Semiconductor Fabrication: From Sand to Chip · Module 2
From Sand to Wafer
Chips start as ordinary sand. This module follows silicon from quartz to a polished wafer that is ready for the fab.
What you will learn
- Why silicon became the standard material for chips
- How silicon is purified to electronic grade
- How a single crystal is grown and cut into wafers
- What the numbers in a wafer specification mean
In plain language
Silicon is the second most abundant element in the Earth’s crust, after oxygen. In nature it is bound to oxygen as silica, the main ingredient of quartz sand. To make chips, the silicon is separated from the oxygen and purified until fewer than one atom in a billion is foreign.
The pure silicon is then grown into one single crystal, in which every atom sits in its place in a repeating pattern. The crystal is sliced into thin round wafers, and each wafer is polished to a mirror finish. A standard wafer today is 300 mm across and less than a millimeter thick. It will carry hundreds of chips.
Core lesson
Why silicon
Silicon is abundant and inexpensive, and its band gap of 1.12 eV lets devices work well above room temperature. Its most useful property is its oxide. Heating silicon in oxygen grows a layer of silicon dioxide that is stable, insulates very well, and forms a clean boundary with the silicon beneath it. Germanium was used for the first transistors, but its oxide dissolves in water, which makes reliable devices much harder to build.
From quartz to polysilicon
Quartz is heated with carbon in an electric arc furnace at about 1,900 °C. The carbon takes the oxygen and leaves liquid silicon that is about 98% pure. This metallurgical-grade silicon is good enough for aluminum alloys and far too impure for chips.
To purify it, the silicon is reacted with hydrogen chloride to form trichlorosilane, a liquid that boils at 32 °C. Repeated distillation removes the impurities. The purified vapor is then passed with hydrogen over thin silicon rods heated to about 1,100 °C, where it breaks down and deposits silicon on the rods. The product is polysilicon, made of many small crystals, with impurities at the level of parts per billion or lower.
ANNLab, University of Arizona
Figure 1. The main steps from quartz sand to a polished wafer.
Growing a single crystal
Most wafers come from the Czochralski method. Polysilicon is melted in a quartz crucible at 1,414 °C under argon gas. A small seed crystal with a known orientation is dipped into the melt and pulled up slowly while it rotates. Silicon freezes onto the seed and copies its atomic arrangement, so the whole ingot becomes one crystal. The pull rate and the temperature set the diameter.
A measured amount of boron or phosphorus is added to the melt to set the resistivity of the crystal. Some oxygen from the crucible also dissolves into the silicon, which makes the wafers mechanically stronger. A second method, float-zone growth, uses no crucible. It gives purer crystals and is used for power devices and detectors.
From ingot to wafer
The ingot is ground to an exact diameter, and a small notch is cut along its length to mark the crystal orientation. Wire saws slice it into wafers. The edges are rounded so they do not chip, and the surfaces are lapped and chemically etched to remove saw damage. Chemical-mechanical polishing then gives one side a surface that is smooth on the scale of atoms. After a final clean and inspection, the wafers are shipped in sealed carriers.
Most wafers for chips have a surface along the (100) crystal plane, because this plane gives the best boundary between silicon and its oxide.
Reading a wafer specification
A wafer is ordered by a short list of properties. A typical line reads “300 mm, p-type, boron, (100), 1 to 10 Ω·cm, 775 µm”. It gives the diameter, the carrier type, the dopant, the surface orientation, the resistivity, and the thickness. Flatness and the number of surface particles are specified as well.
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.
Silicon has the diamond cubic structure. Its repeating unit is a cube with an edge of 0.543 nm, called the lattice constant, and each cube holds eight atoms.
Worked example
The volume of one cube is (5.43 × 10−8 cm)3 = 1.60 × 10−22 cm3. Dividing eight atoms by this volume gives 5.0 × 1022 atoms per cm3, the number used in Module 1.
Electronic-grade polysilicon is often described as 99.9999999% pure, or “nine nines”. That means about one foreign atom in a billion.
Worked example
One part per billion of 5.0 × 1022 atoms per cm3 is 5 × 1013 impurity atoms per cm3. Wafers are usually doped on purpose to 1015 atoms per cm3 or more, so the unwanted background is at least twenty times lower than the intended doping.
A dopant atom prefers to stay in the liquid when silicon freezes. The segregation coefficient k is the ratio of the dopant concentration in the solid to that in the liquid at the freezing boundary. It is about 0.8 for boron and 0.35 for phosphorus. As the crystal grows, the remaining melt becomes richer in dopant, and so does the crystal that freezes from it.
Cs = k C0 (1 − f)k − 1
Here C0 is the starting concentration in the melt and f is the fraction of the melt that has solidified.
Worked example
A melt holds 1 × 1016 phosphorus atoms per cm3. The first crystal to freeze contains 0.35 × 1016 = 3.5 × 1015 per cm3. When half the melt has solidified, the crystal contains 3.5 × 1015 × (0.5)−0.65 = 5.5 × 1015 per cm3. The doping has risen by more than half, which is why wafers are sorted by measured resistivity. Boron, with k closer to 1, gives a more uniform ingot.
Check your understanding
Try each question before opening the answer. The last question uses the Going deeper material.
Silicon forms a stable, insulating oxide with a clean boundary to the crystal, and germanium does not. Silicon is also abundant and works at higher temperatures because of its larger band gap.
Metallurgical-grade silicon is about 98% pure. Polysilicon has been purified through trichlorosilane until impurities are at parts per billion or lower. Neither is a single crystal.
It provides the pattern. Silicon freezing from the melt copies the atomic arrangement and the orientation of the seed, so the whole ingot grows as one crystal.
The features built on the wafer are only nanometers in size, and some layers are a few atoms thick. Roughness or leftover saw damage would disturb those layers and blur the printed patterns.
Cs = k C0 = 0.8 × 2 × 1015 = 1.6 × 1015 atoms per cm3.
Key terms
- Silica. Silicon dioxide, the form in which silicon occurs in quartz sand
- Polysilicon. Silicon made of many small crystals, either as purified feedstock or as a deposited film
- Single crystal. A solid in which the atomic pattern continues unbroken through the whole piece
- Czochralski method. Growing a single crystal by pulling a seed slowly out of molten silicon
- Ingot. The cylinder of single-crystal silicon produced by crystal growth
- Wafer. A thin, polished slice of the ingot on which chips are built
- Crystal orientation. The crystal plane that lies along the wafer surface, such as (100)
- Segregation coefficient. The ratio of an impurity’s concentration in the solid to that in the liquid during freezing
- Chemical-mechanical polishing. Polishing that combines a chemical reaction with fine abrasion to give a flat, smooth surface
In the fab
Wafer makers employ crystal growers, polishing engineers, and metrology technicians. Inside a chip fab, incoming wafers are checked against their specification before any processing begins.
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).
Previous: Module 1Course homeNext: Module 3, The Cleanroom and Contamination Control
Created by Prof. Zafer Mutlu, ANNLab, University of Arizona.