Who’s Paying For Auto Chip Test?

Testing of automotive chips is becoming more difficult and time-consuming, and the problem is only going to get worse.

There is more to this than simply developing new test equipment or devising a better design for test flow. There are multiple issues at play here, and some of them are at odds with the others.

First, no one has experience using advanced-node chips in extreme environments. The whole idea behind using chips developed at established nodes is that tolerances for heat and noise are well established, and these chips can be tested quickly in their entirety using existing equipment. But that was before carmakers latched onto the idea that chips developed at 10/7nm would be faster and more power-efficient for certain tasks, such as running the AI systems inside autonomous vehicles, and that they could be extended across multiple generations of future vehicles.

Advanced-node chips have never had to function in high heat or extreme cold. Leave your smartphone in the sun inside a hot car and it will tell you that it’s too hot to use. Run a server rack beyond the recommended temperature and it will scale back performance. But that’s not an option for chips used in cars, and every transistor in a 10/7nm chip needs to be fully tested at for those extremes because it will be used for a safety-critical application.

Even chips not directly used for driving functions are being tapped for safety failover. In ISO 26262, one of the requirements is that electronics don’t just die. They need to fail gracefully. But because complete redundancy is expensive, takes up space, and adds weight to the vehicle—particularly on the cabling side—one of the newer approaches is to utilize other electronics in the car. This is sort of like using your television to access the Internet if your computer dies. It doesn’t work as well as your computer, but it’s better than nothing. And because you already own a television, it doesn’t require any extra cost.

The problem is that now your television has to adhere to the same rigorous testing as your computer. And in a car, that means the infotainment system may require the same level of coverage as an engine control unit or the AI brain in an autonomous vehicle. So testing needs to be more robust, coverage needs to be guaranteed to a certain level, and all of this has to work in an extreme environment.

Second, testing needs to be much more specific than in the past. It’s not just about testing a wafer to see if the chips function properly. In complex electronic systems, chips need to be tested in context. There are numerous examples of mobile devices catching fire after a series of operations performed in a particular order. This was so bad that airlines banned certain devices. But imagine what could happen in a traffic jam if always-on finFET-based devices suddenly went into thermal runaway.

So testing not only needs to be more robust, it also needs to be more thorough. In the case of mobile phones, though, there are only a handful of major vendors, and each of them sells hundreds of millions of units. There are literally dozens of companies developing autonomous cars and trucks, and they don’t share information about what goes wrong because the race to Level 5 autonomy is so competitive. So they will likely be selling thousands or millions of units, not billions, and the systems they are developing will have to react to a constantly changing environment that has never been fully mapped or field-tested.

That leads to the third problem. Understanding where the pitfalls are in these systems and how to test for them will need to be both continuous and evolutionary, and that’s much more difficult with limited runs of systems—particularly if they are in a constant state of development. It’s likely that fully autonomous vehicles will not be rolled out en masse for at least a decade, and testing will have to evolve alongside of those changes.

All of this costs money and takes more time in the fab, the packaging house, and in the field. Chipmakers squeezed test budgets down to a few percentage points of the total development cost, but those economics won’t work in the automotive chip world. It will take more time, more equipment, and entirely new approaches to provide the level of confidence required that electronics will function as expected under harsh conditions for extended periods of time. The big question is who’s going to foot the bill for that, and so far that isn’t clear.