Semiconductor Fabrication: From Sand to Chip · Module 11
Beyond Silicon
Silicon will remain the base of the industry, but new materials are joining it. This module looks at what comes next, including the materials we study in our own lab.
What you will learn
- Why very thin silicon channels reach a limit
- Which other semiconductors are already in production and why
- What two-dimensional materials, carbon nanotubes, and graphene nanoribbons offer
- What a new material must achieve before a fab will use it
In plain language
For sixty years, the main way to improve chips was to shrink the silicon transistor. That road is getting harder. To keep control of a very short channel, the channel must also be very thin, and silicon only a few nanometers thick conducts poorly because electrons scatter from its uneven surfaces.
Researchers and companies are therefore studying materials that are thin by nature. Some are sheets one atom or three atoms thick, called two-dimensional materials. Others are tubes or ribbons of carbon about a nanometer wide.
Other semiconductors have already taken over jobs that silicon does poorly. Gallium nitride and silicon carbide handle high voltages in chargers and electric vehicles. Compound semiconductors make the lasers, the LEDs, and the radio amplifiers in phones. None of these replaces silicon. They are used beside it or added on top of it.
ANNLab, University of Arizona
Figure 1. The gate has wrapped further around the channel with each generation. An atomically thin channel is the limit of this trend.
Core lesson
Why the silicon channel reaches a limit
A gate can switch a transistor off cleanly only if the channel is thin compared with the gate length. Each new generation therefore needs a thinner channel. Below about 3 nm, a silicon film varies in thickness by a noticeable fraction from place to place, since a single layer of atoms is already 0.3 nm. These variations scatter the carriers and lower their mobility sharply. A material whose thickness is fixed by its own atomic structure does not have this problem.
Semiconductors already used beside silicon
Silicon mixed with germanium is part of every advanced silicon process. Compounds of elements from groups III and V of the periodic table, such as gallium arsenide and indium phosphide, emit light efficiently and work at very high frequencies. They are used in lasers, fiber-optic links, and radio circuits.
Silicon carbide and gallium nitride have band gaps about three times larger than that of silicon. They withstand much higher electric fields and temperatures, which makes power converters smaller and more efficient. Gallium nitride is also the material of the white LED.
Two-dimensional materials
Graphene is a single layer of carbon atoms in a honeycomb pattern. Its carriers move very fast, but it has no band gap, so a graphene transistor cannot be switched off and cannot serve as a logic switch.
Other layered crystals do have a band gap. Molybdenum disulfide is the best known. One layer is three atoms thick, about 0.65 nm, and transistors made from it switch on and off well. The open problems are practical ones. The material must be grown as a uniform single crystal over a whole wafer, metal contacts to it must have low resistance, and gate insulators must be deposited on a surface that offers few sites for ALD to start. Industry roadmaps list two-dimensional channels as candidates for the generations after stacked silicon sheets.
Carbon nanotubes and graphene nanoribbons
A carbon nanotube is a sheet of graphene rolled into a cylinder about a nanometer across. Depending on the direction of the roll, a nanotube is either a semiconductor or a metal. Carriers move through it with little scattering. The difficulty is to obtain only the semiconducting kind and to place the tubes in dense, aligned rows.
A graphene nanoribbon is a strip of graphene only a few atoms wide. Confining the electrons to such a narrow strip opens a band gap, and the size of the gap depends on the exact width and on the shape of the edges. Ribbons can be built from the bottom up. Small designed molecules are placed on a metal surface and heated, and they link together into ribbons that are identical atom for atom.
In our lab we grow atomically precise graphene nanoribbons on chip-scale samples, transfer them onto other substrates, and build them into transistors and sensors. The open questions are the same as for other new channel materials. They include contacts, clean boundaries with insulators, and control over where the ribbons sit. Our Research page describes this work.
New ways to build
New materials also allow new arrangements. Because many of them can be deposited or transferred at low temperature, they can be added above finished silicon circuits without harming the wiring. This opens a route to stacking layers of devices on one chip. Companies are also developing transistors in which the n-channel and the p-channel device sit one above the other to save area.
From the lab to the fab
A fab will adopt a new material only when it meets a long list of conditions. It must be made uniformly on 300 mm wafers. It must fit within the temperature limits of the flow. It must not contaminate the tools, as Module 3 explained. It needs contacts and insulators that work, and it must reach high yield and long lifetime at an acceptable cost.
Meeting all of these takes a long time. High-k gate insulators and EUV lithography each needed well over a decade of research before they entered production. The students who enter this field now will help decide which of the materials in this module make that step.
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.
Armchair graphene nanoribbons are named by the number N of carbon atoms across their width. Neighboring rows of atoms are 0.123 nm apart, so the width is close to (N − 1) × 0.123 nm. The band gap falls as the ribbon gets wider, and it also depends on which of three families the number N belongs to.
Worked example
A nine-atom-wide ribbon has a width of 8 × 0.123 = 0.98 nm. A five-atom-wide ribbon has a width of 4 × 0.123 = 0.49 nm. Changing the width by a single row of atoms changes the band gap, which is why atomic precision matters.
A useful estimate of how well a single gate controls a channel is the natural length λ.
λ = √( (εch / εox) tch tox )
Here tch and tox are the thicknesses of the channel and the gate insulator, and εch and εox are their permittivities. The gate length must be several times λ for the transistor to switch off well.
Worked example
For a silicon channel 5 nm thick with an insulator equal to 1 nm of silicon dioxide, the permittivity ratio is 11.7 / 3.9 = 3, and λ = √(3 × 5 × 1) = 3.9 nm. For a channel 0.65 nm thick with the same ratio, λ = √(3 × 0.65 × 1) = 1.4 nm. The thinner channel allows a much shorter gate. Wrapping the gate around the channel shortens λ further.
A power device blocks voltage across a lightly doped layer. The thickness that layer needs is roughly 2 V / Ec, where Ec is the electric field at which the material breaks down. Ec is about 0.3 million volts per centimeter for silicon and about 2.5 million volts per centimeter for silicon carbide.
Worked example
To block 1,200 V, silicon needs a layer of 2 × 1,200 / (3 × 105) = 8 × 10−3 cm, or 80 µm. Silicon carbide needs 2 × 1,200 / (2.5 × 106) cm, or about 10 µm. The thinner layer can also be doped more heavily, so the device has far lower resistance when it is on.
Check your understanding
Try each question before opening the answer. The last question uses the Going deeper material.
Its thickness varies from place to place by an amount that is large compared with the film itself. These variations scatter the carriers and lower their mobility.
Graphene has no band gap, so a transistor made from it cannot be turned off.
Its exact width, counted in atoms, and the shape of its edges.
Any two of the following will do. The material must grow uniformly on full-size wafers, fit the temperature limits of the process, leave the tools uncontaminated, have working contacts and insulators, and reach high yield and long lifetime at an acceptable cost.
6 × 0.123 = 0.74 nm.
Key terms
- Mobility. A measure of how easily carriers move through a material in an electric field
- Compound semiconductor. A semiconductor made of two or more elements, such as gallium arsenide
- Wide-bandgap semiconductor. A semiconductor with a band gap well above that of silicon, such as silicon carbide or gallium nitride
- Two-dimensional material. A crystal that is stable as a single layer one or a few atoms thick
- Graphene. A single layer of carbon atoms in a honeycomb pattern
- Molybdenum disulfide. A layered semiconductor whose single layer is three atoms thick
- Carbon nanotube. A cylinder of carbon atoms about a nanometer across
- Graphene nanoribbon. A strip of graphene a few atoms wide, with a band gap set by its width and edges
- Bottom-up synthesis. Building a structure by joining designed molecules
- Heterogeneous integration. Combining different materials or chips into one system
In the fab
Materials scientists, device physicists, and research engineers work on these questions in companies, national laboratories, and universities. Taking part in research as an undergraduate is one of the best ways into this field.
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).
If you would like to work on these materials with us, see our Join page.
Previous: Module 10Course home
Created by Prof. Zafer Mutlu, ANNLab, University of Arizona.