Semiconductors
The Tunable Middle Ground
A Material You Can Argue With
Most materials are firmly one thing or the other. Copper is a conductor – electrons flow through it freely, and there is essentially nothing you can do at room temperature to stop them. Glass is an insulator – electrons cannot move through it, and there is essentially nothing you can do at room temperature to start them. Silicon, the canonical semiconductor, is in between, and that "in between" is not a fence-sitting compromise; it is the property that makes silicon the most useful element on Earth. You can choose how conductive a piece of silicon is, where it conducts, when it conducts, and what triggers the switch. A wire cannot do that. An insulator cannot do that. Semiconductors are the only materials whose conductivity is something you can argue with.
Every digital device in your life depends on this property. The phone in your pocket has roughly 15 billion silicon transistors, each of them a tiny region of semiconductor that has been tuned to switch from conducting to non-conducting and back, billions of times per second, in response to a voltage on a third terminal. Multiply by the number of phones, computers, cars, satellites, factories, and tools, and the modern world is fundamentally a fabric of carefully argued silicon.
Bands and the Gap Between Them
In a single isolated atom, electrons live on discrete energy levels. Bring atoms together into a crystal and those levels broaden into bands – broad ranges of allowed energies, with forbidden gaps between them. The lowest occupied band is the valence band, packed with electrons that hold the crystal together. Above it is the conduction band, where electrons can wander freely if they have enough energy to get there. Whether a material conducts depends entirely on whether the conduction band is reachable.
In a metal, the bands overlap or the upper band is partly filled, so electrons can move with no extra energy needed. In an insulator like diamond, the gap between valence and conduction bands is around five electronvolts – tens of times more than thermal energy can supply – so essentially no electrons make it across at room temperature, and no current flows. A semiconductor is defined by a gap small enough that some electrons do make it across through random thermal kicks, but few enough that conductivity stays low. Silicon’s gap is about 1.1 electronvolts; germanium’s is 0.7; gallium arsenide is 1.4. Different gaps, different behavior, different uses.
Doping: Engineered Impurities
Pure silicon at room temperature has so few free carriers that it is nearly an insulator. The trick that turned silicon from curiosity into civilization is doping: deliberately replacing one in a million silicon atoms with an atom from a neighboring column of the periodic table. Phosphorus has one more outer electron than silicon, so a phosphorus dopant donates a lone extra electron that floats free in the conduction band. The result is n-type silicon, where current is carried by negative electrons. Boron has one fewer outer electron, so it accepts an electron from a neighboring silicon bond, leaving a missing-electron site that propagates through the lattice as if it were a positively charged particle. The result is p-type silicon, where current is carried by positive "holes".
Holes are not actually particles – they are the absence of an electron in an otherwise filled band – but they behave so much like positive particles that semiconductor engineers compute with them on equal footing with electrons. Doping is precise to the parts-per-billion level, set during crystal growth or implanted by accelerating dopant ions into a finished wafer. Wherever you place dopants and at what concentration determines exactly where in the chip current can flow and how easily.
The PN Junction
Put a piece of n-type silicon next to a piece of p-type silicon and something interesting happens at the boundary. Free electrons from the n side diffuse into the p side and recombine with holes; holes from the p side diffuse the other way and recombine with electrons. After a brief moment the diffused carriers leave behind a thin region with no free carriers at all – the depletion region – flanked by exposed dopant ions that build up an electric field. This field stops further diffusion and leaves the junction in equilibrium.
Now apply a voltage. Push electrons toward the n side and holes toward the p side and the depletion region narrows; current flows easily. Push the other way and the depletion region widens, blocking current almost entirely. A pn junction is therefore a one-way valve for current – a diode – and the simplest non-trivial semiconductor device. The same junction is also a solar cell. A photon absorbed near the boundary knocks an electron across the gap, and the built-in field sweeps the electron one way and the hole the other before they can find each other again. That forced separation is the current. Drive the junction the other way, injecting electrons to recombine with holes, and it can emit photons instead. Diode, solar cell, and LED are three uses of one physical structure.
Silicon, though, makes a terrible LED, and the reason is worth knowing. Its band gap is indirect. An electron sitting at the bottom of the conduction band and a hole at the top of the valence band differ not just in energy but in momentum. A photon carries almost no momentum, so it cannot bridge that difference alone. A lattice vibration has to turn up at the same instant to balance the books, and that coincidence is rare enough that the energy usually dribbles away as heat instead. Gallium arsenide and gallium nitride have direct gaps, where the two states sit at the same momentum and recombination produces a photon immediately. Every LED you own is made of something other than silicon, and this is the reason.
The Transistor
A diode is two doped regions; a transistor is three. The dominant transistor today is the metal-oxide-semiconductor field-effect transistor, or MOSFET. It has two heavily doped regions of the same type (the source and drain), separated by a region of the opposite type (the body). On top of the body sits a thin layer of insulating oxide, and on top of the oxide sits a metal gate electrode. Apply a voltage to the gate and the field reaches through the oxide into the body, attracting carriers to the surface and creating a thin conductive channel between source and drain. Current flows. Remove the gate voltage and the channel disappears. Current stops.
That is it. Three terminals: a control input, a current path, and a switch. With one transistor you can amplify a signal. With two you can build a NAND gate. With six you can build the cell that holds one bit of cache memory. With a few billion you can build a CPU. John Bardeen and Walter Brattain got the first working transistor going at Bell Labs in December 1947, a point-contact device held together with a slab of germanium and a wedge of gold foil. William Shockley, their group leader, was not in the room; he spent the following weeks working out the junction transistor, the design that turned out to be manufacturable. The three shared the Nobel Prize in 1956, which is why the names are usually said in one breath. Everything since has been about making them smaller, faster, and cheaper.
From Sand to Chip
The starting material is silica sand – the same stuff as a beach. It is reduced to metallurgical-grade silicon, then purified to a level of one impurity atom per billion silicon atoms. From this ultra-pure feedstock, a single seed crystal is dipped into molten silicon and slowly pulled out while rotating, growing a monocrystalline ingot a meter long. The ingot is sliced into circular wafers a fraction of a millimeter thick, polished to optical flatness, and shipped to the fab.
At the fab, the wafer is processed through hundreds of steps over weeks. Each step adds, removes, or modifies one layer: oxide growth, photoresist coating, photolithographic exposure through a patterned mask, etching, dopant implantation, metal deposition. The mask patterns, ultimately written by extreme-ultraviolet light at a wavelength of 13.5 nanometers, place features a few tens of nanometers apart – far below the wavelength of visible light, and finer than most viruses are wide. The whole process happens in cleanrooms with fewer than a hundred dust particles per cubic meter, because a single speck on the wrong layer ruins a billion-transistor die.
Out the other end come finished wafers covered in identical chip dies. Each die is tested, and the working ones are sliced apart, packaged, and shipped. The cost per transistor has dropped by a factor of about a hundred million since the 1960s. There is no other product in the history of human manufacturing whose cost per unit has fallen by that much.
Quantum Effects in Modern Chips
Modern transistor channels are only tens of atoms long. The node names do not tell you this, and are widely misread as if they did. A process sold as "3 nm" contains nothing that measures three nanometers: the spacing between neighboring gates is closer to forty-five, and the finest metal lines run about twenty-four apart. The number stopped being a physical dimension long ago and became a generation label. What genuinely improves each time is how many transistors fit in a square millimeter.
Even at those honest dimensions, classical descriptions break down and quantum mechanics becomes a dominant engineering concern rather than a curiosity. Electrons can quantum tunnel directly through the insulating gate oxide, leaking current and wasting power. They can tunnel through the channel between source and drain even when the transistor is supposed to be off, raising static power. Leakage is also why clock speeds stopped climbing in the mid-2000s. Past a few gigahertz, the power a chip burns doing nothing swamps the power it spends computing, so manufacturers began adding cores instead of megahertz.
Silicon dioxide, the traditional gate insulator, was thinned to a few atoms to keep the gate in control, and there it ran out of room. The way past was to stop using it. Since the late 2000s the insulator has been hafnium oxide, which holds off the same field in a physically thicker layer, because what the gate actually needs is capacitance rather than thinness.
The other response has been to redesign the geometry. Rather than a flat channel under a flat gate, modern transistors are three-dimensional fins (FinFETs) or nanowires wrapped on all sides by gate material (gate-all-around or GAA), giving the gate more leverage over the channel and pushing tunneling back. Each generation has bought another factor of two in density at the cost of dramatically more complex fabrication. The whole industry runs on the assumption that something can always be invented to push past the next quantum-mechanical wall, and so far it has worked.
Beyond Silicon
Silicon is convenient, abundant, and well-understood, but it is not always the best semiconductor for a given job. Gallium arsenide and indium phosphide have higher electron mobility than silicon and dominate microwave electronics and high-end optical communications. Gallium nitride and silicon carbide have wider band gaps that let them handle high voltages and high temperatures; they are quietly transforming power electronics in electric vehicles and renewable energy infrastructure. Indium gallium nitride is what makes blue and white LEDs possible, and won its inventors the 2014 Nobel Prize in Physics.
Newer materials are being investigated for the post-silicon era. Two-dimensional crystals like graphene, molybdenum disulfide, and tungsten diselenide are atomically thin sheets with unusual electronic properties. Carbon nanotubes can carry enormous current densities and have been demonstrated as transistor channels. Topological insulators conduct only on their surface, with quantum-protected current flow that may eventually enable low-power electronics. None of these has yet displaced silicon at scale, because silicon’s manufacturing infrastructure is one of the most refined ever built and any replacement has to compete with sixty years of accumulated process engineering. But the underlying physics is general: any material with a tunable band gap is a candidate, and silicon’s lead is not eternal.
From Band Gaps to a Civilization
Semiconductors are the cleanest example of a recurring pattern in physics: a small, abstract piece of quantum theory – in this case, the fact that electron energies in a periodic crystal form bands separated by gaps – turns out to underwrite an entire civilization’s worth of technology. The band-gap concept was understood in the 1930s. The first transistor came in 1947. Within decades, the cumulative impact of that single piece of physics had reorganized industry, communications, science, warfare, and daily life more thoroughly than any prior technology. The chain from "electrons in a crystal lattice form bands" to "you can carry a supercomputer in your pocket" is one of the most consequential cause-and-effect arcs in human history, and almost every link in the chain is straightforward physics.
