Semiconductor Fabrication: From Sand to Chip · Module 5
Oxidation and Dielectrics
Insulating layers isolate, protect, and control the devices on a chip. This module covers how silicon dioxide is grown and why newer insulators have joined it.
What you will learn
- How a thermal oxide grows on silicon
- How dry and wet oxidation differ
- The jobs that oxide layers do in a chip
- Why the gate insulator changed from silicon dioxide to high-k materials
In plain language
Heat a silicon wafer in oxygen and its surface turns into a thin layer of glass, silicon dioxide. This glass insulates very well, bonds tightly to the silicon, and can be grown to a chosen thickness with great precision. More than any other property, this oxide made silicon chips practical.
Oxide layers separate transistors from each other, protect the silicon surface, block dopants from places where they are not wanted, and form the insulator under the gate of a transistor. As transistors shrank, the gate oxide became so thin, about five atomic layers, that electrons began to leak straight through it. In 2007 the industry started to replace it with new insulators based on hafnium.
Core lesson
Growing an oxide
Wafers are loaded into a quartz tube furnace and heated to between 800 and 1,100 °C in a flow of gas. In dry oxidation the gas is oxygen, and silicon reacts with it to form silicon dioxide. In wet oxidation the gas is steam, and the reaction also releases hydrogen.
The reaction happens at the boundary between the silicon and the oxide. Oxygen or water must first travel through the oxide that is already there. The boundary therefore moves down into the wafer as the oxide grows, and it is always formed from fresh silicon that has never met the room air. This is why the boundary is so clean. Growth uses up silicon. For every 100 nm of oxide, about 45 nm of silicon is consumed.
Dry and wet oxidation
Dry oxidation is slow and gives the densest oxide with the best electrical quality. It is used for thin, critical layers. Wet oxidation is much faster, because water dissolves in the oxide far more readily than oxygen does, and it is used for thick layers. At 1,000 °C, one hour of wet oxidation grows roughly 430 nm of oxide, while one hour of dry oxidation grows roughly 70 nm.
ANNLab, University of Arizona
Figure 1. Oxide thickness against time at 1,000 °C, calculated with the Deal-Grove model and published rate constants. Growth slows as the oxide thickens.
Both curves bend over with time. While the oxide is thin, growth is limited by the reaction at the boundary, and thickness rises in step with time. Once the oxide is thick, growth is limited by how fast the oxidant can cross it, and thickness rises only with the square root of time.
What oxides do
- The gate insulator separates the gate from the channel of a transistor.
- Isolation oxide fills trenches etched between transistors so that neighbors do not affect each other.
- A masking oxide blocks dopants during diffusion or implantation.
- A passivation oxide protects the silicon surface and ties up broken bonds.
- Deposited oxides insulate one layer of metal wiring from the next, as Modules 8 and 9 describe.
Oxide quality
Not every silicon bond at the boundary is satisfied. The leftover broken bonds, called interface traps, can catch and release carriers and disturb a transistor. A final anneal at about 400 °C in a gas that contains hydrogen ties up most of them. Sodium ions are another danger because they drift through oxide under an electric field. Engineers check oxide quality by measuring how the capacitance of a small test capacitor changes with voltage.
High-k dielectrics
A transistor needs a large gate capacitance so that the gate controls the channel firmly. For decades the way to get it was a thinner oxide. Below about 2 nm, however, electrons pass through the oxide by quantum tunneling, and the leakage current rises steeply with each further step.
The answer was a material with a higher dielectric constant, written k. Silicon dioxide has k = 3.9, and hafnium dioxide has k of about 20. A hafnium dioxide layer can be several times thicker than a silicon dioxide layer and still give the same capacitance, and the thicker layer leaks far less. These films are not grown from the wafer. They are deposited one atomic layer at a time, as Module 8 explains. They were introduced together with metal gates, which replaced the polysilicon used until then.
Going deeper
These sections add the equations behind the core lesson, each with a worked example. They assume first-year algebra and can be skipped without losing the thread of the course.
Oxide thickness x after time t follows a simple equation.
x2 + A x = B (t + τ)
B is called the parabolic rate constant and B / A the linear rate constant. Both depend on temperature and on whether the gas is oxygen or steam. The time τ accounts for any oxide present at the start.
Worked example
For wet oxidation at 1,000 °C, A = 0.226 µm and B = 0.287 µm2 per hour, with τ = 0. After 2 hours, x2 + 0.226 x = 0.574. Solving the quadratic gives x = 0.65 µm, or 650 nm.
To compare gate insulators, engineers quote the thickness of silicon dioxide that would give the same capacitance. This is the equivalent oxide thickness, or EOT.
EOT = t × (3.9 / k)
Worked example
A hafnium dioxide film is 3 nm thick with k = 20. Its EOT is 3 × 3.9 / 20 = 0.6 nm. A real silicon dioxide film of 0.6 nm would be about two atomic layers thick and would leak far too much to use.
Silicon has 5.0 × 1022 atoms per cm3, and silicon dioxide has about 2.2 × 1022 molecules per cm3. Each molecule of oxide needs one silicon atom, so the thickness of silicon consumed is 2.2 / 5.0 = 0.44 times the oxide thickness.
Worked example
Growing 100 nm of oxide consumes 44 nm of silicon. The top of the oxide ends up 56 nm above the original silicon surface, and the boundary lies 44 nm below it.
Check your understanding
Try each question before opening the answer. The last question uses the Going deeper material.
Wet oxidation is much faster. Dry oxidation gives a denser oxide with better electrical quality, so it is used for thin, critical layers.
At the boundary between the silicon and the oxide. The oxidant must travel through the existing oxide to reach it.
The oxidant has to cross more oxide to reach the silicon. For a thick oxide this transport limits the rate, and thickness rises only with the square root of time.
The silicon dioxide had become so thin that electrons tunneled through it. A high-k material gives the same capacitance with a physically thicker layer, which leaks far less.
EOT = 2 × 3.9 / 16 = 0.49 nm.
Key terms
- Thermal oxidation. Growing silicon dioxide by heating silicon in oxygen or steam
- Dry oxidation. Oxidation in oxygen, slow and of the highest quality
- Wet oxidation. Oxidation in steam, fast and used for thick layers
- Gate insulator. The thin insulating layer between the gate and the channel
- Interface trap. A broken bond at the boundary between silicon and oxide that can catch carriers
- Dielectric constant. A measure of how much an insulator increases capacitance, written k
- High-k dielectric. An insulator with a dielectric constant well above that of silicon dioxide
- Equivalent oxide thickness. The thickness of silicon dioxide that would give the same capacitance
- Tunneling. The quantum effect that lets electrons pass through a very thin insulator
In the fab
Furnace engineers and thin-film metrology technicians work with these layers every day. Oxide thickness is usually measured with an ellipsometer, which finds the thickness from the way the film changes reflected light.
Further reading
J. D. Plummer and P. B. Griffin, Integrated Circuit Fabrication: Science and Technology, Cambridge University Press. The authors offer a prepublication version that is free to read on the Stanford book site (opens in a new tab).
Our Cleanroom Videos on the Training page include a short video on ellipsometry.
Previous: Module 4Course homeNext: Module 6, Doping by Diffusion and Ion Implantation
Created by Prof. Zafer Mutlu, ANNLab, University of Arizona.