Auto Chip Reliability Opens Door To Other Industries

Auto chips are finding traction in aerospace, industrial, and even consumer applications.


Digital chips in the semiconductor industry evolve from each other. Ideas flow into each other over the years, with occasional big leaps in evolution. The term ‘evolution’ fits because one chip evolves to perfectly optimized for one industry niche.

But what happens when one industry’s chip becomes a useful for other industries because it is more cost-effective than what is being used in that industry? That’s exactly what’s happening with automotive chips.

Rigorous attention to reliability and supply chain accountability in automotive electronics hasn’t gone unnoticed. Chips developed for automotive applications are supposed to last 15 years or more under extreme stress from heat, cold and vibration.

“The qualification that you need to go through from an automotive point of view gives you a huge benefit in industrial applications and data centers,” said Jaime Broome, automotive business and product management at Imagination Technologies. “Industrial and data center are the two use cases. There is some overhead with automotive in those two use cases. And there’s more tolerance in the industrial and the data center world for more consumer-based parts. But given the choice, I would always obviously want to go down the route of more secure and tolerant silicon.”

Data centers and automotive have parallel needs. “The use of things like error correcting algorithms (ECC) is born and brought out of data centers, from running computers in data centers and seeing bits flip,” said Broome. “That has a complete overlap synergy with automotive, with the whole Toyota unintended acceleration issue with bit flipping cosmic rays.”

Broome talks about the Toyota unintended acceleration case as a turning point in automotive. “It tells you why their motivation in the automotive industry now is about making sure that you’ve tested and that you’ve verified what they’re doing to a high level. That story is what makes all of our engineers and myself run cold.”

It’s also something that has caught the attention of NASA. The aerospace industry knows about rad-hard and ECC, too. Cosmic rays and solar wind are flipping more bits in space than on Earth. Budgets for space flight, however, are usually large, and designs need custom redundancy you can’t buy off the shelf.  For that reason, military avionics and space craft are different than industrial and automotive. But the space industry is expanding, and how it finds its chips may change.

The self-driving car EDA company AImotive has started working with the satellite industry to craft AI in satellites for real-time image processing. AImotive makes neural network hardware accelerator IP for self-driving cars. “Automotive is all about safety, reliability, and robustness over extended temperature range,” said Tony King-Smith, executive vice president of marketing of AImotive. “That’s the reason why the collaboration started. There’s a common set of interests there. AIWare was designed from the ground up to be robust. It was designed from the start to be effective in the presence of errors, and to be able to recover very efficiently. All those characteristics made it very suitable for a spacecraft-type application. And this synergy between the automotive industry and space is very interesting.”

Nextchip is using AIware in a chip, and will get their first samples of an automotive-grade, extended-temperature-range device with an AIware engine by the end of the year.  “If you could take that device then, to what extent would that be sufficient to put in a payload and what sort of useful work could it do? This comes back to space industry always looking for ways to be able to access technologies where just doing custom chips for spacecraft is a pretty expensive way to go,” said King-Smith. “To repurpose an automotive chip that’s already designed for extended temperature range, designed to be reliable and robust, fully documented, these are all tick boxes for the space industry. That’s why we’ve got this collaboration going here. It’s using what we’re already doing anyway in automotive and re-purposing it for space.”

Space actually may be harsher on hardware, but there aren’t as many regulations — at least not yet. “Automotive or the semi consumer space — and automotive becomes consumer at some point — requires regulation,”  said Imagination’s Broome. “And while that exists in avionics, it does not exist in space as much. When something becomes regulated, people get very scared. And there’s a high cost involved, so people test against the regulated environment quite heavily.”

Automotive chips have had a big impact in other markets, as well.

“Now that big digital chips are at the heart of automotive, it raises the bar on security and safety and reliability, which was never the case for a computer or an iPhone,” said Marc Swinnen, director of product marketing for the Semiconductor Division of Ansys. “Now, suddenly, it has raised the profile of those things. Techniques like ISO 26262 are a step forward for the semiconductor industry. We never had to worry to that degree on reliability. And at the same time, the automotive industry is having to grapple with the realities of using large digital chips at the heart of their system, which also is something new for them. So it is a bit of learning in both directions.”

Yet, automotive electronics was not created in a vacuum. “If you look at what happened with Imagination’s journey in the IP market, especially with a company like Texas Instruments, we made our mark with Texas Instruments, with Nokia making mobile phone application processors,” said Broome. “And that was a chip designed to run a smart phone even before the world knew how to run a smart phone. It took a certain other company to come along to tell the world how that was done. But the basic framework of that chip was there, and that formed into automotive. Avionics and military are the forefathers of a lot of what was in the industrial and automotive industry and a lot of the safety standards. Things came out of that, especially with regard to military-spaceflight-era avionics.”

Broome said the same thing happened with Renesas. “The first time around, the first Imagination boom with the Dreamcast console formed into another automotive company, called Renesas. In our 15- to 20-year history, we have adopted automotive in our DNA without us even knowing it. Our two major customers have drip fed us this reliability. Now everybody’s talking about safety all of a sudden, but it’s a bigger picture — how quickly can your device detect a crash and come back up? We’ve been doing that for years.”

Automotive, which for years was considered an aging technology market, is once again pushing the boundaries of what’s possible. “Automotive has emerged as one of the growth engines of the market,” said Raanan Gewirtzman, CBO of proteanTecs. “Now, of all the different electronics, it has become relevant for everyone. It basically incorporates a wide range of systems that can be migrated from other applications since it is becoming a mini datacenter on its own.”

The flexibility of a chip for multiple industries, however, is happening by design now. Groq, the AI startup working on its Tensor Sensing Processor (TSP), is developing chips to be flexible for different use cases. The first ASIC implementation of the TSP architecture yielded a computational density of more than 1 teraop/second/mm² of silicon. The chip is a 25 x 29mm on 14nm. The company claims its architecture is simple enough to scale up for supercomputers and data centers, while scaling down for automotive. “Our architecture has that kind of range and characteristics that makes it very interesting for autonomous vehicles,” said Bill Leszinske, vice president of product and marketing at Groq. “We do have a lot of interest in that space. One of the things is the chip is deterministic.”

Reliability in transition
Behind the scenes, the automotive supply chain is in the midst of a major transition. The AI systems that are used in ADAS, and which ultimately may be used in autonomous vehicles, are still under development. For the first time, this includes chips developed at leading-edge process nodes and advanced packaging.

“Of the products that are going to be targeted to automotive, not all of them are leading-edge nodes. The trailing edge is still important because these devices are well established and reliable for many applications,” said Chet Lenox, senior director of industry and customer collaboration at KLA. “But for advanced features on the leading edge, we must change the way that we approach process control in order to meet automotive requirements. Generally, we have approached leading-edge nodes with a consumer-oriented mindset where a part-per-million failure rate is acceptable. Simply applying that same methodology in Automotive is not going to achieve the part-per-billion failure rate that is required.”

The industry is well aware of these issues. “We’re developing techniques that allow us to not just utilize data in the fab for yield improvement, but also for screening for potential reliability failures in the field,” Lenox said. “Machine learning is a big part of this, as it enables effective analysis of the huge quantities of data being generated. In general, inspection and metrology are used primarily for process control in the fab. This is not only to control excursions, but also to figure out what the yield-relevant defect Pareto looks like, and then drive down the leading defects to improve the yield. Traditionally, the final arbitrator was sort, yield, and final package test. If that was green, you feel fairly confident you can ship the product. Advanced node devices contain billions of transistors, and electrical test will suffer from incomplete coverage, untestable gap areas, as well “latent defects” that are not detectable. A part can pass electrical test, and a defect mode can activate later in the car and cause a failure. The combination of un-activated latent reliability defects and test coverage gaps needs to be filled for automotive. In-line inspection is our best tool to screen these out.”

Security has a role in this, as well, particularly as devices become more heterogeneous. “There’s a drive to create more hardware accelerators, which are not peripheral to the design,” said Serge Leef, program manager in DARPA’s Microsystems Technology Office. “That grows the attack surface on the hardware side. A missile guidance may be implemented in software, but that’s not fast enough. So, this is an obvious candidate for hardware acceleration using dedicated hardware. That transition from software to hardware is a new shift.”

The challenge is that every piece of hardware is additional IP that needs to be tested for reliability and security. And in mil/aero, that’s complicated. It requires tracing every component in the supply chain. Automotive is following similar guidelines for roughly the same reasons.

“In automotive, they’re looking at traceability of IP,” said Simon Rance, head of marketing at ClioSoft. “Every artifact can be traced. If a reference is not referenced by anyone else anymore, you scrub it. You need to have enough traceability, and you need to be able to test everything. A lot of stuff gets left behind because it doesn’t meet these requirements.”

He’s not alone in seeing that. “Traceability is now essential in AI, blockchain, automotive and 5G,” said John O’Donnell, CEO of yieldHUB. “You need to be able to search for die exactly where they were on a wafer.”

That data also needs to be kept for at least as long as a device’s lifetime, as well, said O’Donnell. So in automotive, that means some data needs to be kept for up to 20 years or more, because failures in the field may show up a decade or more later.

“The higher the quality, the better the chances that your new product or your already shipping product, can deal with all the changes in production over the years because you’re going to keep producing that over X number of years and things change. They may have moved the fab from one country to another, you don’t know. It’s like a living creature — production. It’s not a fixed thing,” said proteanTecs Gewirtzman. “So, we provide a way to measure, to keep monitoring, keep seeing what’s in there and making sure the right level of requirements are met for whatever requirements are there, whether it’s 26262, ASIL A up to ASIL D. It’s now up to them and the industry that they’re serving to run Deep Data analytics to meet those requirements and do that with the lowest cost, the best efficiency.”

Digital automotive chips are appealing to other industries because they have measured, certified reliability, safety, and security built in, but that reliability is still a work in progress. How well automotive chips can live up to their certified reliability over the next 15 or 20 years requires traceability, testing improvements, analytics and continual infield monitoring. While the semiconductor, automotive, test, and analytics industries work this problem, other industries are watching and may benefit.

—Ed Sperling contributed to this report.


BillM says:

Not sure Automotive is helping mil-aero applications. For decades, mil-aero had very rigid criteria that stressed the operating conditions and also needed to have alpha particle protection (which I do not think automotive ever required) along with very long 15-30 year production life cycles and low volumes. The one area that this might help aerospace/military is increasing the volumes so the cost for mil-aero parts decrease.

Concerning the other areas: industrial and consumer: yes, this can help them.

Leave a Reply

(Note: This name will be displayed publicly)