What is ion implantation, and how does a fab shoot atoms into silicon?

Ion implantation is how a chip fab turns atoms such as boron or phosphorus into charged ions, speeds them up with an electric field and drives them straight into a silicon wafer. It is the most common way to "dope" silicon today: to add a tiny amount of other atoms on purpose, deciding which small regions conduct electricity and how well. The two ends of a transistor in a chip are often made this way.

Why shoot atoms into silicon at all?

Very pure silicon barely conducts. It needs a tiny amount of other atoms added, and that is called doping: phosphorus (or arsenic) makes n-type silicon, boron makes p-type. Only a little goes in, yet the way it conducts changes completely. What is a semiconductor? explains why, starting from the beginning.

A chip doesn't want the whole wafer to conduct. It wants "this tiny patch conducts, the patch next to it doesn't." Put small n-type and p-type regions side by side, add a gate, and you have a transistor, a tiny switch controlled by electricity (see What is a transistor?). So doping means the right atom, in the right place, in the right amount, at the right depth. Ion implantation is the tool for that, and in a fab it has an area of its own (for what all seven areas do, see What do the seven process areas of a chip fab do?).

How do the atoms get "shot" in?

Think of throwing pebbles into a sand pile: the harder you throw, the deeper a pebble buries itself; the more pebbles you throw, the more end up in the pile. An ion implanter throws atoms, and it uses electricity to do it.

1. Give the atoms a charge: a gas containing boron or phosphorus goes into the tool's ion source and becomes charged atoms, which are ions. Once they carry a charge, an electric field can push them.
2. Speed them up with an electric field: the field keeps pushing the ions, faster and faster, and they form a thin ion beam. The whole path is under vacuum so the ions don't hit air on the way.
3. Cover the whole wafer evenly: the beam has to sweep across the whole wafer so every spot gets about the same amount.
4. Count while shooting: the ions carry charge, so the ions arriving are an electric current. The tool measures that current, multiplies by time, knows how much has gone in, and stops when it's enough.

Inside the silicon, an ion knocks silicon atoms aside and slows down as it goes, until it stops at some depth.

Energy and dose: how deep, and how many?

The two most important numbers in implantation are energy and dose. Together with the type of atom, they are written in the recipe the tool runs.

• Energy sets the depth: the harder the push, the deeper the ions go. Typical depths are mostly between 10 nanometers and 1 micrometer (a thousandth of a millimeter).
• Dose sets the amount: the dose is how many atoms go into each small patch of area. Per square centimeter (about the size of a fingernail), a low dose is about a hundred billion atoms and a high one can reach ten million billion, a hundred thousand times more. The more go in, the better that patch conducts.

So every implant step has to state three things clearly: which atom, how much energy, how big a dose. Get any one of them wrong and the region comes out wrong.

How do you implant one place and not another?

The beam sweeps the whole wafer, but only some small regions should be implanted. The answer is to cover the rest with photoresist first. Photoresist is a thin film that changes when light hits it; litho prints the pattern onto it and opens up the places to be implanted, so the resist that's left works like a stencil (how the printing works: What is photolithography?). Ions that hit the resist stop inside it; only ions that hit an opening get into the silicon. After the implant, the resist is removed.

Photoresist can serve as the stencil because implantation doesn't need high heat. Early doping put wafers into a hot furnace containing the dopant and let the atoms slowly seep in on their own, and photoresist can't survive that heat. Implantation "shoots" the atoms in, so both how many and how deep are tightly controlled, and that's why it largely took over.

A chip isn't implanted only once, either. N-type and p-type regions are done separately, different parts of the transistor each get their own implant, and each new place to implant usually needs a new stencil printed first. In one process flow, implantation often happens around ten times, or more.

Why heat the wafer afterwards?

Silicon atoms start out lined up like a neatly built brick wall. One ion racing in can knock more than a thousand silicon atoms out of place along its path. After the implant, that layer is a mess, and most of the implanted atoms sit in the wrong spots, not conducting yet.

So the wafer is heated after implantation, which is called annealing. At high temperature the silicon atoms have the energy to line up again, and the implanted atoms take the place of silicon atoms. Only then do they "get to work" and make that region conduct. Annealing can be done by baking in a furnace, or with rapid heating that lasts only seconds (see What does a furnace do in a chip fab?).

Heat has a cost too: at high temperature the atoms move a little further sideways and deeper, and the smaller the structures, the less of that they can take. This account, which can't be refunded once spent, is the thermal budget. So an anneal has to be hot enough, but not too long.

Why is a wrong implant so hard to fix?

Because atoms that went in can't come out. They are mixed into the silicon, sitting in line with the silicon atoms, and there is no way to pick them out one by one.

Compare it with litho. If a litho pattern comes out misaligned, then as long as nothing else has been done, the resist can be stripped and printed again; that is rework (see Why is the lithography area in a chip fab yellow?). Implantation has no such way back: wrong energy, and the atoms sit at the wrong depth; wrong dose, and the region conducts the wrong amount; the wrong kind of atom, and even n-type and p-type can be reversed. Wafers like that mostly have to be stopped, measured and judged on whether they can still be used, and if not, scrapped.

On top of that, a wrong implant looks exactly like a right one from the outside. Often it only shows up in measurements, or even when the electrical tests are run much later. So the most important check in implantation comes before the run: atom, energy and dose, checked one by one before the wafers go into the tool.

How do you know the implant was right?

You can't see the atoms, but you can measure how well that layer now conducts, and the most common number for that is sheet resistance. If too few atoms went in, the sheet resistance comes out high; if too many, it comes out low.

A common approach is to implant a monitor wafer at regular intervals: it never ships and exists only to check the tool. After the implant, it is heated so the atoms get to work, and then four probes in a row measure the sheet resistance. Many points are measured across one wafer to see how even the implant is; the numbers are logged every day, and when they drift to one side, someone checks the tool.

The tool counts while it shoots, too, but if the part that measures the current is off, it stops at that wrong count and believes the dose was just right. That is why both the tool's records and the monitor wafer's sheet resistance get looked at. For more on measuring, read How does a chip fab know each step went right?.

In fab24hr

In fab24hr, the production fab's front-end bay has an ion implanter, IMP-01. Tap "Clock in", choose "A day in the fab", then on the "Your role today" card tap "Production fab", swipe to the process engineer and tap "Implant" under the name: you're the implant process engineer, and the atom, energy and dose in the recipe are yours. On the late-night shift, IMP-01 stops by itself halfway through a lot, and the monitor wafer's average sheet resistance is within spec while the map is all in stripes; on the morning shift, after the beam current is retuned, the sheet resistance comes out slightly high while the tool says the dose was just right; on the afternoon shift, two finished lots have their resist stripped and go into the furnace for annealing tonight; on the evening shift, you check the recipes of the three lots due tonight line by line, and two lines don't match the work order. On the same tool you can also be the implant equipment engineer, looking after the ion source, the vacuum and the cup that measures the beam; the R&D line has an implant role too. The companies, people, tool numbers, lot numbers and figures in the game are fictional; what ion implantation is and why a wrong implant can't be undone follow public, general sources. For other jargon, see Chip fab jargon in plain English.