Materials Science · Joining, coatings and surface engineering

Coatings and deposition concepts

Useful background: Processing routes and microstructural evolution; Composites, fibres and interfaces; Cracks, fracture and toughness. Each worked case explains the quantities it uses.

Back to the chapter

After reading, you can

Connect a deposition route with coating structure and coverage.
Explain how directional arrival and sequential surface reactions produce different coverage limits.
Distinguish deposited amount, local thickness, film continuity and interfacial attachment.
Evaluate a spatial thickness record and a conditional cycle window without extrapolating a flat witness to hidden regions.
Choose additional observations needed to connect a deposition route with a useful coating.

A deposited layer grows through a process

Coatings can form from vapour, liquid, particles or reaction at a surface. Arrival, attachment, transport and consolidation influence porosity, texture and adhesion. The same nominal composition can form different structures through different routes.

A substrate is the body receiving the coating; the interface is its contact with the new material. Deposition adds material, but the route must also create a connected film and an attached interface. A surface can acquire the right element while still containing separated islands, pores or a weak boundary. Thickness describes a local distance normal to the surface. It does not by itself describe whether a barrier is continuous.

Follow four questions: what reaches each location, what remains or reacts there, how material can rearrange, and what structure is left. Arrival flux is an amount reaching unit area in unit time. Adsorbed species are temporarily attached at a surface; surface diffusion lets them move along it. Their supply and motion compete with burial by later material. These mechanisms explain why a route name or nominal composition cannot specify the result.

Directional physical deposition: access precedes attachment

In evaporation, heating a source produces vapour that travels to the substrate and can condense there. In a sufficiently dilute gas, many particles reach it without a gas collision. A source seen from one face may be hidden from another. This is a useful representative physical vapour deposition (PVD) route; sputtering instead ejects source atoms by energetic bombardment. Their arrival-angle distribution and subsequent surface processes can differ. Neither name guarantees one coverage profile.

Consider an intentionally extreme geometric model: parallel vertical arrivals, sticking at first contact, with no scattering, re-emission or surface diffusion. An exposed horizontal ledge receives material. A visible horizontal trench floor also receives it; a vertical wall receives no direct normal flux. Material cannot turn a corner in this model. A partly blocked floor receives less. A vertical wall is not automatically coated just because it is physically close to a coated floor.

Real films can redistribute material through re-emission, surface motion or other processes, so this ideal shadow picture is a limit rather than a universal PVD prediction. Sputtering can have a broad arrival distribution. Rotation may change which faces see a source, but an inaccessible deep feature can remain poorly supplied. A geometric proposal needs verification on the actual part.

At the smaller scale, nuclei are small stable growing regions. Separate regions can grow together, leaving boundaries or gaps depending on the material and conditions. Protrusions can intercept supply and shadow valleys; sufficient surface motion may smooth some unevenness. Columnar morphology means elongated growth features, whereas crystallographic texture means a preference among crystal orientations. A column shape alone does not establish a crystal orientation: an amorphous film has no crystal texture. These distinctions require structure measurements, not an inference from a smooth exterior.

Original schematic of the deliberately ideal vertical-arrival model, not a measured film or an ALD profile. Parallel arrows end on exposed horizontal ledges and the directly visible floor of an open trench. Bright thick horizontal strokes mark places receiving direct normal flux. The uncoated vertical walls have normals at right angles to the arrival direction and receive zero direct normal flux in this limit. Geometric sizes and coating stroke widths are schematic. The model excludes gas scattering, re-emission and surface motion; real redistribution can change the result. A coated floor therefore does not prove coated walls.

Sequential surface reactions: saturation still requires access

Atomic layer deposition (ALD) is a contrasting chemical route. In a representative two-reactant cycle, precursor A reacts with available surface groups; excess gas and products are removed; precursor B reacts with the changed surface; gas removal prepares the next cycle. A precursor is a chemical species supplying part of the deposited material. A self-limiting half-reaction stops when its available reactive sites are consumed under the chosen conditions. It is different from simply letting a continuous arriving flux build thickness for longer.

Once an exposed location has saturated in a half-cycle, further exposure need not keep adding material there. That offers a way to supply deeper locations while the outer surface waits. Both reactants must nevertheless reach and react at the appropriate surfaces throughout the feature; by-products and excess reactants must also be removed. A narrow passage with a large internal area can consume supply near its entrance. Saturation at a witness outside it says nothing by itself about saturation far inside.

Conformal means that the coating follows the underlying shape with comparable normal thickness over the locations being considered. ALD can give very good conformity, but the mechanism does not guarantee it for every chemistry, pore or exposure. A cycle is not a promise of exactly one complete atomic plane. Nucleation, site availability and the surface chemistry can make early growth differ from later growth; even self-limiting chemistry can leave a discontinuous initial film. Verify any constant growth-per-cycle approximation over its stated range.

A primary platinum/alumina-membrane study illustrates the limit: the investigators changed exposure and temperature and examined internal coverage using cross-sectional elemental observations together with neutron scattering. Planar and internal thickness estimates were not identical, and some central coverage remained nonconformal. This is evidence for inspecting internal structure, rather than a universal setting or proof that any ALD-coated recess is uniform.

Coverage must match the geometry

Recesses, edges and shadowed regions may receive different thicknesses. A coating can alter roughness and residual stress as well as chemistry. Check spatial thickness and defects where the intended protection or optical response is needed.

A witness is a separate sample used to monitor a process. It is useful when its relation to the relevant part location is established. The original recess question therefore needs spatial evidence: compare the exterior, entrance, deeper walls, floor, edges and obstructed regions as appropriate. Use local thickness measured normal to each surface, and state the measurement footprint. An average through a large patch can hide a smaller bare or thin region.

The next record is entirely fictional. Locations W, E and D are an exposed flat witness, an upper recess wall and a deeper wall. The part geometry and coating chemistry are fixed. At 200 cycles, two different trial exposure histories give the values below. W is not a third wall or an estimate of whole-part average thickness. The invented local thickness screen is 80–110 nm, inclusive, at the sampled sites. It is not a barrier standard or a safety criterion.

Invented resolved local thicknesses at 200 cycles; no uncertainty is included in this first comparison.
Location	History A / nm	History B / nm	What the observation can establish
W: exposed witness	100	100	Both have the same exposed value.
E: upper wall	80	95	Different internal profiles occur despite the same W.
D: deeper wall	25	85	A fails the stated local screen; B passes at these sampled sites.

A duration adjustment cannot repair every spatial profile

One record at 200 cycles does not establish a linear growth law. For the next calculation only, separately suppose repeated observations support hᵢ = gᵢN for each of these three locations throughout 100 ≤ N ≤ 700 cycles, with fixed geometry and chemistry, negligible uncertainty and no initial offset. Here hᵢ is local thickness in nm, gᵢ is its constant thickness increment in nm/cycle and N is an integer cycle count. This invented local fit is not a general ALD law or a statement that a partly starved process is saturated.

The supplied increments are A: (0.500, 0.400, 0.125) nm/cycle and B: (0.500, 0.475, 0.425) nm/cycle in W, E, D order. Each site must satisfy 80 ≤ gᵢN ≤ 110. Divide by its positive gᵢ to obtain its allowed cycle interval, then intersect the intervals. A common interval means that the sampled thickness constraints are compatible within the model; it does not demonstrate film continuity, adequate chemistry, adhesion or a working process.

History A cannot meet the three-site screen by extending cycles alone: its deep-wall minimum needs at least 640 cycles, while its witness permits at most 220. At 640 cycles the witness would be 320 nm. The ratio h_D/h_W stays 0.25 under the stipulated fixed profile. Improving access or changing the process history could change this ratio, but would require new observations and a new fit. For history B the deepest wall sets the minimum and the witness the maximum; integer counts 189 through 220 satisfy this particular screen.

If a recess becomes narrower during deposition, the transport and local increment can change. If a new surface delays nucleation, the zero-offset model can fail. If growth is comparable to the feature width, fixed geometry needs reconsideration. These are reasons to limit a fitted model, not to extrapolate a passing witness more confidently.

Original fictional thickness records and separately stipulated constant-increment model. Upper chart compares history A as solid blue bars and history B as orange hatched bars, at 200 cycles: witness W is 100 nanometres for both, upper wall E is 80 versus 95, deeper wall D is 25 versus 85. All bars use a common zero-to-120-nanometre scale. The shaded 80-to-110-nanometre band is the invented local screen. Lower chart gives cycle intervals within the independently stated fitted domain 100 to 700 cycles. For A, W allows 160 to 220, E 200 to 275, and D 640 to 700 after clipping to the fitted domain, so their intersection is empty. For B, W allows 160 to 220, E approximately 168.42 to 231.58, and D approximately 188.24 to 258.82; their intersection is approximately 188.24 to 220, or integer cycles 189 through 220. These are constraints on the same cycle count, not consecutive stages. Passing them concerns sampled thickness only, not complete coating function.

Amount, thickness and useful function are different accounts

A deposited mass is an integral amount rather than a local coverage map. For a single known solid of density ρ_s, its solid volume is m/ρ_s. If material covers area A, the dense-equivalent mean thickness is h_eq = m/(ρ_sA). This is the thickness the same solid volume would make if spread densely and uniformly. It need not be any actual local thickness.

If the coating has a known uniform pore fraction φ and covers all of that area, the geometric mean thickness would instead be h_geo = h_eq/(1 − φ). This extra relation assumes void and solid volumes add and the same mass belongs entirely to that coating. Pores increase geometric volume at fixed solid amount. A missing patch, a density change or material outside the counted area invalidates the simple inference. No mean value proves the least-coated hidden location.

To decide what a coating does, connect the intended function to evidence. A barrier requires relevant continuity and defect observations; a conductive or optical film requires its appropriate response and structure. Interface attachment depends on the substrate surface and growth history. A weak intervening layer can detach while the coating itself remains thick. Coverage evidence and interfacial evidence answer different questions; a thickness screen cannot replace a measured functional or adhesion response.

Sampling and uncertainty constrain the claim. A destructive cross-section can expose internal geometry, but only at the cut and measurement footprint. An area-averaging technique can complement it without resolving every local defect. Record site selection and uncertainty. Additional hidden sites, rough edges, pores, chemistry, structure and relevant response may change the conclusion. A process route guides which failure to look for; it does not certify a service life.

Worked examples

Work through the case

A flat witness sample has acceptable thickness. Does it establish coating coverage inside a deep recess?

The recess has different access and geometry.
Deposition can be direction- or transport-dependent.
Measure representative hidden regions.

No. A flat witness alone may not represent that location.

Use the hidden-location record

Apply the invented 80–110 nm local screen to the three sampled sites at 200 cycles. Compare deep-to-witness coverage for A and B.

History A gives W = 100, E = 80 and D = 25 nm. W and E lie in the inclusive window; D is 55 nm below its minimum. The three-site screen therefore fails.
History B gives W = 100, E = 95 and D = 85 nm, all within the window at these sites. Its smallest margin is 85 − 80 = 5 nm.
Deep-to-witness ratios are 25/100 = 0.250 for A and 85/100 = 0.850 for B. They are spatial thickness ratios, not fractions of the part area proven covered.
Both witnesses read 100 nm. Their agreement cannot distinguish the internal histories; B is a candidate under this sampled thickness screen, not proof of a defect-free functional coating.

A fails at D; B passes the stated sampled screen. Ratios 0.250 and 0.850 describe these two profiles only.

Intersect local constraints before adding cycles

Using the separately stipulated constant increments, is there an integer cycle count meeting 80–110 nm at all three sites for each history?

For A, W permits 160 ≤ N ≤ 220; E permits 200 ≤ N ≤ 275; D permits 640 ≤ N ≤ 880. Only N up to 700 is within the stated fitted domain, and there is no common interval even there.
For B, W permits 160 ≤ N ≤ 220; E permits 168.421… ≤ N ≤ 231.579…; D permits 188.235… ≤ N ≤ 258.824….
Intersect B’s intervals to get 188.235… ≤ N ≤ 220. An integer count must therefore satisfy 189 ≤ N ≤ 220. At 189 the three thicknesses are 94.5, 89.775 and 80.325 nm; at 220 they are 110, 104.5 and 93.5 nm.
Increasing A to 640 cycles makes D = 80 nm but W = 320 nm, violating the upper limit. Changing its spatial access, rather than this single duration variable, is needed before testing a revised candidate.

No count works for fixed-profile A. B has the conditional integer window 189–220 cycles; further coating and process evidence is still required.

Interpret an amount without inventing a local minimum

A fictional coating contains 8.00 mg of one solid with ρ_s = 4.00 g/cm³ over a counted 10.0 cm² area. Find its dense-equivalent mean thickness. What changes if uniform φ = 0.20 is separately known?

Convert the mass: 8.00 mg = 0.00800 g. Solid volume is 0.00800/4.00 = 0.00200 cm³.
Divide by area: h_eq = 0.00200/10.0 = 0.000200 cm. Since 1 cm = 10⁴ µm, h_eq = 2.00 µm. This is a different illustrative account from the 200-cycle record.
With full coverage and uniform φ = 0.20, geometric volume is 0.00200/0.80 = 0.00250 cm³ and h_geo = 2.50 µm. The solid amount is unchanged.
The same mass could be unevenly distributed, leaving a hidden site bare. Neither value establishes a local minimum, film continuity, the actual density without independent evidence, or adhesion.

The dense-equivalent mean is 2.00 µm; the extra uniform-porosity assumptions give 2.50 µm geometric mean. Neither is a coverage map.

Try the reasoning

Does composition alone establish adhesion?
Show the answer and reasoning
No.
Interface preparation and growth conditions matter.
Why inspect edges?
Show the answer and reasoning
Coverage and stress can differ there.
Local defects can control service failure.
In the ideal vertical-arrival, first-contact model, a visible horizontal trench floor is coated. Does this imply its vertical walls are coated? Explain what changing the arrival distribution could do.
Show the answer and reasoning
No: the ideal vertical beam has no direct normal flux onto the vertical walls.
The floor faces the beam and the walls do not. Oblique arrivals or redistribution may supply a wall, while shadowing can still block deeper regions. A changed model requires new geometric and coverage evidence.
The same 80–110 nm window is required. A new fixed profile has g_W = 0.50 and g_D = 0.30 nm/cycle. Can any count satisfy both sites?
Show the answer and reasoning
No. D needs at least 266.667… cycles (267 integer); W permits at most 220.
The common interval is empty. Equivalently, h_D/h_W = 0.60, while any pair meeting this window requires at least 80/110 = 0.72727… for that ratio. At N = 220 the deep site reaches only 66 nm. This ratio criterion is necessary for these two ordered sites, not proof about other locations.
For history B, suppose a fourth sampled site has g_X = 0.35 nm/cycle under the same fitted-domain assumptions. Does B’s former common window remain?
Show the answer and reasoning
No. X needs N ≥ 228.571… (229 integer), above W’s maximum 220.
At the largest former count X is only 77 nm. Adding an observed site can invalidate a sampled screen; the original three sites have not changed. A three-site pass never established all-site coverage.
At 200 cycles, B’s deep-site estimate is 85 nm. Suppose a defensible uncertainty analysis only establishes that its true value lies within 78–92 nm. Does it establish the 80 nm minimum?
Show the answer and reasoning
No: the supplied interval includes values below 80 nm.
The point estimate lies within the screen, but the lower bound does not. An uncertainty model or additional evidence is needed for the decision. An interval is not automatically a probability distribution or an assurance about unmeasured locations.
The 8.00 mg solid amount is now assigned to 20.0 cm² instead of 10.0 cm², with unchanged solid density. Find h_eq and, only with the same separate uniform φ = 0.20 assumption, h_geo.
Show the answer and reasoning
h_eq = 1.00 µm and h_geo = 1.25 µm.
The solid volume remains 0.00200 cm³. Dividing by twice the area halves each mean. The calculation does not demonstrate that the new area is actually all covered.
An exposed witness has reached a repeatable growth-per-cycle plateau as precursor A exposure is increased. Can you conclude both half-reactions saturate inside a newly narrowed recess?
Show the answer and reasoning
No. The observation concerns one reactant history at one exposed location.
A changed passage can change access. Reactant B, internal supply, removal, nucleation and the actual internal coating still need evidence. A surface-limiting mechanism is compatible with incomplete supply elsewhere.
Two films have identical sampled thickness and composition. One is continuous; the other has open gaps between growth regions. Can their barrier behaviour be inferred to match?
Show the answer and reasoning
No. An open path can bypass otherwise thick coating.
Thickness and nominal composition do not encode connected defects. Compare the relevant continuity/transport response and interfaces. Columnar shape also does not establish crystallographic texture or adhesion.

What to carry forward

Processing establishes coating architecture.
Validate coverage on the actual geometry.
Directional access and surface reaction saturation impose different local growth limits.
A flat witness, mean amount and local thickness answer different questions.
Intersect local constraints only within a separately supported model; more cycles cannot repair every coverage ratio.
Passing sampled thickness constraints leaves continuity, interface attachment and useful response to be demonstrated.

References

Teaching resources and research supporting this lesson’s explanations and examples.