More Data, More Problems In Automotive

The race toward increasing levels of autonomy is being hampered by competitive concerns over sharing data across the automotive supply chain.

Pushing past the initial ADAS levels into full autonomy is expected to take more than a decade, but the infrastructure for those systems, and making sure all assisted and autonomous vehicles work with other vehicles, is under development today. Still, that has turned into a big challenge in its own right, in part because everything is new and in part because the data is not consistent across the supply chain.

Carmakers are intent on getting all of this right, of course. The negative publicity from recalls and the actual cost of fixing problems can be enormous. But as cars increasingly are electrified and ever-more complex electronics are used to replace mechanical functions, OEMs and Tier 1s are being confronted by massive amounts of unstructured data, as well as data silos between the supply chain companies that make it more difficult to utilize that data effectively.

The auto industry has both long supply chains and decentralized design and manufacturing ecosystems, which don’t mesh easily with functional safety and increasing electronic content. Qualification times are long, and prompt response to electronics quality and reliability issues is essential. This is made worse by the push toward centralized logic at the most advanced process nodes, and the lack of history about how devices developed at those nodes will perform over time under harsh conditions.

Rapid response to quality/reliability issues requires two things — a collaborative business framework across the companies involved and the right data from the design and manufacturing supply chain. Volkswagen and other automakers have begun tackling the first issue with the GIINA/PASRASC (Global Industry Initiative New Automotive, Platform for Automotive Semiconductor Requirements Along the Supply Chain) framework, which enables working across business boundaries on common issues. You can expect to see more partnerships and collaboration efforts happening.

Fig. 1: Volkswagen’s Global Industry Initiative New Automotive. Source: International Test Conference presentation by Andreas Aal, semiconductor strategy and reliability at Volkswagen.

“The entrance of companies such as Panasonic has shifted some of the dynamics, where we’re seeing more formal partnerships driving investment (such as GM-LGChem), as opposed to individual deals for automotive programs,” said Jeff Phillips, head of automotive marketing at National Instruments.

With a more collaborative environment this also pushes to get data into the hands of the engineers to improve yield, quality and reliability.

What’s enabled with today’s analytics
Manufacturing test data analysis has been the grist between the designer’s imagination and silicon performance. Whether inside an integrated device manufacturer’s system or the fabless/foundry ecosystem, engineers have been leveraging silicon manufacturing data in the front end and back end to identify trouble spots.

But this has been relatively easy compared to what’s ahead. Level 1 to 3 ADAS modules typically use mature semiconductor process technology.

“The Level 1 to 2 applications of sensor technology have become fairly well-known,” said NI’s Phillips. “That means volume is driving most of the concerns and the factors that impact volume — price point, functionality, leverage.”

In other words, companies need to keep their eye on the processes.

“The issues that they face are around tool control, incoming material control and excursion detection,” said Dennis Ciplickas, vice president of advanced solutions at PDF Solutions. “So the trick is to make sure that it stays in control due to extrinsic factors as opposed to intrinsic marginalities.”

With this older technologies, package failures and PCB solder ball connectivity contribute the largest pareto bar to a car’s electronic systems’ quality and relatability issues.

To meet the needs of ADAS levels 4 and 5, automakers will use more advanced process nodes, which come with their associated manufacturing silicon challenges. In addition, to support these hands-off capabilities, the system engineering requires considerations beyond just the silicon product. Designed as system modules, the hardware includes electro-mechanical, optics and silicon, along with system software that needs to be integrated into these units.

Analytics extending up and down stream
For advanced assisted driving features, the computational needs for AI/ML require 10/7/5nm process nodes for logic ICs. Those typically are complex architectures, which also need design validation for security and functional safety requirements. In addition, they require firmware updates to run the final integrated electronic module, as well as to stay current with shifting regulations and technology across the industry.

Level 3, which is aimed at keeping a vehicle in its lane, requires cameras with optics, and LiDAR and/or radar. All of that technology needs to work seamlessly with other technology in a vehicle, as well. To fully support the automotive supply chain, data analytics companies have been extending their reach both downstream and upstream, both with in-circuit technology, which enables more in-depth analysis capabilities, and with analytic capabilities for electronic boards and complete modules.

Prior to ramping silicon, engineers need to validate that the silicon works correct in the end system. “On-chip monitors can check the CPU usage/load and can be checking the quality measure of the system,” said Gadge Panesar, CTO of UltraSoC. “These ECUs will only be loaded up to 50% so you can see, ‘How does my code run, how is this behaving?’ The same thing applies to updating the firmware.”

The emphasis on silicon quality has been growing over the past couple years. Defects are now measured in parts per billion rather than parts per million, and the leading German automakers mandate that parts last 18 years with zero defects.

“With advanced assisted driving applications today and autonomous applications in the future, customer concerns relate to production quality at time zero, as well as the longevity of that quality, meaning reliability spanning the system’s life,” said Raanan Gewirtzman, chief business officer at proteanTecs. “In order to achieve DPPB targets, the industry must adopt a new approach of looking inside the electronics. On-chip monitoring is the first piece of the puzzle, allowing continuous measurement of previously-inaccessible information. Machine learning and analytics complete the picture to make sense out of all this new data at the push of a button, so that users can actually take corrective action at every stage. This enables chip and system manufacturers to qualify new products with a higher level of certainty, screen products for quality issues at a finer resolution and weed out any defective parts, ‘walking wounded’ or latent defects. Only by guaranteeing the highest possible chip coverage and a parametric approach to testing, will the industry be able to achieve near-zero defect rates. The great thing is that while the systems are deployed in the field, the electronics continue to be monitored, so any sign of degradation over time is immediately detected and alerted on. This gives automakers the chance to perform predictive maintenance in their fleets without reaching the point of unexpected failures.”

How that will pan out isn’t entirely clear. Advanced node chips never have been used for that length of time in any market. Almost all chips developed at leading-edge geometries historically have been used in controlled environments with tightly controlled temperature and minimal vibration.

“That’s what’s so great about on-chip monitoring in the field,” Gewirtzman said. “It factors in not only the design and manufacturing process, but also the use conditions and the interaction between the three. Key mechanisms associated with advanced ICs will degrade with time while operating in their respective application. We can catch that degradation as a precursor before it reaches the tipping point of failure.”

Others agree. “Now, the stability of the technology and the inherent marginalities in it will be different, which leads to different fail modes, unknown fail modes, and needing to understand what those are in order to have the right countermeasures at the design level and system level to insure safe operation,” said PDF’s Ciplickas.

Automakers keen to uncover these issues will be driving the extension of data analytics requirements due to the heightened safety concerns for ADAS levels 4 and 5.

“The manufacturing design is highly complex,” said Uzi Baruch, vice president and general manager of the automotive business at OptimalPlus. “Now when we get the safety feature, we may have to scrap something we normally could use because there’s less tolerance to high-variability manufacturing.”

This is where using data that in the past has been siloed becomes so critical, and it is causing a flurry of activity across the industry. PDF Solutions, for example, has added capabilities in test and package assembly data analytics. OptimalPlus, meanwhile, recently launched data analytics for auto camera modules. And partnerships by Japan’s Denso with UltraSoC and OptimalPlus signal the rising value of partnerships across this space.

On-chip monitoring
On-chip circuitry eases both silicon test manufacturing and silicon debug by providing observations to internal VLSI operations, internal signals, and within-die process variability. Over the past 10 years design engineers have more systematically architected in these features.

“ICs for automotive have increased their design complexity,” said Tamar Naishlos, director of marketing at proteanTecs. “We are now dealing with nano-scale production processes and advanced packaging for what is one of the most safety-critical uses. More power and performance are needed to drive the logic-intensive applications, and that puts a strain on reaching the climbing quality and reliability requirements. So now you have an equation of tradeoffs. Chip designers are challenged to find the balance, but there’s always a cost. We need to break that equation and allow for higher levels of performance, at restrained power consumption and die area, while simultaneously improving quality and reliability. This can only be achieved by gaining visibility into every IC during design, manufacturing and while in the field.”

This isn’t just about engine functionality anymore, either. “Consider the microcontroller for your steering wheel,” said Gewirtzman. “Today’s IC manufacturing analytics are no longer enough. Car OEMs and brand owners must adopt a deep data approach to ensure in-field failure prediction.”

On-chip Agents/monitors could reduce the field risk by targeting types of on-die data collected.

“We are going there as well with on-die silicon monitors,” said Andrzej Strojwas, chief technologist at PDF. “Collecting the data from sensors is very important in automotive because of the various sources of electrical and physical stress. This ties into a full set of sensors.”

It also adds another dimension to the analytics, which is time. Consider the temperature measured by a sensor is “z.” Another data point would be how long it operated at that temperature. This kind of data over a car’s lifetime can boost the automotive supply chain’s understanding of abnormal operations and ultimately help predict maintenance needs.

But what data is essential, and what is useless, isn’t understood yet. “Using monitors for managing products in the field, I think it’s still in the infancy in terms of what the automakers will use these IP like UltraSoC has,” said Panesar.

Whose data?
That’s only part of the problem. Not all data is available to all companies, despite the fact that some of it might be very useful for analyzing potential issues.

Data in the factories, in the silicon, and data collected in a car are a data analytics company’s dream. But today, the easiest way to get access to the data in a car is literally to remove a module and ship it to whoever needs that data, said National Instruments’ Phillips. As a result, National Instruments is working with automotive customers on data management challenges. Discussions on automotive systems occur regarding the volume of data, tradeoffs between edge and central computing, and managing transmission.

Semiconductor manufacturers are well-versed in big data management, but that isn’t necessarily true for the rest of the supply chain, particularly those developing large electronic modules.

“The semiconductor portion has much more advanced analytic solutions than what we see farther down the supply chain,” said OptimalPlus’ Baruch. “Especially in power electronics, they are doing this in a siloed approach. Sharing data from final test in semiconductors with test in PCBs will begin to break down the silos.”

PDF’s Ciplickas agrees. “Connectivity between sites enables a richer analysis that can drive, process or test and measurement. Let’s assume that they can share the data, then they can improve.”

With on-chip monitors this improvement extends into the field. “Today these are simple things, like watching the health of the car, monitoring the ECU, i.e. not overheating of the car,” said Panesar. “But as more complex cars some along (sic ADAS driving), you need to understand the behavior of the system and look to the next level for safety issues when things wear out.”

Developing solutions for hands-off driving (ADAS level 4 and 5) requires advanced semiconductor technology, which further heightens the need for collaboration. Companies that break down silos between data sources will play an important role in reducing risk with autonomous vehicles.

Conclusion
To move from ADAS Levels 1, 2 and 3 up to Levels 4 and 5 will require a much better bridge between IC manufacturing data and the data generated by the electronic systems and modules in a vehicle. This is easier said than done. The pressures of safety and security in Level 4 and 5 vehicles, combined with electronics complexity, necessitates collaboration across businesses to share data to enable joint solving of quality and reliability issues.

Along the way, there are many other problems to solve. For example, automotive makers demand a 10 ppb quality level, but how that will happen at price points automakers demand isn’t clear. At ITC 2019, presentations from Infineon and OnSemi engineers shared their efforts to improve quality at wafer and final test for the automotive sector. This naturally requires more test vectors, more test conditions, more test methods, and, hence, more time spent in manufacturing. Yet feedback from vehicles usage in the field is needed to inform semiconductor companies on where to spend that test time.

None of this happens overnight. It took Volkswagen two years to prove that mechanical stresses deepened the impact of SRAM cell stability latent defects, according to a recent presentation by Andreas Aal, who is in charge of the company’s semiconductor strategy and reliability.

The trifecta of fast, cheap and good quality, as any engineer will share, is often impossible. With machine learning and a super linear growth in collected data, analytics capabilities may be up to this challenge. But with siloed data in the supply chain and business contracts limiting open communication, ADAS level 4 and 5 cars may not even get out of the driveway.

Related Stories
Automotive Knowledge Center
Top stories, special reports, white papers, videos, blogs and more on Automotive Electronics
Auto Industry Shifts Gears On Where Data Gets Processed
How to manage massive amounts of data in real time still isn’t clear, but it can’t be done in the cloud.
5 Major Shifts In Automotive
How new technology developments will change the trajectory of the automotive industry.
Planning For Failures In Automotive
Is it better to build expensive parts that are highly reliable, or redundant cheaper parts?
Safety Critical Design In Automotive
Finding faults at the chip and system level.
Automotive System Design
How to build and update chips in cars.