PVD

Physical vapor deposition is how a fab lays down most of its metal. A solid source material is physically knocked or boiled into a vapor inside a vacuum chamber, and that vapor condenses onto the wafer as a thin film. There is no chemical reaction building the film from gas precursors the way there is in CVD; the atoms that land on the wafer are the same atoms that left the source. That physical, line-of-sight nature is the whole story of what PVD is good at and where it struggles.

Sputtering is the dominant form of PVD in modern logic and memory fabs. A target of the material you want to deposit is held at negative potential, argon is bled into the chamber, and a plasma forms. Positive argon ions accelerate into the target and eject target atoms by pure momentum transfer, like a break shot on a pool table. Those ejected atoms travel across the chamber and stick to the wafer. Because the ejection is mechanical rather than thermal, sputtering handles high-melting-point metals and alloys that evaporation cannot, and it preserves alloy composition faithfully.

The magnetron is what makes sputtering fast enough to be practical. A ring of magnets behind the target traps secondary electrons in a racetrack close to the target surface, so they ionize far more argon per electron before escaping. That dense local plasma raises the sputter rate by an order of magnitude at lower pressure, which also means fewer gas collisions and a more directional flux arriving at the wafer. Nearly every metal-deposition sputter tool in production is a magnetron tool.

Sputter yield sets the deposition rate, and rate sets throughput.

where $R$ is the film growth rate, $Y(E)$ the sputter yield in ejected atoms per incident ion at ion energy $E$, $J_{ion}$ the ion current density at the target, $n_{film}$ the atomic density of the deposited film, and $\eta_{transport}$ the fraction of ejected atoms that survive the trip to the wafer. The magnetron earns its place by raising $J_{ion}$ an order of magnitude; ionised PVD and long-throw geometry earn theirs by raising $\eta_{transport}$ toward one.

Reactive sputtering turns PVD into a way to grow compound barriers. Add nitrogen to the argon and sputter a titanium or tantalum target, and the film that lands is TiN or TaN rather than the pure metal. These conductive nitrides are the diffusion barriers and liners that keep copper from poisoning silicon, and they are a core PVD workload alongside the aluminum, tungsten, and copper-seed depositions.

Step coverage is where the line-of-sight nature bites. Because sputtered atoms arrive along straight paths, a deep, narrow via sees plenty of arriving flux at its mouth and very little at its bottom and sidewalls. The result is an overhang at the top that can pinch off into a keyhole void before the feature fills. Fabs fight this with collimators, long-throw geometry, and ionized PVD, where the metal flux is itself ionized and steered straight down the feature by a substrate bias. Even so, PVD is a poor choice for filling high-aspect-ratio structures, which is why conformal ALD and CVD took over barrier and fill roles as features shrank, leaving PVD to seed layers, contacts, and blanket films.

Read PVD through a line-of-sight-and-momentum lens rather than a generic thin-film lens. The moment you picture atoms flying in straight lines from a target, everything else follows: why it deposits high-melting metals and alloys faithfully, why reactive sputtering gives you the copper barriers, and why the same straight-line flux that makes it simple also makes it the wrong tool for filling a deep via.