What Is Thin-Film Deposition? CVD, PVD and ALD Explained
Thin-film deposition is the step that puts very thin films on the surface of a silicon wafer. A chip is built from many stacked layers: films that block electricity, films that carry it, and semiconductor films. There are several ways to lay them down. CVD uses reacting gases, PVD knocks atoms loose from a solid, and ALD builds a film up in tiny steps. Engineers choose a method by the film's material and shape.
What you'll learn
- Why a chip is a stack of thin films, and how thin those films are
- The difference between "growing" a film by thermal oxidation and "stacking" one by deposition
- How CVD, PVD (sputtering) and ALD work, and when each is used
- Why it is hard to coat the inside of deep trenches and holes evenly
- Plating for copper wires, how deposition fits with removing and polishing, and who makes the tools
Prerequisites: it helps to know that chips are made by repeating three jobs: add a film, copy a pattern onto it with light, and remove what is not needed. See "How Are Chips Made?" for the details.
In everyday life
The chip in your phone is not one flat slab. On top of tiny switches called transistors sit many layers of wiring. Films between those layers keep the wires apart. Making each of those films, one by one, is the job of deposition.
In the industry, the step is called deposition, often shortened to "dep." You will also see the phrase "deposition equipment" in news about chipmaking tools.
Why build a chip from thin films?
The films in a chip fall into three broad groups by job. The first group does not conduct electricity. These films separate wires from each other and parts of a transistor from each other. The second group is metal films that conduct well and become the wiring. The third group is semiconductor films, which form parts of transistors and other devices.
A chip is made by covering the whole wafer with one of these films, then cutting it into a pattern. This is repeated over and over. Each layer has its own material and shape, so deposition is repeated once for every layer.
Some of these films are astonishingly thin. A transistor has an electrode that receives the voltage and a path that carries the current. A very thin insulating film keeps the two apart (see "What Is a Transistor?"). In 2004, Intel announced that this film was 1.2 nm thick in what it calls its "65 nm generation" of manufacturing technology.
A nanometer (nm) is one millionth of a millimeter. According to Intel's 2007 description, 1.2 nm is about five atomic layers. That is about 1/70,000 of a human hair, which is about 0.09 mm thick. Of course, film thickness varies widely with the film's job.
Two ways to make a film: "grow" it or "stack" it
There are two broad ways to make a film. One is to change the silicon of the wafer itself into a film. The other is to bring material in from outside and let it settle on the surface. The second way is called deposition.
The classic example of the first way is thermal oxidation. Silicon is heated to a high temperature, around 900–1,200 °C, in oxygen or water vapor. The silicon at the surface combines with oxygen and turns into a film of silicon dioxide (SiO₂). This is the same substance as the main ingredient of glass, and it does not conduct electricity.
A thermal oxide film grows by using up the wafer's own silicon. That is why part of the film sits below the original surface. According to lecture notes from the University of California, Berkeley, about 44% of the film's thickness comes from silicon that was converted.
Thermal oxide films have few gaps, and their boundary with the silicon is very clean. But they can only form where there is silicon to use as raw material. This method cannot put a film on top of metal wiring, for example. The high heat also spreads impurities that were added earlier (see "Dopant Diffusion in Silicon").
So most films are made by deposition. The next sections look at three main deposition methods. The difference between using oxygen and using water vapor is covered in "Dry vs Wet Oxidation."
CVD: gases react on a hot surface
In CVD, source gases react chemically above the wafer. The solid they produce settles on the wafer as a film. CVD stands for chemical vapor deposition. For example, silane is a gas made of silicon and hydrogen. When it is heated, it breaks down and leaves a silicon film behind.
The reaction happens mostly on the hot surface of the wafer. The gas moves all around the wafer and gets into dips in the surface. So even a surface with steps can be covered with a fairly even thickness. Most insulating films are made by CVD.
Some films need high temperatures if heat alone drives the reaction. Turning the gas into a plasma lets the film form at a lower temperature. This is explained in "Plasma in Semiconductor Manufacturing."
PVD: knock atoms loose and let them land
PVD carries atoms of the material straight to the wafer, without a chemical reaction. PVD stands for physical vapor deposition. The most common type is sputtering.
First, the air is pumped out of a chamber. Inside sits a plate of the material you want as a film. This plate is called the target. Argon gas is turned into a plasma, and positively charged argon particles slam into the target. The impacts knock atoms out of the target. They land on the wafer facing it and build up a film.
Sputtering is well suited to metal films. Swap the target, and you can handle many metals, such as aluminum and titanium. Its weakness is that atoms fly from the target in almost straight lines. So films have trouble reaching the walls deep inside a hole. For how it differs from heating a material until it evaporates, see "Sputtering vs Evaporation."
ALD: very thin layers, one cycle at a time
In ALD, two source gases take turns flowing over the wafer to build a film. ALD stands for atomic layer deposition. It is a relative of CVD, but the two gases never flow at the same time.
First, the first gas flows in, and its molecules stick to the surface. Once the sticking spots are filled, no more molecules can stick. The reaction stops by itself. This is called "self-limiting," and it is the key feature of ALD. Leftover gas is pumped out before the next gas comes in.
Next, the second gas flows in. It reacts with the first gas on the surface and forms a very thin film. This is one cycle. Each repeat makes the film a little thicker. One example is a tool at Binghamton University in the US. There, aluminum oxide grows 0.085 nm per cycle.
The name says "atomic layer," but one cycle usually adds less than one layer of atoms. The molecules crowd each other out for space, which leaves gaps. The thickness per cycle also changes with the gases and the temperature.
Because ALD is self-limiting, sending in extra gas barely changes the thickness added per cycle. As long as the gas reaches it, even the bottom of a deep hole gets the same thickness as a flat surface. Another strength is that the number of cycles sets the thickness. The trade-off is time. At the rate above, a 10 nm film needs more than 100 cycles.
So ALD is used for films that are especially thin and need precision. A key example is that thin insulating film between a transistor's electrode and its current path. In 2007, Intel switched this film from a silicon-dioxide-based material to one containing hafnium (high-k / metal gate). ALD is used to lay down such ultra-thin films with an even thickness.
The principle behind ALD was found independently in the Soviet Union in the 1960s and in Finland in the 1970s. In 1974, Finland's Tuomo Suntola and a co-inventor filed a patent, aiming to make thin, flat displays. In 2018, Suntola received the Millennium Technology Prize for this work.
Coating deep trenches and holes evenly
The surface of a chip is not flat. It has countless trenches for wiring and holes that connect upper and lower layers. How many times deeper a hole is than it is wide is called its aspect ratio. The narrower and deeper the hole, the higher the aspect ratio.
When you coat a trench, how thick is the film on its sides and bottom, compared with the flat surface above? This measure is called step coverage. The closer it is to 100%, the more evenly the film covers everywhere.
In PVD, atoms fly in almost straight lines. So the film builds up thick around the mouth of the trench, and the deep walls stay thin. The rim can even grow outward and close off the opening first. Then a hollow space is left inside. It can make a wire thinner or break it. This hollow space is called a void.
The deeper the hole, the harder it is to coat evenly. One example is 3D NAND, used for phone storage. It stacks the memory parts in many vertical layers, and films must coat the walls of deep holes. In places like this, methods like ALD, which reach deep inside with the same thickness, become important.
Films that copy the crystal: epitaxial growth
Most CVD films do not have their atoms lined up in order. But under the right conditions, a film can copy the atomic order of the silicon below. It then grows as a neatly ordered crystal. This method is called epitaxial growth.
It is used, for example, to make a crystal layer with different properties on the wafer surface. Even in the most advanced transistors, parts such as the regions where current enters and leaves are made this way. For how it differs from ordinary CVD, see "Epitaxy vs CVD."
Copper wires are filled in by plating
For a long time, the metal for chip wiring was aluminum. In 1997, IBM announced a chip with copper wiring. IBM says copper has about 40% lower electrical resistance than aluminum. Lower resistance cuts signal delay and the heat made in the wires (see "What Is a Resistor?").
Copper is hard to shape by depositing a film and then cutting it away. So trenches in the shape of the wires are dug into an insulating film first, and then filled with copper. The filling uses electroplating. The wafer is dipped in a liquid with copper dissolved in it, and a current is passed. Copper then builds up on the surface little by little.
Before plating, a thin metal film is added to stop copper from seeping into the films around it. Next comes a thin copper film that carries the plating current. Both are usually laid down by PVD, such as sputtering. So even a single wire combines several deposition methods.
Plating adds enough copper to overflow the trenches. The extra copper is removed by polishing the surface flat. This way of making wires is called the damascene process: dig trenches, fill them, then polish so only the wires remain. You can compare it with shaping wires by cutting in "Damascene vs Subtractive Etch Interconnects."
Adding, removing and polishing work as a team
Deposition alone cannot form a circuit. After a film covers the whole wafer, a pattern is copied onto it with light. Then the unwanted parts are removed with gases or liquid chemicals, leaving the shape. This removal step is called etching.
Stacking layers leaves bumps and dips on the surface. If they remain, some spots will be out of focus when the next pattern is copied. So a film is laid down a little thick to fill the dips. Then the surface is polished flat with a pad and a liquid that contains fine abrasive particles. This polishing step is CMP. It is also the step that removes the extra copper in copper wiring.
The thickness of each film is measured along the way. One kind of tool shines light on the film and works out the thickness from how the reflected light changes (see "Ellipsometry vs Reflectometry").
Who makes the tools?
Deposition tools are specialized for each method and type of film. Below are the main makers, described only as each company describes itself (as of October 2026). For the industry as a whole, see "Semiconductor Manufacturing Equipment."
- Applied Materials (US): lists ALD, CVD and PVD tools in its product lineup.
- Lam Research (US): its annual report for fiscal 2025 describes tools such as plating systems that fill copper wiring.
- Tokyo Electron (Japan): lists deposition tools in its product lineup.
- ASM International (Netherlands): said in a September 2025 investor presentation that it kept its leading share in ALD.
- KOKUSAI ELECTRIC (Japan): presents tools such as batch deposition systems, which process many wafers at once.
Common misconceptions
Myth: Films are put on by brushing on a liquid, like paint
In fact, most films are made from gases, or from atoms flying through a chamber with the air pumped out. Liquids are used only in some steps. Examples are copper plating, and spinning the wafer to spread a chemical thinly. The myth arises because in daily life, a "film" or "coating" brings paint to mind.
Myth: ALD is the most even, so every film should be made by ALD
In fact, ALD adds only about 0.1 nm per cycle, so thick films would take far too long. Engineers pick the method that suits each film. Thick insulating films use CVD, and most metal films use PVD or plating. The myth arises because it is easy to assume the most precise method suits everything.
Myth: Every oxide film is added on top
In fact, a thermal oxide film also grows below the original surface. Even when both are silicon dioxide, a film stacked on top by CVD forms differently and behaves differently. The myth arises because both are described as "putting on a film," so they look like the same process.
By the numbers
| Number | What it measures |
|---|---|
| 1.2 nm | Thickness of the insulating film under the electrode in transistors of Intel's "65 nm generation" (announced in 2004). About five atomic layers |
| About 44% | Share of a thermal oxide film's thickness that comes from converted silicon (UC Berkeley lecture notes) |
| 0.085 nm | One example of the aluminum oxide thickness added per ALD cycle (a Binghamton University tool; varies with conditions) |
| About 40% | How much lower copper's electrical resistance is than aluminum's (according to IBM, which announced a copper-wired chip in 1997) |
Lay a film only a few atoms thick over the whole of a 300 mm wafer, with the same thickness everywhere. Do it even at the bottom of narrow, deep holes. That is what makes deposition hard.
Learn more
- Explainers on the process: "How Are Chips Made?", "Plasma in Semiconductor Manufacturing", "Semiconductor Manufacturing Equipment"
- Side-by-side tables: "CVD vs PVD", "ALD vs CVD", "LPCVD vs PECVD", "Copper vs Cobalt vs Ruthenium Interconnects"
- Try the numbers: with the thermal oxide thickness calculator, you can change the temperature and time and see how the oxide thickness changes.