The Need For Traceability In Auto Chips

Someday your car will drive itself to a repair shop for a recall using a scheduling application that is both efficient and can prioritize which vehicles need to be fixed first. But that’s still a ways off. Proactive identification of issues is not yet available.

To be ready for that, today’s data analytics systems need to begin supporting targeted recalls, enabling predictive maintenance and identifying in-field quality issues sooner so that automakers can be ready for ADAS levels 4 and 5. In-field quality issues often stem from either new reliability failure modes or unmodeled mechanical stresses that impact a circuit performance. But without a deeper dive into the data, automakers and Tier 1 and 2 suppliers can’t respond appropriately.

Traceability, the ability to have a unique ID for an electronic device, provides the foundation for connecting the data currently siloed at the individual manufacturing steps. SEMI rolled out a new standard last year that enables the requirements that the auto industry will need to connect the data dots. By combining that with data sharing, automakers can be in the driver’s seat if they insist on traceability.

Fig. 1: Device traceability across the supply chain. Source: SEMI

Finding a needle in an electronic haystack
Today your car’s check engine light stays on if there’s an issue, requiring an unexpected trip to the repair shop. Ignoring that could result in damage to your car or to yourself. Today, you receive a postcard or an email informing you that your car needs a replacement part. With the specter of the Takata airbag safety issue lingering seven years, and the advent of ADAS driving levels 4 and 5, automakers will need to become more proactive on recalls and predictive maintenance.

Recalls often stem from reliability issues involving a single component in a complex system. In the case of Takata airbags, the properties of gas used in the airbag changed over the lifetime of the product. And in the case of modern automotive electronics, that recall could involve any number of nuanced issues and components.

The biggest problems today relate to package failures and board solder bumps. But given the intricate relationship between IC manufacturing processes and the mechanical and temperature stresses experienced in automobiles, the exact cause of failure still isn’t so easy to diagnose.

Consider the impact that nano-indentions (a mechanical stress) on CMOS device characteristics. Volkswagen observed a quality issue with three different IC vendors at the board level. In a study presented at the 2018 International Reliability and Physics Symposium, researchers demonstrated the change in transistor properties in the presence of nano-indentations using a 28nm CMOS SRAM cell. When the pressure was off, the transistor went back to normal. This is taking the impact of mechanical stresses down to a whole new level. The stresses have always existed, but now they are having an impact on current technologies.

With advanced semiconductor technologies, no one knows what new failure mechanisms await the automotive electronics supply chain.

“As vehicle computing requirements move into the newest technologies, silicon failures will become much higher than in the past,” said Dennis Ciplickas, vice president of advanced solutions at PDF Solutions. “Life safety is one thing, but product quality is another thing. You might have things that break all the time, so product quality might be a significant shorter-term driver of connecting data across the manufacturing steps.”

To meet the stringent expectations of quality in the automotive space, feedback from further down the manufacturing process will be essential. But you can’t test for what you don’t know about.

Furthermore, with in-field data from the car you can look at the semiconductor and electronics supply chain data for a better signal of potential issues in the field and determine measures to mitigate risks, However, to take a car’s data and tie it back to its manufacturing genealogy, to understand why something fails, to discern if there’s a larger pattern in place. All of that requires traceability.

Tracing Silicon Devices Manufacturing Origin
None of this has been lost on companies that supply chips or other electronics components into safety-critical applications.

“The big issue for all of our customers is traceability,” said John O’Donnell, CEO of yieldHUB. “You need to be able to search for a die in the database and see where it was in wafer sort, where it was in terms of modules, and see how that die performs versus the rest of the population. Everybody is looking for that, and right now they aren’t always sure how they’re going to do it. But they do realize they have to do it. We have the data in modules and final test and wafer start consistency. Our customers have access to the module data. We promise them that within seconds they can trace that data from the module all the way down to wafer sort. There could be a common failure in the field that wasn’t apparent because you didn’t have enough test data. But now you can correlate across all of these dice if you know where these modules they ended up in the field. That will be possible very soon. So now if you have a die that behaves in such a manner, and all the other die behave that way, you should be able to tell which modules they each went into.”

The SEMI Traceability Standards committee has been working on a framework for traceability across the supply chain. In SEMI T23 (approved in 2019), a unique Device ID is required from wafer to multi-chip devices that can be propagated through the supply chain. In the works is the ability to add security and counterfeit protection by sharing the data via a block-chain technology. So, the providence of the electronic device is known.

Semiconductor IDMs and foundries have long used chip IDs to track defect metrology and to provide feedback mechanisms to wafer fabrication. They use chip IDs to feedforward wafer sort test data to assembly steps and to package level test. Package-level IDs that move out to the customer can be tracked further by the customer. But package IDs are not required to be linked to the chip IDs, because there has been no requirement to have these IDs readable by anyone.

If compliant to SEMI T23, then a manufacturer’s Device IDs can be shared across the supply chain.

This is of value to end users of silicon that have high-reliability requirements, such as network backbones, medical implants and automakers. If every supplier has a Device ID and it’s readable at subsequent manufacturing steps and in the end-use system, that facilitates analytics of electronic devices. Data between suppliers then can be joined for deeper dives into the data.

SEMI T23 enables this fusion of data by specifying minimum requirements. To be traceable, the standard specifies the device supplier must have a unique Device ID per device, and that ID must be externally readable. In addition, it requires the manufacturing and test data from the device supplier to be able to connect to the Device ID. If met by all suppliers touching the electronic device, the standard it makes possible to identify test coverage gaps, to support targeted recalls of suspect parts, and to share data across the supply chain.

The standard is agnostic as to Device ID implementation. SEMI maintains a number of specific device IDs (T7, T9, T19, E142) and IEEE specifies Electronic Chip ID 1149.1-2013. There is no requirement to provide the data with the Device ID. That could be handled by individual NDA’s between suppliers or from a third party.

So, what does all this mean for the auto maker?

Consider a part that fails in the automaker’s system. At their return center they can read all device IDs. An MCM (multi-chip module) shipped to the PCB manufacturer has following device IDs:

MCM_Test_Device_ID
MCM_Assembly_Device_ID

1. DieA_Sort_Device_ID
     DieA_Fab_DeviceID
2. DieB_Sort_Device_ID
     DieB_Fab_DeviceID
3. DieC_Sort_Device_ID
     DieC_Fab_DeviceID
4. DieD_Sort_Device_ID
     DieD_Fab_DeviceID

This capability enables the automaker to state, “I have these failures,” to the previous manufacturing step, and state, “Here is the Device ID lists associated with each failure.” That allows them to draw the manufacturing genealogy of the failure per device. Once the commonality is determined, that can be fed forward to identify devices in the supply chain manufacturing process, thereby preventing a defective part to get into a car — or identifying parts already in cars that need to be recalled.

Multiple data analytic vendors state the value of connecting data across the supply chain that enable identification of field quality and reliability issues.

“There’s a time correlation with Level 1-2 electronics and new manufacturing/packaging technology,” said PDF’s Ciplickas. “As packaging has gotten more complex and the volume of chips have gone up such that the probability of having an issue in that technology has gone up. Then traceability and the ability to join the data, wafer and package test data and the lot and assembly level history of failure, material batches and so on becomes important. This data is needed to understand the problem, the root cause and how pervasive that problem is in order to do the right recall.”

Still, this is just one piece of the overall puzzle. Multi-chip modules and multi-die packages add yet another level of traceability requirements because not all of the chips may come from the same vendor.

“Automotive in particular has a lot of thinking to do about how to how to engage electronics because they are going to be going up in value and content in cars will go up like crazy over the next few years, especially with ADAS and electrification,” said Rich Rice, senior vice president of business development at ASE. “So, they’re going to have to think about what’s the best way. We’re starting to see decisions made not necessarily on the cheapest thing that’s available that connects all the dots and connects all the circuits, but something that is a little bit more reliable, that can reduce failure and return rates.”

Adding traceability into those types of approaches is likewise increasingly valuable. “In-use data has the car guys excited because ultimately it can mean no more recalls,” said Doug Elder, vice president and general manager of the semiconductor business unit at OptimalPlus. “This is all about predictive models. So, if a camera dies on a certain date, what happened to make that camera die? Was there a change in the process steps? And this goes beyond just the part. Was it a solder machine or a tester that drifted? What we’re able to do is add up the tolerances for error. And if you can get the automakers to use a common platform that goes all the way back through the supply chain, then you can start tying all of this stuff together.”

Traceability enforcement
As an industry that tightly manages its supply chain, automakers should welcome a traceability standard that enables management of goods and thus the data associated with them. While the prospects of driverless cars might be viewed as long-term factor, just managing the quality of parts provides enough impetus today. Semiconductor devices are used throughout the car not just in ADAS electronic modules, for example fuel injection and brake systems. These too can impact safety.

“Products qualified for automotive need to meet audit requirements for the total quality management,” said Ron DiGiuseppe, Synopsys automotive IP marketing segment director. “Even though there’s an expectation of 10 defects per billion over the 16-year life of the car, the capabilities for determining source of failure need to be in place.”

More issues to resolve
Software adds another layer of complexity to all of this, particularly involving traceability. The amount of software in a car is growing significantly. Not all of that software is exposed, however, and not all of it can be fixed, and the almost incessant changes to algorithms makes it very difficult to trace when changes cause problems in hardware functionality. So far, this is outside the scope of much of the hardware industry.

“If you have software running and have an error in the algorithm, that might not match another algorithm,” said Rahul Singhal, AI architect at Mentor, a Siemens Business. “At this point, the only way around that is to add redundancy and a system-level approach. The OEM that manufactured the system that the car manufacturer buys has to provide proper guidelines for that. If you have a software error, it may depend on the individual application.”

On top of that, even though standards are developed, it doesn’t mean that manufacturers will implement them. As they prepare for ADAS levels 4 and 5, automakers already have a need for traceability to improve quality. Each defective part discovered further down the manufacturing process has an estimated 10X increase in cost. With a history of tight control of their supply chain and more electronics suppliers entering that chain, the automakers can demand that everyone in their supply chain meet the SEMI T23 standard. Automakers have started to join SEMI which may signal that demand is coming.

What’s next?
While data from all sources theoretically can be used to provide nuanced insights into electronics parts behavior, enabling such capabilities comes down to the cost of enabling traceability as well as trust between suppliers to share the data.

At the very least, there is much more discussion about traceability than in the past. “Medical devices have long had the motivation to demand traceability, on automotive side its cost related, reputation for sure, but cost related primarily,” Ciplickas said. “At what point is it cost effective to implement all of this.”

Others agree. “Reliability and traceability are essential in AI running in the cloud, blockchain, automotive, 5G,” O’Donnell said. “You need very fast traceability aligned with easy extraction to run machine learning software, along with the ability to factor in whatever environmental variables you may have.”

At this point, the sharing of data between suppliers in the supply chain is not the norm. Even SEMI T23 states that data with the Device ID is not required, and recognizes that sharing data may require non-disclosure agreements between individual suppliers. This creates a bureaucracy in sharing data from the wafer fab up to the car in the field, while also limiting the current data analytics eco-system.

Still, when data is aggregated across all sources everyone in the automotive supply chain has the potential to learn, make improvements, reduce costs and produce higher quality products. Traceability is necessary but not sufficient to make it all happen.

Related Stories
More Data, More Problems In Automotive
Data is becoming more useful and timely, but not everyone has access to it.
Automakers Changing Tactics On Reliability
Focus shifts to more data-centric approaches as chip content increases.
Gaps Emerge In Test And Analytics
Comparison data is required for understanding drift and AI changes, but that’s not so simple.