What is semiconductor IP, and why do chips contain circuits someone else designed?

Semiconductor IP is a block of circuit design, made and verified in advance, that others can license to put inside their own chips. It's also called silicon IP, an IP core or an IP block. Think of putting up a building: nobody invents their own elevator. You buy a ready-made one and install it by the manual. What's sold is the design, not a chip; the buyer wires it up to the circuits they drew themselves and hands the whole thing to a fab, which makes it as a single chip.

One thing to clear up first: silicon IP is sometimes shortened to SIP, which differs from SiP (system in package) only in the case of one letter. They're two different things. Silicon IP is a circuit design that still lives on a computer; SiP is finished chips packed together with resistors and capacitors into one small module (What is SiP?).

Why is it called "intellectual property"?

Intellectual property is something someone spent time and money to come up with; to use it, you need their permission and usually have to pay. That permission is a license. The chip made on the wafer still belongs to the buyer; the IP is just one block inside it.

Which parts of a chip are often bought?

The SoC in a phone is like a building, with the processor, graphics and communications each on their own floor (What is an SoC?). With more than ten billion transistors, hardly any team draws every block from scratch. Blocks that are often bought:

• Processor cores: the part that works through a program step by step.
• Memory: small blocks of memory inside the chip that hold data.
• Memory interfaces: circuits that pass data back and forth with the memory outside the chip.
• High-speed interfaces: circuits that send data to other chips, to a screen, or down the cable in a charging port. Both kinds of interface have to follow the industry's shared specifications, or other people's devices can't connect.
• Analog circuits: circuits that handle signals that change smoothly, such as the clock circuit that sets the beat for the whole chip, or a circuit that turns signals like sound into numbers (Analog vs digital chips).
• Standard cell libraries: the smallest building blocks, individual logic gates and tiny circuits that remember a 0 or a 1. When design software turns a circuit into logic gates, these are what it uses; sometimes the fab provides them, sometimes they're bought from another company (How are chips designed?).

What's the difference between soft IP and hard IP?

• Soft IP (a soft core): what you get is a circuit description written in a hardware description language, like a recipe in a cookbook. The buyer's own design software turns it into logic gates and decides where they go.
• Hard IP (a hard core): what you get is an already laid-out pattern, with every transistor and every wire in place, like a room built in a factory and delivered in one piece. It's drawn for one fab's particular process generation.

	Soft IP	Hard IP
What you get	A circuit description in text	A finished layout
Another fab, another process generation	Yes, by building it again	Not as is; it has to be redrawn and verified again
Speed, size, power	Depend on how the buyer builds it	Known in advance
Typical examples	Digital circuits such as processors	Analog circuits, high-speed interfaces

Analog circuits are usually hard IP: they're sensitive to the size and placement of every transistor, so if someone else rearranges them, they may not work right. Some people also name a third kind in between, "firm" IP.

What are the steps to using a block of IP?

Step 1: Choose the IP

You compare function, speed, power and area, and check whether there's a version for the fab and process generation your chip will use. That matters because a chip can only be made with one process. If it doesn't match, you wait for someone to make a version that does, or change fabs.

Step 2: Agree on a license

The contract sets which products it can go into and how you pay: a common setup is a license fee up front, then a royalty on every chip sold, though some deals are one-time payments. The seller spent a long time making and verifying the circuit, and spreads that cost by selling it to many companies. Without a deal, you don't get the design files, and you can't put it in a chip you sell.

Step 3: Get the design files and documents

You get more than the IP itself: models for the design software, documents on how to connect and configure it, and often a set of test programs to check it with. That way the buyer doesn't need to understand every wire inside and can connect it as a "black box". Miss something in the documents, connect one signal wrong or set one option wrong, and it may not show up until much later.

Step 4: Put it in your design and verify the connections

You wire the IP to your own circuits and simulate and check the whole chip together (What is chip verification?). The IP was verified on its own, and once it's connected, the two sides may not handle the clock, the power or restarting the same way. Miss that, and every block looks right alone but fails when connected; find it after tape-out, and you have to change the design and remake the photomasks.

Step 5: Tape out with the whole chip

The IP goes to the fab together with the rest of the chip's design; this is tape-out (What is tape-out?). The IP isn't a separate chip: it's printed on the same set of masks and the same wafers as every other circuit. If nobody checked before tape-out that it's the latest version with all known problems fixed, every chip made will carry that problem.

Why buy it instead of designing it yourself?

• Time: taking a complex circuit from nothing to fully verified takes several people a long time, and a chip that comes out late may miss its moment.
• Already verified: the seller has verified it, and some IP has already been made into real chips and measured. When many companies use it, problems tend to be found earlier.
• Follows the specification: every detail of an interface has to follow a shared specification, and buying one that's been built and tested makes mistakes less likely.
• People on what makes you different: buy the blocks that are the same for everyone, and save your effort for the blocks that make your chip different.

What are the risks of bought IP?

• Tied to a process: hard IP only works in the process generation it was drawn for. To change fabs or generations, you wait for someone to make a new version, and there may not be one.
• Waiting on someone else: the design inside belongs to someone else. You may not see the details and often can't change it yourself, so when there's a problem, you report it and wait for the seller to find and fix it.
• Paying for as long as you sell: with a royalty, the more you sell, the more you pay.

How does silicon IP relate to SoCs and chiplets?

The bigger the chip, the more it relies on IP: an SoC is built from many blocks, some bought, some designed and verified earlier inside the same company and reused. In the supply chain, IP sits at the very top, one step before anyone designs the whole chip (What does each part of the semiconductor supply chain do?).

Chiplets split one big chip into several pieces that are made separately and then put together (The news says "advanced packaging". What is it?). Chiplets placed together have to speak the same language, and the industry is pushing for shared interface specifications so that chiplets made by different people have a chance to connect. That interface circuit can be a block of bought IP too.

In fab24hr

There's no silicon IP in the game, and no chip-designer role; the main story, The life of a chip, starts after tape-out. Tap "Clock in" and choose "The life of a chip". The first time you play, you see the prologue, "Tape-out": "Glowmere", the first chip a small hearing-aid company has designed itself, is finished and checked, then goes to the fab to become a set of photomasks.

The game doesn't say what blocks Glowmere is divided into or which were bought. A real hearing-aid chip often has an analog circuit that turns the microphone's sound into numbers, and a clock circuit that sets the beat for the whole chip; circuits like these are often bought IP. Chapter 1, "How to Split 12", is the day before wafer start. You're the R&D integration engineer on the R&D line, seeing whether this design holds up the first time it's made.

The companies, chip and people in the game are fictional; what silicon IP is and how it's used come from public, general sources, and every company does it differently. For other jargon, see Chip fab jargon in plain English.