What is new product introduction (NPI), and who takes over at each step from a new chip's pilot run to volume production?

New product introduction (NPI) is the stretch that takes a newly designed chip from the first time it is made in a fab to steady, high-volume shipments. At many companies, NPI also means the person or team who ties that stretch together. Plenty of factories that make products use the term; in the chip business it usually means the stretch where a new chip goes into production.

The path is like a relay race: design, the fab, test, reliability, the customer and the production fab each run one leg. The person running the NPI is like the coach, planning where each handoff happens and watching every one.

How is introducing a new product different from developing a new technology?

How long does a new chip technology take to go from a paper to mass production? is about a new way of making chips, such as a new material or structure, moving from the lab to the factory, which often takes more than ten years. This article is usually about something else: a newly designed chip made on a process that is already in volume production. The process has been proven. The question is whether this one design comes out steadily on this line.

So the time scale is very different. One trip through the fab takes about three months on average, as the industry often puts it (see How long does a wafer take to make?). Even if the first version is right, going from the first wafer start to volume production often takes many months; each revision means another trip through the fab, so it takes longer still.

How does a new chip go from pilot run to volume production, step by step?

Step 1: The customer tapes out, and the photomasks are made

• What happens: The company that designed the chip (the fab's customer) hands over the final design files. This is the tape-out. Then a full set of photomasks is made (see What is tape-out?).
• Who takes over: It passes from the designers to the fab. Whoever runs the NPI first draws up the schedule: when the masks arrive, when the wafers start, when samples are due.
• Why: Every stage waits on the one before. Without a plan, the tools and people can't fit the work in when it arrives, and every stage waits longer.

Step 2: The first pilot run

• What happens: The first lot of wafers gets its wafer start and moves through the steps one by one. A new product's earliest lots are its pilot run (see What is a wafer start?). On a process already in volume production, the first lot often runs straight in the production fab; when a new step is needed, it often goes to the R&D line first (see How is an R&D line different from a production fab?).
• Who takes over: The people at each step in the fab. In a production fab, pilot lots often share the same tools with production lots. Whoever runs the NPI checks every day where these lots are and whether any is on hold.
• Why: Making a few first shows whether this design and this line get along. A pilot run has few wafers, so if one lot goes wrong, a big piece of the data is gone.

Step 3: Results go back to design and process

• What happens: When the wafers are done, WAT comes first, to see whether the process drifted. Then wafer sort tests every chip, the results become a wafer map, and the yield is worked out (see What is WAT?). Packaged chips also go on boards and run as if in real use: cold, hot, and at higher and lower voltages.
• Who takes over: Problems that look like the process go to the fab's engineers; problems that look like the design go to the designers.
• Why: The same kind of failure can come from the design or from the process. Send it to the wrong people and both sides waste their time.

Step 4: A revision, or tuned conditions

• What happens: If the problem is in the design, the design is changed, the affected masks are remade and another lot starts. This is a revision. If the problem is in the process, that step is fixed, or its conditions are tuned within the spec.
• Who takes over: The designers or the fab's engineers. Whoever runs the NPI redoes the schedule and tells each leg how late it now is.
• Why: Each revision costs several more months, so it pays to think through everything to fix at once. A process already in volume production is used by other products too, so it can't be changed casually for one chip (see What is change control in a chip fab?).

Step 5: The reliability tests

• What happens: Some of the chips run for a long time at high temperature and raised voltage, to see whether any fail early. A common one is HTOL: 1,000 hours, about six weeks (see What is reliability testing?).
• Who takes over: The reliability lab and QA. When the customer can't wait, a risk start comes up: starting wafers in the production fab before the tests have finished.
• Why: A chip that works when it's new may not still work after years of use. Earlier test data from the same process can sometimes be looked at too.

Step 6: The customer approves samples

• What happens: The customer (or its own customer, the company that builds products with this chip) puts samples into a product and tries them. Only once they're approved does the volume order come in.
• Who takes over: The customer. Whoever runs the NPI carries the customer's questions back to the fab, and the answers back out.
• Why: Whether a chip really works only shows once it's in the real product. Industries such as cars also have a public standard process: a full set of documents goes to the customer for approval before volume parts can ship.

Step 7: The conditions are fixed and handed to volume production

• What happens: Every step's tools and conditions are written into the POR, the version that volume production follows, and it's locked. Then a review goes through the list item by item: yield, reliability, documents, training, the customer's approval. When the handover is from the R&D line to the production fab, this is the transfer review (see What does an R&D integration engineer do with a new chip's first lot?).
• Who takes over: The production fab's Manufacturing, Process, Equipment and Production control, who from that day make it and schedule it by the POR.
• Why: The people taking over may never have been part of the pilot run. They can only make the same chip from what was written down; whatever the list missed becomes the production fab's problem.

Step 8: In volume production, yield climbs further

• What happens: Wafer starts grow lot by lot, and yield climbs stage by stage: find the most common kind of failure, find its cause, fix it, then move on to the next (see Why do bigger chips have lower yield?).
• Who takes over: The production fab's engineers, and the product engineers who look after this chip's data (see Is a PE a process engineer or a product engineer?).
• Why: Experiments to find causes take up tool time and compete with shipments for capacity. Papers on production ramp-up point out that the earlier something is learned, the more it is worth.

What does the person running an NPI do all day?

Job titles differ between companies: NPI engineer, new product introduction engineer or project manager, or it may be part of the job of an R&D integration, process integration or product engineer. Many companies cut the path into several gates; at each one, a meeting checks that everything due has arrived before the work moves on. The daily work comes down to four things:

1. Set the schedule: when each stage starts, when it's due, and who is waiting on whom.
2. Follow every lot: a pilot run has only a few lots, so where each one is and what it measured both need to be known.
3. Run meetings and get each problem to the right person: every problem needs an owner and a date for an answer; the ones nobody clearly owns are the ones most likely to be dropped.
4. Write down every decision: why something was changed this way, and who agreed. When someone asks months later, it can be found.

They don't need to be able to do every step, but they need to understand what each step is telling them. For how to tell a customer clearly what is going on, read What does the customer liaison at a test-and-assembly house do?; for who schedules the lots once volume starts, and what departments a fab has, see What does production control do in a chip fab? and What departments does a chip fab have?.

In fab24hr

The main story, The life of a chip, is the introduction of one fictional hearing-aid chip, "Glowmere", which keeps changing hands along the way. The prologue is the tape-out: Glowmere's design is made into a set of photomasks. Its process has one extra, new etch step, and the first lot runs on the R&D line first. Chapter 1, "How to Split 12", is the day before wafer start, and you are the R&D integration engineer. In chapter 3, "The shape of a failure", you read the WAT and three wafer maps. In chapter 4, "Two Weeks Early", you go through the eight-item checklist the night before the transfer review; one item is the customer sign-off, and the customer also wants the chips two weeks early. In chapter 5, "The Last Box", an R&D process engineer has to fill in the last box of the POR. After that, Glowmere goes on to the production fab, the test-and-assembly house, and finally into someone's hands.

To play just one of these people, tap "Clock in", choose "A day in the fab", then on the "Your role today" card tap "R&D line" and swipe to the R&D integration engineer. The companies, chip, people, lot numbers and schedule in the game are fictional; what new product introduction is and who takes over at each step come from public, general sources, and the actual split of work differs between companies. For other jargon, see Chip fab jargon in plain English.