What is a probe card, and how does every chip on a wafer get tested?

A probe card is the tool used for testing chips on the wafer: a board with many very fine needles, laid out to match the positions of a chip's pads. It touches down on the pads all at once, sends the test machine's signals into the chip, and carries the results back. Before the wafer is cut, every chip on it takes this electrical test. That is called wafer sort, also wafer probing, wafer test or CP.

What wafer sort does as a whole, and how it differs from final test after packaging, is covered in How are chips tested?. This article stays with the needles: how they touch down, why they get dirty, and why each kind of chip needs its own card.

Where does a probe card sit?

The small patches of metal on a chip's surface are its pads; power and signals go in and out through them. The pads sit right next to each other, often only a few tens of micrometers apart, about the width of a hair. The test machine's connectors are far bigger. The probe card sits between the two, a bit like an adapter:

• The test machine: the computer end. It follows the test program, sends signals and compares the results with the right answers. It connects through a large box called the test head, which sits right on top of the probe card.
• The probe card: at heart, a circuit board. Its outer edge connects to the test head, and its wiring packs closer and closer as it runs to the center, where it reaches the needles.
• The wafer: it lies on the chuck of a probe station (also called a wafer prober). The chuck moves front, back, left and right, and also pushes up.

So the signal goes: test machine → test head → probe card → needle tip → pad, and the result comes back the same way.

Step by step: how does a probe card test a wafer?

Step 1: Build a card for this chip's pads

• What: The needles are placed to match the position of every pad on this kind of chip, so that each needle tip lands on exactly one pad.
• Why: The needles are fixed to the card and can't move by themselves, and every kind of chip lays out its pads differently.
• Otherwise: The needles land off the pads, nothing connects, and good chips measure as bad.

Step 2: Load the wafer into the probe station and line it up

• What: A robot arm puts the wafer on the chuck, and cameras find where the chips and the needle tips are. Once they are lined up, the first chip moves to just below the needles.
• Why: A pad is only a few tens of micrometers across. Tilt the wafer a little and hundreds of needles go off target together.
• Otherwise: The needles land on the edges of the pads, the contact comes and goes, and the same chip gives different results twice in a row.

Step 3: Press the needles down, then slide a little

• What: The chuck pushes the wafer up until the pads touch the needle tips, then a little further. The needles are springy, so they bend slightly and each tip slides a short way across its pad, leaving a tiny scratch called a probe mark.
• Why: Many chips have aluminum pads, and aluminum grows a very thin oxide as soon as it meets air. That oxide barely conducts, so the tip has to scratch through it to reach the metal underneath, like scratching the silver coating off a scratch card.
• Otherwise: Press too lightly and the tip doesn't get through: the resistance reads high and the numbers jump around. Press too hard and the mark goes too deep, damaging the pad and the layers below it, so wire bonding on that pad later may not hold.

Step 4: Test one chip, move to the next, or test several at once

• What: When the test machine has gone through every item, the chuck drops, moves to the next spot and pushes up again. Many cards touch several chips at once, and each position uses its own set of needles on the card.
• Why: A wafer often holds hundreds or thousands of chips. One at a time is too slow.
• Otherwise: Testing one chip at a time takes several times longer per wafer. But with several at once, if one set of needles gets dirty, failures keep showing up in the same position.

Step 5: Draw the results as a wafer map

• What: Each chip is sorted into a bin by its result, and the results are drawn by position as a wafer map.
• Why: After the wafer is cut, the next machines follow the map and pick up only the good chips for packaging. The shape the bad ones make is also a clue to what went wrong.
• Otherwise: Bad chips get packaged too, and the cost of packaging them is wasted. For reading the map, see What is WAT?.

What kinds of probe needles are there?

Type	What it looks like	Why it's used
Cantilever	Needles reach out from the card at a slant, like tiny fishing rods, with the tips bent down	Quick to make and cheaper, still very common; suits chips whose pads run around the four edges
Vertical	Needles stand straight up and down	They can be arranged in a full grid and reach pads in the middle of the chip
MEMS	Needles are "grown" with the same kind of methods used to make chips, such as lithography and plating	Finer, denser and more precisely placed, for chips with the most and the closest pads; the denser and finer a card, the more it usually costs

Why do the needles need cleaning so often, and what else is checked?

Every touchdown scrapes off a little oxide and metal, and some of it sticks to the tip. Over many touchdowns, the contact resistance keeps creeping up. So after a set number of touchdowns, the needles are pressed a few times onto a cleaning pad next to the probe station to scrub the debris off. Cleaning also wears away a bit of each tip, so how often to clean is a trade-off.

On top of cleaning, the card is measured regularly:

• Planarity: hundreds or thousands of needle tips have to sit at the same height. The shortest one needs a lot of pressing to make contact, and then the longest one gets pressed too hard.
• Alignment: the front-back and left-right position of each tip slowly drifts.
• Contact resistance: when it gets too high, it's time to clean.

When the needles have worn down too far, they are repaired or the card is replaced.

Why does a dirty needle look like a process problem?

A dirty tip makes the resistance read high, which looks a lot like a wafer that really was made wrong: a high resistance can mean the process went off, or simply that the needle didn't make good contact. Good chips then get rejected as bad; test engineers call this overkill.

That's why, when the numbers look wrong, people often suspect the measurement first and the process second. A few clues:

• The failures are always in the same position: of the positions tested together, the same one comes out low in every lot. It's very unlikely that every lot of wafers happens to fail in that one spot.
• Cleaning the needles and probing again fixes it: if the numbers come back, it was the card. If the chip still fails, then it's time to look at the process.

Probing again can't go on forever, though. A chip sitting right at the pass line will pass sooner or later if it's probed enough times, which amounts to loosening the limit, and every touchdown leaves one more probe mark on the pads. Wafer sort only answers whether a chip is good now; for whether it will fail soon after it goes into use, see What is burn-in?.

Why does each kind of chip need its own card, and why is it so expensive?

Because the pad positions belong to the chip. A different product, or a redesigned or shrunk version of the same chip, moves the pads, and that usually means a new card.

Making a card isn't simple either. A card may have several thousand needles or more, and each one has to land precisely on a pad a few tens of micrometers wide. Many cards also have their needles fitted by hand, one at a time. And each kind of chip needs only a small number of cards, so they can't be mass-produced like ordinary parts to spread the cost: it's small batches of many different designs. That's why a card is reused for a long time.

In fab24hr

Tap "Clock in", choose "A day in the fab", then on the "Your role today" card tap "Test-and-assembly house" (see What is an OSAT?) and swipe to the test engineer. On the late-night shift (shifts follow the time on your device), the probe card touches four chips at a time, and Position 3 comes out low in every lot. Do you deal with the card first (clean the needles or swap in the spare card), or retest the failed chips first? And how many times do you retest?

In the main story, The life of a chip, chapter 3, "The shape of a failure", wafer 9 from the R&D line has a contact resistance above spec, and you decide whether to probe it again or write it up as a process problem. In chapter 10, "Standby Current", when Ms. Zheng, the senior test engineer, sees the failures, the first thing she asks is also "Did we measure it wrong?"

In the codex's Machines list, the probe card and the wafer prober both have Take it apart, and the probe card's See it work plays it step by step, from taking out the card and touching down to cleaning the needles.

The test-and-assembly house, the chip, the people and the numbers in the game are fictional, and four chips per touchdown is the game's own setting; what a probe card is and how it's used come from public, general sources. For other jargon, see Chip fab jargon in plain English.