Thin Films, Epitaxy, and Superlattices

From a buried layer to a grown film

The last guide sandwiched a five-nanometre slab of one semiconductor between two others and called it a quantum well — electrons penned into a two-dimensional sheet. Strip away the confinement story and look at what you physically have: a thin film. A layer, anywhere from a few atoms to a few hundred nanometres thick, sitting on a far thicker slab called the substrate. In a bulk crystal every atom is surrounded on all sides; in a film, one dimension is so short that the free surface on top and the interface with the substrate below stop being a rounding error and start running the show. It is the surface-to-volume argument from the first guide of this rung, now sharpened to a single direction.

How is such a film born? Atom by atom, from a vapour or a molecular beam. Arriving atoms land on the surface, skitter around, and either settle where they can lower their energy or hop off again. Whether they spread into a smooth carpet or bead up into islands is decided by a tug-of-war between three surface energies — the bare substrate surface, the new film surface, and the interface between them. Three classic outcomes follow: layer-by-layer growth (Frank-van der Merwe) when the film wets the substrate; three-dimensional islands (Volmer-Weber) when it does not; and layer-then-islands (Stranski-Krastanov) when it starts out wetting but strain later forces it to bunch up. Hold onto that last one — strain-driven islanding will come back at the end as a trick for making quantum dots.

Epitaxy: growing in registry with the substrate

When a crystalline film grows so that its lattice lines up with — is in structural registry with — the crystal beneath it, that is epitaxy (Greek epi, 'upon', plus taxis, 'arrangement'). The substrate acts as a template: its surface plane and its in-plane atomic spacing are copied by the first layer of film, which templates the next, and so on upward. Two flavours matter. Homoepitaxy grows film and substrate of the same material — silicon on silicon, which is how the pristine active layer of nearly every microchip is laid down — and registry is automatic. Heteroepitaxy grows a different material on top, and now the film inherits a definite orientation relationship with the substrate: a (001) film on a (001) substrate with [100] locked to [100], for instance.

Why care so much? Because a film that copies its substrate can be a single crystal over an entire wafer — no grain boundaries anywhere — which is precisely why the semiconductor industry lives and dies by epitaxy. The junction between the two crystals can itself be graded by how faithfully their lattices meet across it. A coherent interface has every film plane matched one-to-one with a substrate plane; a semicoherent interface mostly matches but slips in a defect now and then to take up the slack; an incoherent interface has given up on registry altogether. Which one you get comes down to a single number: how badly the two lattice parameters disagree.

Misfit, strain, and the critical thickness

That disagreement has a name and a formula: the lattice misfit, f = (a_film - a_substrate)/a_substrate. Put real numbers on it. Grow germanium (a = 5.658 angstrom) on silicon (a = 5.431 angstrom) and f = (5.658 - 5.431)/5.431 = 0.042 — a 4.2 percent misfit. Grow gallium arsenide (5.653 angstrom) on aluminium arsenide (5.661 angstrom) and f is a mere 0.14 percent, a near-perfect match, which is exactly why the AlGaAs/GaAs pair became the workhorse of epitaxial devices. Push to indium arsenide (6.058 angstrom) on GaAs and f leaps to 7.2 percent. That one number, the mismatch of two lattice parameters, decides how a film must behave.

A thin enough film has no choice: to keep registry it stretches or squeezes its own in-plane spacing to match the substrate exactly, storing elastic strain. This is pseudomorphic (coherent) growth, and the interface stays perfect. But the trap is that strain energy piles up with every layer you add — a thicker film stores more of it — while the cost of relieving the misfit with a dislocation is roughly fixed. So there is a critical thickness h_c: below it, staying strained is the cheaper deal; above it, the film relieves misfit by throwing in dislocations. Roughly, h_c scales like b/f — the bigger the misfit, the thinner the film you can grow before it lets go. For a 1 percent misfit, h_c is a few nanometres; for the 4.2 percent Ge-on-Si case it is only a handful of atomic layers. When the film does relax, the leftover mismatch drains into a grid of misfit dislocations lying right in the interface — edge dislocations whose extra half-planes soak up the slack, roughly one every b/f rows (about one every 24 rows at 4.2 percent). The interface has gone from coherent to semicoherent, and the film sits close to its own natural spacing again.

Compute the misfit from the two lattice parameters: f = (a_film - a_substrate)/a_substrate.
Estimate the critical thickness, roughly h_c ~ b/f: a few atomic planes for a large misfit, tens of nanometres for a small one.
Compare your intended film thickness h with h_c. If h < h_c the film grows coherent and strained (a pseudomorphic layer); if h > h_c it will relax.
On relaxing, expect a grid of misfit dislocations lining the interface about every b/f rows — one extra half-plane absorbs one row of mismatch — and the interface turns semicoherent.

EPITAXY  --  film atoms (o) grown on substrate atoms (x), seen edge-on

(a) COHERENT / PSEUDOMORPHIC     (thin film, h < h_c)
      o   o   o   o   o   o        the film's in-plane spacing is
      |   |   |   |   |   |        FORCED to match the substrate:
    --o---o---o---o---o---o--      elastic STRAIN stored, interface
      x   x   x   x   x   x        flawless, zero dislocations
      x   x   x   x   x   x

(b) RELAXED                       (thick film, h > h_c)
      o  o  o  o  o  o  o  o       the film springs back to its OWN
      |  |  |  |  |  |  |  |       spacing; leftover misfit is dumped
    --o--o--o--o--o--o--o--o--     into interface dislocations
      x  x  x  x  T  x  x  x       T = misfit dislocation (edge type),
      x  x  x  x  x  x  x  x       one extra half-plane every ~ b/f rows

Epitaxy in one picture. A film below the critical thickness has no choice but to adopt the substrate's in-plane spacing, storing elastic strain in a flawless coherent interface (a). Grow past the critical thickness and it becomes cheaper to spring back to the film's natural spacing and shed the leftover mismatch into a grid of misfit dislocations lining the interface, roughly every b/f atomic rows (b). The sketch is schematic — a real interface is a two-dimensional grid of such lines.

When the film is not a single crystal: texture

Epitaxy is the aristocratic case, but most films are not grown that way. Sputter or evaporate a metal onto glass, or an oxide onto steel, and the substrate is not a single crystal to copy — often it is not crystalline at all. The film then nucleates as a swarm of separate crystallites and grows into a polycrystal, typically as columnar grains stacked like a forest of pencils standing on end. What saves this from being pure chaos is that the grains are usually not randomly oriented. Growth conditions favour some crystallographic directions over others — the fastest-growing or lowest-surface-energy planes tend to end up facing outward — so the grains share a preferred orientation even while their in-plane directions are scrambled.

That shared preferred orientation is texture, and for films it usually takes the form of a fibre texture: one crystal axis (say [111]) points reliably along the film normal, but the grains are free to spin about that axis. Texture is not a cosmetic detail — because a crystal's properties depend on direction, a textured film behaves like a partial single crystal, stiffer or more conductive or more magnetically aligned along one direction than another. The grain structure, the columns, and the texture together are the film's microstructure, and they are the reason two films of the very same composition can perform completely differently.

You do not have to guess at any of this — a diffractometer reads it straight off. To measure texture, you record a pole figure: pick one reflection, then tilt and rotate the film through every orientation and map where that plane's diffracted intensity concentrates. Random grains give a uniform ring; a fibre texture gives a bright spot or arc at the pole. And the same instrument reads strain: because a coherent film and its substrate have slightly different spacings, they give slightly split Bragg peaks, and a high-resolution scan separates the film peak from the substrate's to a few parts in 10^4. A reciprocal-space map goes further — a fully strained film sits at exactly the substrate's in-plane reciprocal coordinate, while a relaxed film has drifted toward its own. The reciprocal lattice you built two rungs ago is doing all the measuring.

The superlattice: a crystal you design

Now the payoff. If you can grow one epitaxial layer with atomic precision, you can grow a stack of them, alternating two materials over and over: a few nanometres of GaAs, then a few of AlAs, then GaAs again, hundreds of times. The result is a superlattice — an artificial, periodic structure whose repeat unit is not a handful of atoms but a whole engineered layer pair. If GaAs is 6 nm and AlAs is 4 nm, the superlattice has a brand-new period of D = 10 nm stacked along the growth direction. You have, quite literally, designed a crystal: a periodicity that no natural material possesses, built to order layer by layer. Its electrons see a periodic staircase of quantum wells and barriers, and the tailored miniband structure that emerges is the whole point of building it.

Diffraction sees the man-made period as plainly as it sees the atomic one. A period of D = 10 nm is a large distance in real space, so by the inverse-size relationship it prints a small spacing in reciprocal space: a comb of closely spaced satellite peaks flanking each ordinary Bragg reflection, separated by 1/D. Count the satellites' spacing and you read the layer period straight off; it is a routine, exquisitely sensitive check that the growth actually made the stack you designed. And the same misfit story from before still applies to every interface in the stack, which sets a hard ceiling on how thick each mismatched layer can be before it starts throwing misfit dislocations. Push the mismatch far enough — that 7.2 percent InAs-on-GaAs case — and instead of a flat strained film you get Stranski-Krastanov islanding: the strain spontaneously bunches the film into a regular array of tiny defect-free islands. Those are self-assembled quantum dots, a beautiful example of self-assembly closing the loop back to the dots and wells of the earlier guides.

Beware a genuine terminology clash. The word 'superlattice' carries two quite different meanings in materials structure. The engineered layer stack of this guide is one — grown, artificial, with a period of nanometres. But in the order-disorder rung 'superlattice' means something else entirely: when a random alloy like Cu-Au chemically orders itself, the atoms settle onto distinct sublattices and the true repeat of the crystal grows larger, producing extra 'superlattice reflections' in diffraction. That superlattice is thermodynamic and atomic-scale, not grown and nanoscale. Same word, two worlds — the glossary keeps them apart, and so should you. Note too that a superlattice is not the same as a single quantum well: one well is a lone thin film, whereas a superlattice is a periodic run of many wells and barriers coupled together.