useless for XPS and adds a fluorescence background to Raman.

1. Optical inspection and photograph. Free, and it catches the failed runs before you spend instrument time on them.
2. Non-contact optical metrology: ellipsometry, reflectometry, UV-Vis.
3. AFM and profilometry, which touch the surface lightly but leave it usable.
4. XRD and Raman, non-destructive on most inorganic films.
5. Electrical measurements, which may require contacts and therefore commit the sample.
6. SEM with EDS, which usually means coating.
7. Destructive work last: cross-section, indentation, SIMS, adhesion testing. Use the witness coupon here, not the device sample.

Building a capability instead of a shopping list

The common pattern is three separate grants producing three instruments over five years, with no sample preparation bench, no reference standards, no software licences beyond the first year and one trained operator who then leaves for a PhD abroad. The instruments work. The measurements are still not comparable between students, and nobody can answer a reviewer's question about traceability.

• Buy standards with the instrument. A certified step height standard, a reference silicon wafer for roughness, a thickness reference for the ellipsometer and a diffraction standard. These items are inexpensive at purchase and effectively unobtainable once the tender has closed.
• Write the standard operating procedure before the first student. Fixed scan sizes, fixed accelerating voltages, fixed acquisition times. Comparability between years is worth more than any single optimised measurement.
• Decide what you will always send out. Name the two or three techniques you will never own, find a partner laboratory, and build that relationship deliberately rather than treating each shipment as an emergency.
• Budget software and licences across the equipment lifetime, including the analysis package, not only the acquisition software that came bundled.
• Log the deposition parameters and the measurement date together. Films age. A silicon surface measured three weeks after deposition in a humid room is not the film that came out of the chamber.

What Bangladeshi conditions change

• Humidity between deposition and measurement. Through the monsoon, an unprotected metal or nitride surface adsorbs water and oxidises measurably. Store coupons in a nitrogen dry cabinet or vacuum desiccator and record the storage interval as part of the result.
• Power interruption during a run. An etch or PECVD process interrupted halfway is a scrapped wafer and often a chamber clean. Put the tool controller, mass flow controllers and gas panel on an online UPS, and confirm what the system does on power loss before you rely on it. Systems such as the Oxford PlasmaPro reactive ion etcher and Oxford Instruments sputter and PECVD platforms have defined safe states, but only if the control electronics stay alive long enough to reach them.
• Process gas logistics. Specialty gases and the correct regulators have long lead times into Dhaka. Order the second cylinder when you open the first, not when it runs out.
• Lithography environment. Mask aligners such as the SUSS MicroTec systems need temperature and humidity control for repeatable resist behaviour. Resist that has been through three warm days behaves differently from fresh resist, and the resulting line width variation will be blamed on the aligner.
• Sample shipment abroad. Customs paperwork and courier handling add weeks. Build that time into the project plan rather than discovering it at the thesis deadline.

Back to all Insights