Building an automotive chip on a commercial wafer process and then qualifying it to AEC-Q100 is sometimes the only option, but on mature nodes it is usually a false saving. AEC-Q100 shows that sampled parts survived a set of stress tests; it does not make the silicon as robust as a process built for automotive. What you save on wafers you tend to spend again on screening, yield loss, test time and field reliability risk.
In the semiconductor world, there is a conversation that happens in boardrooms every single week. It usually goes something like this:
Management: "The 'Auto Grade' wafer process from the foundry costs 20% more. Why don't we just build this chip on the standard commercial process? We'll test the heck out of it, put an AEC-Q100 sticker on the box and save the money. Apple does it, why can't we?"
It sounds like a smart business call. And sometimes it is. But more often than not, especially with mature technology, it is a trap.
Key takeaways
- Auto process, AEC-Q qualification and functional safety are three different things; none replaces the others.
- At the leading edge, commercial processes are often the only choice, so heavy screening is the price of entry.
- On mature nodes where an automotive process exists, choosing commercial is a shortcut, not a necessity.
- The hidden costs are yield loss from screening, longer test time and higher wear-out risk in the field.
- If you choose commercial, budget for it openly: defect targets, test development and reliability work.
The big confusion: ingredients versus the recipe
I have seen a lot of confusion recently about the difference between a chip that is qualified for auto and a chip that was built for auto. Getting it wrong does not just cost money; it risks reliability. First, we need to stop mixing up three very different things: auto process, AEC-Q and functional safety. Think of it like running a high-end restaurant.
| Concept | The analogy | The engineering reality |
|---|---|---|
| Auto process (the ingredients) | Buying organic, pesticide-free vegetables. | The foundation. This is the foundry flow (for example an automotive variant of TSMC's 16nm family). It uses stricter design rules and process controls, such as more conservative metal widths and oxide margins, to prevent defects physically. |
| AEC-Q (the taste test) | Checking whether the meal spoils after three days. | The qualification. AEC-Q100 is the Automotive Electronics Council's stress-test qualification for packaged integrated circuits. A commercial chip can pass it, but passing proves the sampled lots survived the tests, not how robust the silicon is. |
| Functional safety (the fire extinguisher) | Installing sprinklers in the kitchen. | The architecture. ISO 26262 measures such as ECC, redundancy and safety monitors. They detect and manage failures when they happen, but they do not stop the silicon wearing out. |
What AEC-Q100 actually covers
AEC-Q100 defines stress tests (for example temperature cycling, high-temperature operating life, ESD and latch-up) run on samples from several lots, at an ambient temperature grade chosen for the application. The common grades are:
| Grade | Ambient operating range | Typical use |
|---|---|---|
| Grade 0 | -40 to +150 °C | Engine compartment, close to heat sources |
| Grade 1 | -40 to +125 °C | Most under-hood electronics |
| Grade 2 | -40 to +105 °C | Hot cabin locations |
| Grade 3 | -40 to +85 °C | Most cabin electronics |
Qualification is a gate the design and process must pass once. It says little about the defect rate of every part you ship afterwards; that comes from the process, the test programme and production controls.
The commercial process loophole
Here is where the argument gets tricky. Management often points out that big players use commercial processes for automotive chips all the time. They are not wrong. But there is a catch.
If you are building an AI processor at the leading edge for a self-driving car, you often have to use the commercial process, because the automotive variant of that node arrives years later, if at all. TSMC, for example, announced an automotive 5nm variant (N5A) in 2020 and an early automotive 3nm offering (N3AE) in 2023, each well after the commercial node. Companies at the leading edge therefore take commercial wafers and spend heavily on screening: testing every die hard to find the good ones.
But if we are talking about a GPIO expander or a standard MCU on a mature node (such as 40nm or 65nm), the automotive process usually does exist. Choosing the commercial process there is not a necessity; it is a shortcut.
The hidden costs of being cheap
So what actually happens when you try to save money by using a commercial process for an automotive product? You might save on the wafer, but you pay for it in engineering effort.
1. The needle in a haystack problem (defect rates)
Automotive customers today ask for defect rates in the low parts per billion, with targets such as fewer than 50 defective parts per billion (0.05 DPPM) for many components and zero-defect programmes as the stated goal. That is a very demanding standard. A commercial process typically has a higher natural defect density and less margin. To get commercial wafers down to that level, you have to screen hard, throwing away chips that are probably fine just to be safe. Your yield drops, and your "cheap" chip is suddenly not so cheap.
Illustrative example: suppose the auto process costs 20% more per wafer and gives 90% yield after test. If the commercial process needs extra screening that cuts yield to 78%, the cost per good die is about the same (1.00 / 0.78 is about 1.28, against 1.20 / 0.90, about 1.33) before you add the extra test time and engineering. These numbers are illustrative only; real figures depend on the product and the foundry.
2. The testing burden
Because the manufacturing process is not filtering out as many weak parts for you, your test engineers have to do it. They write complex, aggressive tests such as high-voltage stress testing (HVST), and apply statistical outlier screens such as part average testing (PAT, described in AEC-Q001) and statistical bin and yield analysis (AEC-Q002). In the chip world, test time is money, and every extra second of test on every part adds up over millions of units.
3. The walking wounded
Commercial design rules allow tighter geometry: thinner wires and tighter spacing. A chip built this way might pass every test today. But put it in a car, heating and cooling for 10 to 15 years, and the margins matter. Wear-out mechanisms such as electromigration (metal atoms moved by high current density, accelerated by temperature) and dielectric breakdown progress faster with less margin, while thermal cycling and vibration stress the package and its joints. You might ship a chip that works on day 1 but fails on day 1,000.
How to decide
- Check whether an automotive process exists for the node and foundry. If it does, the default should be to use it.
- Confirm the customer's defect target and mission profile (temperature grade, lifetime, hours of operation).
- Estimate the full cost: wafer price, yield after screening, test time per part, extra characterisation and reliability work.
- Check functional safety needs under ISO 26262 separately; process choice does not remove them.
- Write the decision down, with the risks accepted, so it is a deliberate business call and not a quiet shortcut.
Common mistakes
- Treating AEC-Q100 qualification as proof of automotive-grade quality.
- Assuming functional safety features make up for a weaker process.
- Comparing wafer prices without comparing cost per good, shipped die.
- Planning test development late, after the process decision has fixed the screening burden.
The bottom line
If we decide to go with a commercial process, we need to be honest about what that means. It is not just a paperwork change.
- It means we need to accept higher underlying defect rates.
- It means we need to budget for substantial test development.
- It means we are accepting a reliability risk that competitors using a genuine automotive process do not have.
It is a valid business call to make, but let us not pretend it is the same quality product. In the world of automotive silicon, you usually get exactly what you pay for.
FAQ
Is an AEC-Q100 qualified chip the same as an automotive-grade chip?
Not necessarily. AEC-Q100 shows the part passed a defined set of stress tests. An automotive-grade product usually also means an automotive process, production controls such as PAT, and change management agreed with the customer.
Why do leading-edge automotive chips use commercial processes?
Because automotive variants of the newest nodes arrive later than the commercial versions. Companies that need the newest node accept the commercial process and compensate with screening and design margin.
Does functional safety replace a robust process?
No. ISO 26262 safety mechanisms detect and handle faults; they do not reduce how often the silicon develops them.
Your comments will be moderated before it appears here.