What is doping? How n-type and p-type silicon make a switch that's normally off

Doping means deliberately adding a tiny amount of other atoms to very pure silicon to change how it conducts. Add atoms that have one more electron than silicon, like phosphorus, and the silicon gains electrons that can move: that's n-type. Add atoms with one fewer, like boron, and it gains empty spots: that's p-type. Put n-type and p-type in the right order and you get the switch inside a chip, the transistor: off by default, and on only when a voltage is put on its gate.

Why does pure silicon barely conduct?

A silicon atom has four outer electrons. Think of them as four hands, each holding on to a neighboring silicon atom, so the atoms line up in a neat grid. The electrons are tied up between the atoms. They can't move, so current can't flow. For the version that starts from zero, see What is a semiconductor?.

To move, a tied-up electron first has to jump over a step. Textbooks call that step the band gap. Metals have no step: their electrons can move from the start. Glass and plastic have a very high step, and almost no electron gets over it. Silicon's step is in between: at room temperature, heat now and then knocks an electron over it, but only about one for every few trillion silicon atoms. That's why pure silicon barely conducts.

What happens when you add a few other atoms?

The amount is tiny: at the low end, about one atom for every hundred million silicon atoms; at the high end, about one in ten thousand, and in some places more.

• Add phosphorus, get n-type. Phosphorus has five outer electrons. It takes the place of one silicon atom, holds hands with four neighbors, and has one left over. That electron is held only loosely: its step is much lower than silicon's, so at room temperature most such electrons are free to roam. What moves is a negatively charged electron, so it's called n-type.
• Add boron, get p-type. Boron has only three outer electrons, so one neighbor's hand has no one to hold, and that leaves an empty spot, called a hole. Picture a parking lot with a full row of cars and one empty space. The car behind pulls forward into the space, and the space moves back to where that car was; the next car pulls forward too, and so on. The cars move forward one by one, and the empty space travels backward. What really moves is electrons, but it looks like a positively charged empty spot on the move, so it's called p-type.

The silicon as a whole isn't charged. N and p only say whether what moves inside is an electron or an empty spot.

What happens where n meets p?

Where n-type and p-type meet, current can only go one way. Push from the p side to the n side and it gets through. Along the boundary there is always a thin zone with nothing free to move; push the other way and that zone grows wider, like a wall, and blocks it. A part that does only this is called a diode. Keep that wall in mind ("push the wrong way and it blocks"), because the next section needs it.

Why is a transistor normally off?

A transistor has an entrance (the source), an exit (the drain), and a gate above the middle, with a very thin insulating layer between the gate and the silicon (what it looks like: What is a transistor?). In a common kind, the doping is laid out like this:

1. Normally: the source and drain are n-type, and the big block under the middle is p-type. To get from source to drain, current has to cross two boundaries, n to p and p to n, back to back. Whichever way you push, one of the two is the wrong way, and its wall blocks the current. So the switch is off by default.
2. Put a positive voltage on the gate: through the insulating layer, it pulls electrons to the surface of the p-type silicon and pushes the empty spots away. With enough voltage, the thin layer at the surface has more electrons than empty spots, so it starts behaving like n-type. Textbooks call it the inversion layer.
3. On: that layer is a temporary n-type bridge. Source, bridge and drain are all n-type, there's no wall in between, and current gets through. Take the voltage away and the bridge falls apart, and the switch is off again.

The voltage it takes to lay that bridge is the threshold voltage. The more heavily the p-type block underneath is doped, the more empty spots it has, the more electrons it takes to outnumber them, and the higher the threshold. So doping decides not only n or p, but also how much voltage it takes to turn the switch on.

The other kind of switch works the other way round

The other kind of transistor swaps n and p: the source and drain are p-type and the block underneath is n-type, so it's p, n, p. It's off when the gate voltage is high and on only when the gate voltage is low, and its bridge is made of empty spots.

Most chips today pair the two kinds. One connects the output to the power supply, the other connects it to ground, and only one of them is on at a time. Whether each switch is on or off decides whether the line is connected to the supply (1) or to ground (0). This pairing is called CMOS: see What is CMOS?. Because both kinds are needed, the same piece of silicon needs both a p-type and an n-type "base", often each made as one large doped area, called a well, with one kind of transistor sitting in each.

How does a fab dope silicon?

1. Print a stencil first. Litho opens windows in a layer of photoresist where the doping should go. Only certain small areas should become n or p; if the rest isn't covered, the whole wafer gets changed.
2. Drive the atoms in. Today this is mostly ion implantation: the atoms are turned into charged ions, sped up by an electric field and shot into the silicon, with good control over how many go in and how deep (details: What is ion implantation?). Before that, doping was mostly done by high-temperature diffusion: the wafers sit in a very hot furnace and the atoms slowly seep in from the surface, and some steps still use it today (see What does a furnace do in a chip fab?).
3. Anneal. The implanted atoms have to settle into places where silicon atoms would sit before they add electrons or leave empty spots, and the knocked-about grid has to line up again. Without an anneal, that area doesn't conduct the way it was designed to. But heat it for too long and the atoms wander sideways, so many anneals now heat one wafer at a time for just a few seconds.
4. Do it again somewhere else. The n-type well, the p-type well, the layer that sets the threshold, and the source and drain of each kind of transistor are all implanted separately, so one process flow has many implant steps. How these steps come together into one transistor: How is a transistor made?.

What happens if the doping is wrong?

Implanted atoms can't be taken out, and the wafer looks exactly the same from the outside. Wherever the mistake is, that's where the switch comes out off target:

• Too much in the block underneath: the threshold goes up, the switch is harder to turn on, and it's slow.
• Too little in the block underneath: the threshold goes down and the switch doesn't shut tightly. When the gate says "off", some current still slips through: that's leakage. When the source and drain are very close together, they can even connect underneath, where the gate can't shut them off. A chip has hundreds of millions of switches or more, and if each one leaks a little more, the chip uses power even while it sits unused (see Why do chips use power?).
• Too little in the source and drain: the entrance and exit have more resistance, so less current flows when the switch is on.

That's why fabs measure the sheet resistance of a monitor wafer to check how much went in, and why, once the wafers are finished, WAT measures each switch's threshold voltage and leakage on test keys (see What is WAT?).

In fab24hr

Chapter 2 of the main story, The life of a chip, is "Station by Station": the first five days after wafer start. You're the R&D integration engineer, following the first 12 wafers of a fictional hearing-aid chip, "Glowmere", through the seven areas of the R&D line. On Day 5 you go to the implanter IMP-R1 and first check the traveler against the tool's recipe line by line: the tilt angle doesn't match, because the tool still has the version from the last experiment, which implanted straight on. Once it's fixed, you implant the N-well, put down a new layer of photoresist and implant the P-well; that night the wafers go into the furnace FURN-R1 for the anneal.

You can find it in A day in the fab too: tap "Clock in", choose "A day in the fab", then on the "Your role today" card tap "R&D line", swipe to the R&D process engineer and tap "Implant" under the name. How much goes into Glowmere's source and drain, and how deep, is yours to work out; on the late-night shift, one wafer was run with the wrong recipe and got the dose meant for the "fast" group. The production fab has an implant role too. The companies, chip, people, tool numbers and figures in the game are fictional; how doping works and how a transistor switches follow public, general sources. For other jargon, see Chip fab jargon in plain English.