Semiconductor Materials

Intuition

The empty seat in a full row

Imagine a cinema row with every seat taken. Nobody can move, because there is nowhere to move into. Now let one person leave. The person beside the gap can shift across, then the next, and the next — and although what is actually happening is people moving one way, what you see travelling along the row is the empty seat, going the other way.

A crystal of pure silicon at very low temperature is the full row. Every outer electron is committed to holding two atoms together, none of them is free to roam, and the material barely conducts at all. Warm it up and a few electrons collect enough energy to break loose. Each one that gets away is then free to carry current, and behind it sits an empty place that behaves, in every way that matters to a circuit, like a positive particle drifting the other way. Electronics gives that empty place a name: a hole.

The freed electron and the hole both carry current, and how many of either the material contains depends sharply on temperature. A metal is already unlike that: its free electrons are always present, and the count barely shifts with temperature.

Take that pure crystal and add an almost absurdly small quantity of another element — a few atoms in a million, or fewer — and you can decide, deliberately, whether it ends up with an excess of free electrons or an excess of holes, and how large that excess is. Conductivity stops being a fact of nature and becomes a design parameter, and that is the property the entire industry rests on. The process is called doping, and putting a doped-one-way region next to a doped-the-other-way region is how a diode and every transistor is made.

Silicon is the material of choice even though, by most measures, it is not the best semiconductor available. The reason for that lies outside its electrical properties altogether, and Layer 4 takes it up.

Practitioner

The band gap, and what it decides

What characterises a semiconductor above everything else is its band gap: the energy an electron needs to break out of its bond and become a carrier. Where the highest filled band and the lowest empty band overlap, the material is a metal; where the gap is very large, it is an insulator. Semiconductors sit in between, and in between means a gap of a few tens of times the energy that room-temperature heat makes available.

Silicon's gap is 1.12 eV, and the average thermal energy at room temperature is about 25.85 meV.

Carrier population follows that ratio exponentially rather than in proportion to it. A modest change in the gap, or in the temperature, moves the number of carriers by orders of magnitude, and most of what is characteristic about semiconductors traces back to that one sensitivity.

Against a background that thin, doping swamps the thermal carriers completely. An element with one more outer electron than silicon — phosphorus, arsenic — leaves a spare electron per dopant atom that needs almost no energy to become free, and the material is then n-type, with electrons as the majority carrier. An element with one fewer, boron being the usual choice, creates an easily filled vacancy and gives p-type material, holes in the majority. Dopant concentration sets the carrier count directly, so conductivity becomes a number the process engineer chooses.

Gallium arsenide, with a gap of 1.42 eV, moves electrons faster than silicon and emits light efficiently, which silicon cannot do; it turns up in RF parts and in optoelectronics. Germanium came first historically and has largely gone. Silicon carbide and gallium nitride have taken over the top end of power electronics.

Temperature is no small effect here either. Heat generates carriers, more carriers mean lower resistance, and the fall is steep — the opposite direction to a metal. A thermistor exploits that deliberately, and the same mechanism makes leakage current in every semiconductor device a strong function of temperature.

Engineer

Carriers, junctions, and the limits of the picture

Conduction in a doped semiconductor is carried by both species at once. The majority carriers are the ones doping provides; the minority carriers are the thermally generated ones of the opposite type, present in far smaller numbers. A device's useful behaviour comes almost entirely from the majority carriers. Its temperature-dependent misbehaviour — leakage, gain drift, thermal runaway — comes from the minority ones, whose population is set by that exponential. Germanium shows what that costs.

Doping on its own only sets a conductivity. What turns it into a device is a boundary: bring n-type and p-type material together and carriers diffuse across it, leaving behind a region stripped of mobile carriers and containing a built-in electric field. That depletion region conducts in one direction and blocks the other, which makes it a pn junction, the ancestor of every diode, transistor, solar cell and photodiode. A JFET squeezes a conducting channel with that same depletion region, whose width an applied voltage sets.

Carriers get about by drift and by diffusion. Drift is motion under an applied field, as in a metal; diffusion is motion down a concentration gradient and needs no field at all. Semiconductor devices use both, and the balance between them is what device equations are actually about. Metals effectively use drift alone.

The other material parameter is mobility, which says how fast carriers move for a given field. Electrons are generally faster than holes in the same material, and that asymmetry shows up directly in circuits: an n-channel device is smaller than a p-channel one of the same capability, so the two are not interchangeable in a layout.

A hole is not really a particle, whatever the cinema row suggested. It is the collective behaviour of a nearly full band, described as though it were a single object, and the description holds up because the mathematics comes out identical. Treating a hole as a real positive charge is safe for circuit-level reasoning, and it stops being safe once you have to account for effective mass, band structure or optical absorption.

All of that presupposes an extremely clean starting material. Doping at parts per million only controls a crystal whose unintended impurities sit far below that level, which makes semiconductor silicon among the purest bulk materials ever manufactured. Contamination here is a physics problem rather than a quality one: a stray impurity is an unplanned dopant.

Professional

Why silicon won, and where it is now losing

What settled it was chemistry. Silicon's native oxide, silicon dioxide, is an excellent insulator, grows directly on the crystal, adheres perfectly, and can be patterned. That one property made the planar process possible — grow an oxide, open windows in it, dope through the windows, use the oxide as both the mask and the permanent insulation — and the planar process made the integrated circuit possible. Germanium's oxide dissolves in water and is useless. The MOSFET, whose gate insulator is that same oxide, is the most manufactured object in history because of it.

Silicon's limit is thermal, and its band gap sets it. Leakage rises exponentially with temperature, so above roughly 150 to 175 °C at the junction a silicon device loses the ability to switch off cleanly, and thermal runaway turns from a theoretical failure mechanism into a real one. Maximum junction temperature on a power-device datasheet is that limit written down, and thermal design exists to keep the part under it.

Wide-bandgap materials go after exactly that ceiling. Silicon carbide's gap is 3.26 eV, which is 2.9 times silicon's; gallium nitride's is 3.4 eV. A larger gap brings a much higher breakdown field, so a device blocking the same voltage can be far thinner and therefore far lower in resistance, and it tolerates much higher temperatures. That combination is why electric-vehicle inverters, solar inverters and fast chargers have moved to SiC and GaN despite the cost.

Choosing a material commits far more than the part number. A faster-switching device produces steeper edges, and steeper edges stress insulation, generate more interference and demand better layout. A device that tolerates 200 °C is useless if its package, its solder and its board do not. What the material allows and what a design actually gets are different quantities, and packaging, gate drive and thermal design settle the difference between them.

A single crystal defect in the wrong place kills a die, so the cost of a working chip depends on defect density and die area rather than on what the material itself costs. Large power dies are expensive out of proportion to their size for that reason, and process control rather than materials dominates the industry's economics.

Silicon's band gap is indirect, which means an electron crossing it cannot readily emit a photon, so silicon makes a poor light emitter. Every LED is therefore a compound semiconductor picked for its gap, and the colour of an LED states directly which material it is made from. Absorption is a different matter: silicon absorbs light perfectly well, which is why photodiodes and solar cells are silicon while LEDs are not.