What is fan-out packaging, and how does it move a chip's connections past its edges?

Fan-out packaging (fan-out wafer-level packaging, or FOWLP for short) is a way of packaging a chip that moves its connections out past the chip's edges. Chips that have been cut apart and tested are spaced out, the gaps around them are filled with plastic to make a disc that looks like a wafer, and fine copper lines are built on top, running outward from each chip's pads. That way, the solder balls underneath can spread over an area bigger than the chip. The "fan" in the name is those lines: they spread out from the chip like an open hand fan.

For the basics of packages and solder balls, see "What is chip packaging?" and "What do chip packages look like?".

What is wafer-level packaging?

In ordinary packaging, the wafer is cut apart first, and then each chip is put into its own case. Wafer-level packaging works the other way round: while the chips are still joined in one whole wafer, the protective layer, the pads and the solder balls are all built, and only then is the wafer cut. What comes off is already packaged. The whole wafer is done at once, thousands of chips at a time, and the packages are as small and thin as they get.

What is fan-in, and why can't chips with many connections fit?

When solder balls are built right on the wafer, they can only sit under the chip itself. The chip's pads are often around its four edges; in that case a layer of wiring first carries them inward to fill the underside. That's why it's called fan-in, and the finished package is the same size as the chip.

The catch is that the underside is only as big as the chip. Try some made-up numbers: a chip 3 mm on a side, with a solder ball every 0.4 mm, fits 7 balls in a row and at most 49 over the whole face. A chip that needs more than a hundred connections won't fit.

The balls can't just keep shrinking and crowding together either: the solder pads on a circuit board can't be made that dense, and balls that are too small don't hold well. Making the chip bigger just to fit the balls doesn't pay, because silicon is expensive. So fan-in is mostly used for small chips without many connections.

How does fan-out put solder balls outside the chip?

Cut the wafer first, then give each chip a border of extra space: spread the chips out with a gap between them and fill the gaps with plastic. The ring of plastic around each chip can carry wiring and solder balls too. Take the same 3 mm chip, add 1 mm on every side, and it becomes 5 mm square: 12 balls in a row, 144 over the face, about three times as many.

How is fan-out packaging made? Step by step

One common way is "chip first, face down":

Step 1: Pick out the good chips

• What: every chip on the wafer is first tested with fine needles (wafer sort), then the wafer goes through dicing, and only the good chips are taken.
• Why: every chip that gets packaged costs its share of wiring and solder balls.
• Otherwise: bad chips get packaged too, and the work on them is wasted.

Step 2: Place them face down on a temporary carrier

• What: a round, very flat rigid plate (the temporary carrier) gets a layer of adhesive, and the chips are placed on it one by one, pad side down, with space between them.
• Why: all the pads sit on one flat plane, so once the carrier comes off they are nearly level with the plastic and wiring can be built on them. The space between chips is where the solder balls will go.
• Otherwise: a chip that is placed off or turned slightly won't line up with the wiring built later.

Step 3: Mold it into a "reconstituted wafer"

• What: mold compound (packaging plastic, usually an epoxy) covers the whole tray and fills the gaps, then is heated until it hardens into one disc, called a reconstituted wafer.
• Why: the plastic holds the chips in place and gives each one its border of extra space. Shaped like a wafer, it can then be worked on as one piece.
• Otherwise: the flowing plastic pushes the chips, and it shrinks as it hardens, so chips drift from where they were placed. Silicon and plastic also expand and shrink by very different amounts with heat, so once it cools the whole disc warps, like a cookie that bent in the oven.

Step 4: Remove the temporary carrier

• What: the carrier and the adhesive come off, exposing the pad side, nearly level with the plastic around it.
• Why: the pads have to be exposed before anything can connect to them; the carrier was only a temporary support.
• Otherwise: adhesive left on the pads means the wiring won't connect well, and pulling too hard can damage the disc.

Step 5: Build the redistribution layer to carry the pads outward

• What: coat an insulating layer and open small holes over the pads. Lay down a very thin metal base, coat photoresist, expose it and develop it to mark out the lines, grow copper lines by electroplating, then remove the resist and the leftover base. The copper lines run from the pads out to the outer ring, and if one layer isn't enough, another is stacked on. These layers of wiring are called the redistribution layer (RDL).
• Why: printing them the way a fab's litho does ("What is photolithography?") makes the lines fine and dense, and the whole disc is worked on together.
• Otherwise: the pattern is printed where the chips are supposed to be. If they drifted or the disc warped in step 3, the holes and lines miss the pads, like an overlay error in a fab: some don't connect, and some touch their neighbors.

Step 6: Add the solder balls

• What: solder balls are placed on the copper pads of the top layer and heated so the solder melts and bonds.
• Why: the balls are the feet that get soldered to the circuit board, and they can sit on the ring outside the chip.
• Otherwise: if a ball is too small or falls off, those spots won't connect when the package is soldered to the board.

Step 7: Cut it apart and test again

• What: the disc is cut along the plastic between the chips into single packages, which are tested again (final test).
• Why: each one is now a packaged chip that can be soldered straight onto a circuit board.
• Otherwise: a chip damaged in the earlier steps, or a line that didn't connect, ends up in a product.

What's good about fan-out packaging?

• Room for more connections: the solder balls spread past the chip, so the chip's size no longer limits them.
• Thin: flip chip solders the chip onto a small board ("What is flip chip?"); fan-out builds the wiring right on the chip and skips that board, so it's thin and very common in phones.
• Short paths: in the chip-first way, the pads connect straight to copper lines, with no bumps, wires or board in between.

The lines in a redistribution layer are commonly a few micrometers to about ten micrometers wide (a micrometer is a thousandth of a millimeter), far finer than the lines on a circuit board.

What's hard about fan-out packaging?

• Chip shift and warpage: the two problems from step 3. Every layer of wiring has to line up, so the molding material and conditions have to be just right.
• Good chips lost along with the rest: with chip first, a chip can't come out once it's molded in, so if any later step goes wrong, the good chips are scrapped too. That's why there is also the reverse order, "RDL first": the wiring is built on the carrier and checked first, and then good chips are attached.
• Lines not as fine as on silicon: when chips need to pass the most data between them, a silicon interposer is often used instead.

Can several chips be packaged together?

Yes. In step 2, several chips are placed side by side and molded together, and the redistribution layer also connects them to each other. No separate piece of silicon is needed, so it costs less, but the lines are coarser. For why chips are split apart and put back together, and the other ways to do it, see "What is a chiplet?" and "The news says "advanced packaging". What is it?".

In fab24hr

There is no fan-out packaging in the game. In chapter 11 of the main story, The life of a chip, "The Ball Lifted", you are the assembly process engineer at a test-and-assembly house. The hearing-aid chip "Glowmere" is stuck face up onto the small board of an SiP module (die attach), connected out with fine wires (wire bonding), and then covered in plastic (molding). The note at the wire bonder says one chip has a dozen or so wires; with so few connections, there's no need to move them outside the chip.

The closest thing is molding. Tap "Clock in", choose "A day in the fab", tap "Test-and-assembly house" on the "Your role today" card and swipe to the assembly process engineer; the shift follows the time on your device. On the evening shift, the molding machine has a new mold, and in the two cavities farthest from the inlet, the fine wires have been pushed out of place by the plastic. You choose between lowering the injection speed (too slow leaves bubbles) and X-raying every tray. When a fan-out package is molded, what the plastic pushes is the whole chip.

The die bonder also picks up only the chips that wafer sort marked good, just as in step 1, and the codex's "Machines" lets you take the molding machine and the scanner apart.

The test-and-assembly house, the chip, the people and the machines in the game are fictional; what fan-out packaging is and how it's made follow public, general sources, and the chip size and ball spacing used to count solder balls are made-up numbers. For other jargon, see Chip fab jargon in plain English.