What is flip chip, and how does a chip connect through solder bumps?

Flip chip is a way of connecting a chip to the outside by turning it over. First, a small bump is built on every pad on the chip's surface. Many bumps are made of solder, so people also call them solder balls. Then the chip is turned face down, the bumps are lined up with the pads on a board, and heat solders them on. There are no wires to run one by one: every connection is made at once.

How is flip chip different from wire bonding?

A chip is built on a wafer. The circuits are on its front face, and so are the pads, the small metal patches it uses to take in power and send signals.

In wire bonding, the chip is stuck face up onto a small board (die attach), and metal wire finer than a hair runs from the pads along its four edges to the board, one wire at a time (see "What is wire bonding, and how does a chip connect to the outside?").

Flip chip turns the chip over, face down. Once the bumps are soldered, the chip is both held in place and connected: die attach and wire bonding become one step (glue filled in underneath later makes the hold stronger).

	Wire bonding	Flip chip
Chip's front face	Up	Down
How it connects	Fine wires, one at a time	Small bumps, all soldered at once
Where the pads go	Mostly the four edges	Anywhere on the face
Signal path	Each wire has some length	Only the height of a bump

The idea isn't new. It goes back to the 1960s.

Why turn the chip over?

More connections. Wire-bond pads sit along the four edges, like a table where people can only sit around the rim. Flip chip can use the whole face. Try some made-up numbers: a chip 10 mm on a side, with wire-bond pads every 0.05 mm, has about 800 pads around its edges. Flip-chip bumps are bigger, say one every 0.2 mm, but they can fill the whole face: 50 rows of 50 is 2,500. The bigger the chip, the bigger the gap, because the edges grow only with the side length while the face grows with the area.

Shorter paths, faster signals. The longer a wire, the more a fast-changing signal gets slowed down and distorted, and the more is lost along the way. A bump is only a tiny bit tall, so signals get in and out of the chip fast.

Cooling and space. Once the chip is flipped, its back faces up, so a metal lid or a heat sink can sit right on it, and heat can also flow down through the bumps. With no room needed around the chip for wires, the package can be a little smaller too.

How are the bumps made?

The bumps are usually made while the wafer is still whole, on every chip at once. Roughly:

1. Lay a base: a few very thin layers of metal go on every pad, so the solder will stick and won't react with the metal underneath.
2. Open small holes: a layer of photoresist is coated on, then exposed and developed the same way as in a wafer fab, leaving a small hole over each pad.
3. Fill with solder: electroplating grows solder into the holes little by little (some lines print it instead).
4. Melt into balls: the photoresist is washed off and the wafer is heated once more. The solder melts, and surface tension pulls it into small round balls, like water beading on a lotus leaf.

Traditional bumps are spaced roughly one to two hundred micrometers apart (a micrometer is a thousandth of a millimeter). The even smaller "microbumps" used in advanced packaging are only a few tens of micrometers apart, about the width of a hair.

How does the chip get flipped and soldered?

After the wafer is diced into separate chips:

1. Add flux: a little flux goes on the bumps or on the board's pads, so the solder melts and wets properly when heated.
2. Flip and align: a machine picks up the chip, turns it over, and uses a camera to see both the bumps and the board's pads, then sets the chip down once they line up.
3. Reflow: in an oven, the solder melts and joins the board's pads, then sets as it cools. Melted solder has surface tension, so a chip placed slightly off often gets pulled back to the center.
4. Underfill: a very narrow gap is left between the chip and the board. A line of liquid glue is dispensed along the chip's edge, and it creeps into the gap on its own and fills the space between the bumps, the way water climbs up a paper towel when you dip one corner. Heat then hardens the glue.

What does the underfill protect?

Things expand when they warm up and shrink when they cool, but every material by a different amount. Silicon barely expands. The common boards are mostly made of resin and expand several times as much.

Every time the chip warms up in use and cools down again, the small bumps in between get pulled in two directions. Bent back and forth long enough, they crack, the way a wire you keep bending eventually snaps. The bigger the chip, the harder the bumps far from the center get twisted.

The underfill glues the chip and the board together over the whole area, so the whole face shares the pull, and the solder joints survive many more hot-and-cold cycles. It's different from molding: molding wraps the whole chip in plastic, while underfill only fills the gap under the chip.

What are the downsides of flip chip?

• Cost: building the bumps adds steps, and the board needs very dense pads and very fine lines to meet that many bumps. For a chip without many connections, wire bonding is usually cheaper.
• Hard to repair: the joints are hidden under the chip, so X-rays or ultrasonic scans are needed to see whether they're soldered and whether the underfill filled the gap. Once the glue hardens, the chip is very hard to remove, and a bad one often means the whole package goes in the bin. So before soldering, every chip is tested first (wafer sort), and only good ones are used.
• Heat stress: the pulling described above; the board can also warp as it heats up.
• Decided at design time: the pads and the board both have to be planned for flip chip. Wire bonding is more flexible.

Are flip-chip bumps the same as the solder balls under a package?

No. They're very different in size. The bumps under the chip are the smallest, soldered onto the board. Under that board is a grid of much bigger solder balls that go onto the circuit board (the ones in "What is chip packaging?"). The board in the middle works like an adapter: small, dense bumps on top, big solder balls below.

How is flip chip related to advanced packaging?

The advanced packaging in the news places several chips side by side on a thin slice called an interposer, or stacks them (The news says "advanced packaging". What is it?). Many of those chips are connected by flip chip: each chip is turned over and soldered onto the interposer with tiny microbumps, the interposer connects to the board with bigger bumps, and the board connects to the circuit board with even bigger solder balls. Three layers of balls, each bigger than the one above.

Stacked memory chips are also often soldered layer to layer with tiny bumps (The news says "HBM". What is it?). For connections denser than microbumps, there is also hybrid bonding, which uses no solder and joins the copper pads of two layers directly.

In fab24hr

There is no flip chip 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. On this assembly line, the "Glowmere" chip is stuck face up onto the small board of an SiP module, wire-bonded out with fine wires, and then molded. The note at the wire bonder says one chip has a dozen or so wires. With that few connections, wire bonding fits well.

The closest thing to flip chip is the long reflow oven on the SMT line: solder paste is printed, resistors and capacitors are placed on it, and one pass through the oven melts the solder and lets it set, holding them in place. Soldering a flip chip works on the same idea, only the part being soldered is a chip turned upside down.

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. You get one of four shifts by the time on your device: boards waiting for the wire bonder late at night, the dicing saw's blade in the morning, the chapter 11 pull test in the afternoon, and wires pushed out of place by molding in the evening.

The test-and-assembly house, the "Glowmere" chip, the people and the machines are fictional; what flip chip is and how it's done follow public, general sources, and the chip size and spacing used to count connections are made-up numbers. For other jargon, see Chip fab jargon in plain English.