Auto Chip Test Issues Grow

Semiconductors used in cars have higher quality and reliability requirements than most chips, but they have the same cost and time-to-market pressures.


By Jeff Dorsch & Ed Sperling

Semiconductor suppliers are flocking to the automotive chip market to gain share in fitting out the connected car and the autonomous vehicle. But before those chips are sold to automotive manufacturers and Tier 1 suppliers, they must be tested and certified to meet stringent industry standards.

This is no ordinary testing, though. Assisted and autonomous vehicles are extremely complex systems with a mix of advanced digital logic, various types of processors and accelerators, sophisticated and commodity sensors, and high-speed networks. Nearly all of this technology is bounded by an evolving and growing set of regulations, backed up by little or no history to gauge what works, what doesn’t, and why.

Put in perspective, this is a mix of new and established technology. But much of it is being used for new applications in a safety-critical market, where until several years ago electronics were viewed as part of the user infotainment experience rather than an essential component of a vehicle’s functionality. In addition, there is much more electronic content than in the past. IC Insights forecasts auto chip sales will enjoy a compound annual growth rate of 13.4% from 2016 to 2021, increasing slightly more than Internet of Things devices. The market research estimates automotive IC sales increased 22% in 2017 to about $28 billion, after growing 11% during 2016. It sees automotive IC sales rising 16% this year to $32.4 billion.

IHS Markit, meanwhile, predicts that more than 33 million autonomous vehicles will be sold around the world in 2040. “The first autonomous vehicle volumes — beyond retrofit test vehicles — will arrive in 2019 through driverless mobility services,” reported Egil Juliussen, director of automotive technology research at IHS Markit. “Volumes will surpass 51,000 units in 2021, when personally owned autonomous cars reach individual buyers for the first time, and IHS Markit forecasts estimate nearly 1 million units will be sold in 2025 across shared fleets and individually owned cars.”

Autonomous driving and mobility services, such as Uber, Lyft, and other ride-hailing services, will be key to early adoption and deployment of advanced automotive electronics technology, according to the market research firm.

“Diversity of choice in personal mobility and autonomous driving technologies are both evolving more quickly than ever, but their convergence will have the greatest impact,” said Jeremy Carlson, principal automotive analyst at IHS Markit. “Autonomous mobility services can deliver newfound personal freedom to the young, old, disabled, and others without reliable transportation for everyday needs, but the benefits don’t have to stop there. Fleet operators in big cities, who better understand the lower operational costs of battery electric vehicles, are more likely to employ them to drive higher amounts of vehicle and passenger miles traveled.”

Most parties agree that auto chips present a tremendous opportunity in the near future, and for decades to come, as the vision of self-driving cars and a dramatic reduction in auto accidents and fatalities is realized. And this extends to the tools used to create and test these chips.

Risto Puhakka, president of VLSI Research, estimates the current market for automatic test equipment dealing with automotive semiconductors is worth $120 million a year. There are some blurred lines in auto chip testing, however, because a microcontroller-specific test system could be testing MCUs for a variety of applications, including automotive.

“Automotive reliability requirements are so much higher,” Puhakka said. “So as a result, they end up doing a lot more testing for the chips. And that makes it nice and attractive for test suppliers.”

Fig. 1: The transition to autonomous driving. Source: Renesas

Challenges ahead
That also presents several big challenges for chipmakers. Because these chips are being used for safety-critical applications, they have to be much more thoroughly tested than chips that end up in a server rack or in a smart phone. This includes components specifically developed for automotive applications, such as camera sensors, LiDAR and radar. But it also includes components such as microcontrollers, analog chips, and power management ICs.

“Whenever you deal with automotive, traceability of requirements is higher,” said Tim Kogel, application engineering manager and solutions architect at Synopsys. “Functionality needs to be better defined, and testing and models need to be kept in the loop. This requires much more holistic system-level design.”

That, in turn, requires a lot more time and effort, but car companies are insistent that costs need to be kept down. The general consensus is that all of the electronics in a car should cost as little as $2,000 to $5,000 for lower-end autonomous vehicles.

The semiconductor industry has been able to meet these kinds of cost reductions in the past by amortizing NRE, manufacturing and packaging of a single design with multiple derivatives across billions of units. That model doesn’t work for automotive electronics, though. There are about 90 million cars sold every year, compared with about 2 billion smartphones. On top of that, people are not likely to be buying or leasing a new car every year, but many consumers are buying the latest smartphone model of the year. So while while the amount and value of electronics in vehicles is constantly on the rise, global sales of automotive vehicles themselves generally grow at single-digit rates.

To make matters worse, the amount of testing required (as well as chip-package-board design and hardware/software verification) is significantly higher because these chips need to last longer and meet stringent minimum standards for performance in a safety-critical market. The only way to reduce costs at that scale is to whittle them down across the supply chain. And on the test side, one of the solutions that has been winning favor is system-level test.

System-level test is already being used at the most advanced nodes because of the sheer number of components that need to be tested in each chip.

“The lower the node count, the greater the number of transistors,” said Anil Bhalla, senior manager at Astronics Test Systems. “This is where system-level test gives you some advantages. With built-in self-test (BiST), you may have 99.5% coverage, but you’re still not testing millions of transistors. And the reality is more like 85% coverage. And if you guard-band, which is how many companies have dealt with this problem in the past, you’re eating into your headroom. At 10nm, 33% of headroom is eaten up by guardbanding at 100 millivolts.”

Astronics isn’t alone in seeing this trend. “As engineers, we guard-band around design,” said George Zafiropoulos, vice president of solutions marketing at National Instruments. “But if you guard-band everything, you stack up inefficiencies. If you can decrease guard-banding with certainty of reliability and performance, that’s a huge value.”

System-level test can help significantly here because it tests components in context. “You can compensate for tool equipment variation and process variation,” said Bhalla. “Many of these parts are expensive, and you don’t want to throw good parts away. This allows you to test a device in conjunction with another device. So with ‘this constraint’ maybe it interacts with ‘these products.’ With ATE, you’re getting patterns. With system-level test, you’re combining a device with the final product. That can tell you if a device is good. Or if it’s tentatively not good, you can go back and re-test to see whether the issue is tool variability or temperature variability or metrology variability. And you can look for patterns in different lots.”

This intersects with the automotive world in multiple ways. Cars are a complex system of systems. Many of those systems are interdependent, and some of the components in these systems will be quite advanced. One of the strategies of automotive companies is to develop chips at the latest process nodes because they want to maximize the lifespan of a design. This is particularly true for some of the infotainment chips and the AI logic, where the latest nodes can provide performance or power improvements and help to delay obsolescence.

“You’ll see architectures in the future where you’ll have what’s almost a supercomputer in a car,” said Wilhelm Radermacher, executive advisor at Advantest. “It is being connected with a camera or imaging to central systems, where it can learn together with other cars what is going on in the environment. This is a very exciting time for the technology, and it’s a very rich opportunity. We will see a lot more different kinds of test. We will see system-type tests built into systems and chips. But that’s going to be connected to an external test subsystem, where you go from an early bring-up in hardware all the way to manufacturing based on a set of common architectures.”

Testing in context
To add efficiency into this process involves a deeper understanding of the test data, though, and the best way to do that is to systematically mine the data for errors and aberrations. That’s easier said than done in many cases, however.

“With automotive test, the testing of components is much tighter,” said Michael Schuldenfrei, CTO at Optimal+. “Car companies are requiring 100% burn-in on all parts. There is some pushback, because it’s very expensive. But the bigger problem is that we’re not seeing people using data for quality the way it should be used. This is all being done at the component level, and companies are not sharing data across the supply chain.”

That makes it harder to understand the reliability of parts in context of a system, and it makes it difficult to understand where the real problems are and how test resources should be used for optimal results.

“We’re dealing with a no-trouble-found syndrome,” said Schuldenfrei. “But you don’t know that for sure until you understand the interaction of components.”

NI’s Zafiropoulos agrees: “The state of the art right now is just acquiring data. The next phase involves what does big data allow you to do. Can you find something useful that you were not looking for? The next wave is whether software more intelligently can point out trends and correlations.”

And this is one of the big shifts underway in the test world. Context matters, and it matters not just for the hardware. Looking 5 to 10 years out, when self-driving cars are expected to become more widespread around the world, software security will become more crucial to automotive electronics, more than device security, said VLSI’s Puhakka. “The problem is more at the system level. I’m sure the technology will be there, way above the chip level.”

All of this is hard enough to resolve with a single device, such as a phone or a computer, but automotive technology and standards are in a state of almost constant change. There are continuing advancements in advanced driver-assistance systems and automated driving, and the entire ecosystem is in flux as companies figure out which pieces they will develop themselves, which will be outsourced, and how quality can be maintained or improved throughout an evolving supply chain.

“ADAS and secure autonomous driving are resulting in really just an explosion of electronics,” said Steve Pateras, product marketing director for automotive semiconductor test solutions at Mentor, a Siemens Business. “Autonomous and ADAS functions are also resulting in very strong growth in the need for better quality and reliability in those electronics. Automotive historically has been designed using trailing-edge technology — small MCUs, using robust processes, so the quality and reliability was often not difficult to achieve or maintain. With autonomous vehicles, we’re now moving towards very large, complex devices that are needed for all aspects of autonomous and ADAS, from a large computational point of view, from networking. We’ve gone from the trailing edge to essentially leading edge in the automotive space.”

Auto chip testing is now more oriented toward what Pateras calls in-system test – what other vendors usually term system-level test. In-system test offers faster testing with its limited scope of testing, providing better data compression and more efficient test. Auto chip testing also benefits from running logic built-in self-test (BiST), with pseudo-random patterns, scan chains, and go/no-go testing.

“At the moment, the focus remains on device test, but this is changing,” said Scott West, system-level test product marketing manager at Advantest. “Automotive radar technology is here now, and while it’s currently being seen primarily in premium-brand vehicles, the goal is to bring down the unit cost so that it becomes standard equipment throughout the automotive industry. To do this, a number of challenges must be addressed, including solving of the complexities associated with testing.”

He pointed to a number of factors that need to be considered, from noise to context. But he added that a key issue, which often is overlooked, is a shortage of test engineering knowledge. This is particularly evident with 5G communication, which is expected to be used for analyzing autonomous vehicle data that does not require real-time response.

“This is perhaps the most critical factor of all,” said West. “The number of engineers working in millimeter technology is relatively small, and companies wanting to enter the space can’t simply materialize engineers versed in radar technology to help them with product development – particularly when the primary emphasis in most engineering programs is digital technology, rather than analog/RF. This means that talent is expensive, which can put a real damper on what companies are able to do. We need competent engineers to be trained that are strongly motivated and passionate about millimeter-wave.”

As with many other forms of electronics these days, automotive chip testing is becoming more complex, difficult, and time-consuming. The quality and reliability of auto chips can’t be compromised because human life depends on the fail-safe redundancy of advanced automotive electronics. But the complexity, the economics of testing in a safety critical environment, and the newness of this technology all add up to some difficult challenges.

There is general agreement that test will take on an increasingly important role in assisted and autonomous driving. But how that test will be done, by whom, and whether companies are willing to pay for it are not so clear at the moment. What is evident is that test is changing along with all automotive technology, but what the final requirements will look like may take years to fully comprehend.

Related Stories
Changes Ahead For Test
Testing microprocessors, microcontrollers, application processors, and other system-on-a-chip devices grows more complicated.
Tech Talk: System In Package
Why advanced packaging is so important for autonomous driving and the semiconductor industry.
Toward System-Level Test
What’s working in test, what isn’t, and where the holes are.