What you’ll learn:
- How to take advantage of a changing automotive ecosystem.
- The steps one smaller fabless ASIC vendor took to design and certify a customized automotive ASIC.
- The key takeaways from going through this design process.
The move to electric vehicles, and more recently advanced driver-assistance systesm (ADAS), has turned what was a very established, static industry on its head. As has been written about (EnSilica included), this provides opportunities for smaller electronic system vendors outside of the “tier-1” club to gain inroads into the automotive supply chain.
Such a transition, coupled with the speed of innovation required by the automotive manufacturers, means that many of the traditional tier-1 suppliers will often lack the necessary skills to create solutions for these new technologies. It’s creating a gap for new and innovative companies to enter.
Earlier this year, EnSilica announced a new custom automotive ASIC (see figure), which had completed the final PPAP and Part Submission Warrant declaration (PSW), entered mass production. It went through the design and certification in less than 24 months.
In this case, our customer was the tier-1 supplier, working with them to fill the gap in their skillset and support their OEM customer. For sensitivity and confidentiality reasons, we will not name either.
The part in question was a complex mixed-signal ASIC that uses a bipolar-CMOS-DMOS (BCD) process with high-voltage transistors, combining extensive monitoring and fault-detection circuits with duplicate redundancy on key functions. It was created to be used in EVs, although the technology is equally suited to hybrids and hydrogen fuel-cell-powered vehicles.
In addition to the well-understood AEC-Q qualification, which is undertaken over three production lots along with extensive reliability testing, this design was designated safety-critical and had to follow a study of safe launch planning and the use of part average testing in manufacturing. The safety aspect also led us into the design methodologies and audits as prescribed in ISO 26262, all of which we had to learn from scratch.
Here we summarize some of the key elements of the project that can help you take advantage of the opportunity.
The Challenge Faced
Before we begin, a bit of background. The primary issue for this particular ASIC was time. The tier-1 supplier’s timescale was very aggressive by automotive standards, which was exacerbated by the need for high analog complexity (including high voltages and extreme precision). This narrowed the field for ASIC design agencies that had the required skills.
A secondary challenge was cost. The OEM exclusively makes high-end luxury vehicles, which means the volumes of ASICs produced are smaller than if they were used in a more mass-market brand, such as Ford or BMW, and with that comes reduced economies of scale, meaning costs need to be well managed.
How We Selected the Supply Chain
EnSilica is a relatively small fabless ASIC vendor. Therefore, we had to first come up with a credible supply chain that would demonstrate to the customer that we could manage them and get the appropriate support to a high-volume automotive component.
Most of the large wafer foundries offer an automotive option on their mainstream processes. This (typically) means a smaller list of equipment is used, with tightened in-line process controls to reduce the likelihood of defects and, perhaps more importantly for automotive, to reduce batch-to-batch variation. Assembly vendors also operate similar automotive variants for the packaging schemes.
This left the decision of where to do the test development and the qualification. Test and qualification of this kind of device requires the collection of a lot of data over a large number of tests. It’s done over three wafer production lots and tens of thousands of devices. All of it needed to be undertaken in parallel with our own, and the customer’s, design verification.
Unlike processes and packages where we’re essentially buying a well-proven product, the entirety of the qualification and characterization cycle is all unique to this project. After some evaluation, we chose a supplier that could support all of these activities in one location and could demonstrate a track record in doing this kind of project with small fabless companies.
What We Had to Learn from a Device Perspective
As an ASIC vendor, we routinely develop digital and mixed-signal ASICs for a range of customers across safety and non-safety critical applications, including automotive. However, this vehicle manufacturer was new to EnSilica and required us to quickly understand the system-level functionality and the company’s concerns.
Understanding the fine detail of complex systems takes time. Thus, we had no option but to perform this system-level learning. The task was undertaken by our modeling group in parallel with our silicon group running the ASIC component-level development.
Expertise Can Embed Yourself Higher Up the Supply Chain, Too
EnSilica has end-to-end expertise covering digital IC design, analog IC design, and systems design. We used this to engage with both the tier-1 and the vehicle manufacturer in as many areas as possible - not just on the ASIC definition and development, but on the electronic control unit’s (ECU’s)—which would be made by the tier-1 supplier—definition and development as well.
What We Had to Learn About Designing for Safety (26262 References)
The end application was safety-critical, and the ASIC was required to support the ASIL-D safety level. In addition to the core functionality, a considerable amount of additional safety-related functionality needed to be defined and implemented for the ASIC.
Much of this fault-detection functionality could only be defined and tested at the ECU system-level. Furthermore, as the core component within the ECU in this safety-critical application, the development process needed to be fully documented with regular FMEAs (failure mode and effects analyses) during the development and the fault metrics regularly tracked with FMEDAs (failure modes, effects, and diagnostic analyses).
The Journey Through to First Silicon
For the OEM, many aspects of the automotive ECU could only be validated after silicon was available. Therefore, from the project’s start, there was a race to get to first silicon. Development time for the mixed-signal ASIC was eight months with a three-month silicon manufacturing cycle, meaning the ASIC was soldered down in the ECU after just 11 months from project start.
Feature Creep Must be Avoided
One of the many challenges with this project was that the detailed definition of the ASIC—the specification—needed to be written in parallel with the ASIC development work. In essence, the moment the ink dried on a section of the ASIC specification, then the design work for that part of the ASIC commenced.
Taking this approach means that feature creep happens easily. And avoiding it is only possible via system-level modeling undertaken by a small, highly focused team.
Pre-qual Should be Undertaken
It was always intended, given the tight timescale, to create a specification and initial design that could undergo a partial redesign to get the device fully centered. But this also means that, unless you can be sure these changes would be minor and not affect form fit or function, you’re not allowed to start your automotive qualifications, which require final production tooling.
To solve this problem, without wasting time between first and final silicon, we undertook a cut down version of an AEC-Q qualification that looked at the riskier sections: the burn-in, early-life failure rate (ELFR), package integrity discharge (ESD), and latch up (LU). To do this required significant cost to design and procure the necessary tooling. We would consider it money well spent, though. In fact, in this case, a small number of potential issues with the product’s design and manufacture were uncovered that needed improvement before going to final silicon.
Second Silicon and Qualification
First samples of the final version of the silicon were delivered to EnSilica at the beginning of April 2020, just as the COVID-19 pandemic had forced most of the world into lockdown. This kicked off a large number of parallel activities including:
- Design evaluation inside EnSilica.
- Design evaluation by our customer.
- Debugging ATE program to get a release candidate to test all of the samples required for AEC-Q100 qualification.
- Planning and execution of the full schedule of tests over the three production lots as required.
- Full PVT characterization and correlation to end applications.
- PPAP and PSW delivery.
Summary
We learned much on this journey at many levels. The three most important takeaway lessons were:
- Ensuring a frequent and detailed technical information exchange, and a review that encompassed the entire ECU design, not just the ASI.
- Having a tight and detailed internal design and specification review process, particularly as we were aware that feature creep was inevitable given the timescales.
- The pre-qual that we performed on first silicon, although seeming costly at the time, was vital in getting to final silicon faster.
This article couldn’t really be completed without discussing the impact of COVID on the latter stages of qualification and test development. Without a doubt, not being able to travel to the test and qualification partner for more than a year has impacted the efficiency of the project and led to some basic errors and misunderstandings that have in turn caused delay or extra work.
Yet despite this, we were still able to complete the final PPAP and PSW declaration and entered mass production in less than 24 months from starting the project. It demonstrates that there are opportunities for smaller electronic system vendors to gain inroads into the automotive supply chain as the move to electric and ADAS vehicles marches on, turning what was a very established industry on its head.