🔬 Semiconductors: How Chips Are Made

What you’ll learn

1. A Chip Is Printed, Not AssembledEstablish the course's through-line — that chips are printed in batches rather than assembled part by part — and set out exactly what this course covers and what its two sibling courses cover instead.The intuition that a chip is built by placing billions of tiny components is not merely wrong in scale but wrong in kind: nothing is ever placed. Every transistor on a wafer is made simultaneously, by treating the whole surface at once through a stencil, which is why cost depends on the number of steps rather than the number of transistors. That single fact — printing, not assembly — explains the economics, the fabs, the yields and Moore's Law, and it is the thread this course follows from sand to finished package.
2. Why Silicon?Understand that silicon won not because it is the best semiconductor but because of a chemical accident: its oxide is a superb, stable insulator that grows directly on the crystal.Silicon is not the fastest semiconductor — germanium is faster and gallium arsenide faster still — and it was not the first to be used. It won because silicon dioxide, the oxide that forms when you simply heat silicon in oxygen, is a hard, stable, high-quality insulator that grows perfectly bonded to the crystal beneath it. That single property provides both the stencil that masks the doping and the gate insulator inside every transistor, and germanium's oxide is water-soluble.
3. The Useful In-BetweenUnderstand what makes a material a semiconductor — a band gap small enough to cross but large enough to control — and why being mediocre at conducting is exactly the point.Whether a material conducts depends on whether electrons have accessible empty states to move into: metals have them freely, insulators have a gap far too wide to cross, and semiconductors sit in between with a gap small enough that a modest nudge can push electrons across it. That in-between position is the entire value, because a material that always conducts cannot be switched off and one that never conducts cannot be switched on. Silicon's usefulness comes from being controllable, not from being good.
4. Ruining the Crystal on PurposeUnderstand doping as the deliberate introduction of precisely-controlled impurities, and see why an industry obsessed with purity spends its effort putting specific dirt back in.Pure silicon is nearly useless, so fabs contaminate it deliberately: adding phosphorus, which has one spare electron per atom, makes n-type silicon with mobile negative carriers, while adding boron, which is one electron short, makes p-type silicon whose carriers are mobile positive vacancies called holes. The doping levels are extraordinarily dilute — parts per million or less — which is exactly why the starting crystal must be extraordinarily pure, since stray contamination would swamp the signal. Both types remain electrically neutral overall, and the entire art is in which regions get which dopant.
5. The Junction That Only Goes One WayUnderstand how simply placing p-type and n-type silicon in contact creates a one-way valve for current, and see that the boundary rather than the materials is where the device lives.Where p and n meet, mobile electrons diffuse across and fill nearby holes, leaving behind fixed ions whose charge builds an internal electric field that eventually stops further diffusion — creating a carrier-free depletion region with a built-in voltage across it. Applying an external voltage one way collapses that barrier and current flows freely; applying it the other way widens the barrier and almost nothing flows. Nothing was assembled to achieve this: the valve is created purely by the boundary, which is why doping different regions of one crystal is enough to build devices.
6. The SwitchUnderstand the MOSFET as a switch controlled by an electric field through an insulator, and see why the fact that its gate draws essentially no current is what makes billions of them possible.A MOSFET places two n-type regions in a p-type body, back-to-back junctions that block current in both directions by default. A metal gate sits over the channel between them, separated by a thin layer of silicon dioxide, and a positive voltage on that gate pulls electrons into the region beneath, temporarily inverting it into a conducting channel that links source to drain. The gate is insulated, so it draws essentially no steady current — it commands by field alone — which is why a chip can hold billions of switches without melting.
7. Sand to Crystal to WaferFollow silicon from ordinary sand to a single flawless crystal, and understand why the entire wafer must be one uninterrupted lattice.Silica is reduced to metallurgical silicon, purified chemically through a gas phase to reach semiconductor grade, then grown by the Czochralski process into a single crystal boule up to two metres long and weighing hundreds of kilograms — pulled from the melt on a rotating seed so the entire ingot inherits one atomic orientation. The boule is sawn into wafers and polished to an extraordinary flatness. The crystal must be unbroken because grain boundaries trap and scatter carriers, so a transistor straddling one would not behave as designed.
8. The Planar IdeaSee the planar process as the invention that made integrated circuits possible, by turning device-making into a sequence of flat, whole-wafer operations sorted by an oxide stencil.Jean Hoerni's planar process at Fairchild in 1959 kept everything flat and used a grown oxide layer both to protect the silicon and to mask where dopants could enter, so devices were defined by patterned windows rather than by handling. Robert Noyce saw that if devices were flat and covered in an insulator, metal could simply be deposited on top to wire them together — making a whole circuit in one piece of silicon. That is the moment the chip stopped being assembled and started being printed, and every process step since has been a variation on treat-everything-through-a-stencil.
9. The City of WiresUnderstand that most of a modern chip is not transistors but the many-storey metal network connecting them, and see why wiring — not switching — became the limiting factor.Transistors occupy only the bottom layer of a chip; above them sit ten or more levels of metal interconnect built in the back end of line, a stacked road network of vias and wires that carries signals and power. Modern chips have up to eleven or more metal levels produced in over 300 sequenced processing steps, and copper replaced aluminium via the damascene process because shrinking wires get proportionally more resistive. As features shrank, the delay of getting a signal across the wires came to dominate the delay of the switching itself.
10. Wafer, Die, PackageFollow the finished wafer through test, dicing, packaging and binning, and understand why the package exists and why chips of different speeds are often the same design.After 300-plus steps over roughly 11 to 13 weeks, the wafer is probed to find which dies work, then sawn apart — and only the good dies are packaged, because packaging a known-bad die wastes money. The package exists to translate between a die's microscopic pads and the millimetre-scale outside world, to carry heat away, and to protect the silicon mechanically. Binning then sorts identical designs by the speed and voltage they actually achieved, which is why a slower processor is frequently the same chip that simply tested less well.
11. Yield Governs EverythingUnderstand yield as the industry's true economic engine, and see why a single particle can kill a chip and why big dies are punished so severely.Because the fab's cost is per-wafer and nearly fixed, profit is decided almost entirely by what fraction of dies work — manufacturers are secretive about yields, but they can be as low as 30%. A chip has no redundancy, so a single stray particle landing in the wrong place can kill an entire die, which is what the cleanroom exists to prevent. Doubling die area more than doubles the chance of catching a defect, so yield falls sharply with size, which explains both why fabs are so obsessive and why the industry is moving toward smaller chiplets.
12. Moore's Law Was Never a LawLand the through-line by seeing Moore's Law for what it was — an economic observation about the cost-minimising number of components, kept alive by investment rather than by physics.Moore's 1965 paper 'Cramming more components onto integrated circuits' observed that the complexity giving the lowest cost per component had been doubling roughly every year, and he revised this to about two years in 1975; the familiar 18 months came from David House, not Moore. It is explicitly an empirical relationship rather than a scientific law — a claim about economics that nothing in physics obliges the universe to honour. It held for decades because the industry treated it as a target and spent whatever it took, which is exactly why it is now slowing as those costs rise.

Questions this course answers

Grounded in trusted sources

• 3 nm process — measured pitches and the naming convention — https://en.wikipedia.org/wiki/3_nm_process
• Doping (semiconductor) — https://en.wikipedia.org/wiki/Doping_(semiconductor)
• Integrated circuit packaging — https://en.wikipedia.org/wiki/Integrated_circuit_packaging
• Integrated circuit — https://en.wikipedia.org/wiki/Integrated_circuit
• Interconnect (integrated circuits) and the damascene process — https://en.wikipedia.org/wiki/Back_end_of_line
• MOSFET — https://en.wikipedia.org/wiki/MOSFET
• Monocrystalline silicon and the Czochralski process — https://en.wikipedia.org/wiki/Monocrystalline_silicon
• Moore's law: the 1965 paper, the 1975 revision, and why it is an empirical relationship rather than a scientific law — https://en.wikipedia.org/wiki/Moore%27s_law

Every Wunder lesson is built from real, reputable sources — never invented.

Related Science courses