Automakers are right in demanding quality, but wrong in their approach.
There is a huge problem brewing in the automotive world. Automakers are demanding quality parts, but they’re using methods and strategies developed in the steel age when suppliers were metal benders, not developers of advanced electronics.
Automakers are correct in that the quality of electronics is poor. A 2018 report by J.D. Power showed that overall car reliability is improving year over year. But it also showed that audio/communication/entertainment/navigation (AECN) remains the biggest source of complaints for the third year in a row. Meanwhile—and of much more interest to the semiconductor industry—driver-assistance issues increasing 20% per year. That number is still low, because there aren’t too many of these systems on the road compared to the total number of cars, but those numbers likely will rise significantly as these technologies show up in more vehicles.
Automakers have responded to these complaints by strengthening their failure in time (FIT) requirements for suppliers. Rather than parts lasting 7 to 10 years, they now demand zero defects for 18 years. Alternatively, they are demanding zero incidents per part, which allows suppliers some wiggle room because it means that redundancy or possibly a failover could reduce those incidents to zero.
But their solution to these complaints is basically a knee-jerk reaction to one of the most fundamental shifts in the history of automobiles. Moreover, it shows just how large the disconnect is between automakers and their tier 1 and tier 2 suppliers, and modern electronics. Automakers are doing exactly what has proven successful over the past 70 years, which is to squeeze their suppliers rather than rethink their business model.
Prior to Tesla, all of the big improvements in automotive engineering were weighted much more heavily on the mechanical side than electrical—overhead camshafts, anti-lock brakes, fuel injection, airbags. The chips used in these systems are primitive compared to the electronics in a robotic vacuum cleaner, and they don’t wear out quickly because many of them consist of basic circuitry developed at 0.25 micron processes or larger. They are heavily insulated to protect them from engine noise and vibration, and the lines/spaces are so large that heat isn’t a major problem.
Adding advanced electronics, including 7nm AI systems for driver assistance and autonomous driving, requires a whole different approach. For one thing, these systems are so new that no one has ever actually observed how they behave over time. And while simulation and accelerated testing are helpful, try simulating a 7nm AI chip used intermittently on a rutted dirt road in the Mojave Desert, or one used pre-monsoon season in Bangalore when humidity is at its highest.
There are solutions to these problems, but they all eat into profits. Redundancy costs more than failover. Squeezing suppliers costs less than adding in-circuit monitoring and continual testing of more components. And relying solely on suppliers that comply with stringent but outdated automotive regulations at a given price point, rather than trying to solve these problems with different architectural and packaging approaches, is like driving blindfolded on a twisty mountain road. You don’t know what problems you will encounter.
The basic issue is that carmakers are still thinking like carmakers, when they need to be thinking like high-tech companies. There are always solutions to quality problems, but they’re not always simple in electronics. And demanding that advanced electronics need to have defectivity rates in the parts per trillion range shows just how much these carmakers still need to learn.
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
I suspect that it will be easier to teach a techie how to build cars than to teach a car maker how to be a techie.
A slightly different issue.
Car makers hope that a decrease in battery cell prices will enable them to make EVs work (cost wise) but that’s not quite true. Cell price declines will enable an increase in range and, more importantly, charging rates (miles added per unit of time). For as long as charging rates are limiting the SAM, range will keep increasing.
The cost savings need to come from everywhere else. They can increase efficiency, work on aerodynamics and weight to reduce the spending on cells but they got to work on everything else and find ways to cut costs and this includes electronics. Integrating, rethinking the entire architecture and maybe even making their own silicon – VW spends on R&D 10 times more than Tesla and if Tesla can invest in this area, the big guys can too.
They need to focus on value not costs and they can not count on cells getting cheaper. And they got no time to hesitate, EVs are going to grow much faster than they expect. There is an increase in a awareness once sales reach 2-4% share and well designed EVs are better cars than ICE. Just watch growth rates in markets that are well ahead of the curve – global sales for new energy vehicles were just above 2% last year but China was at over 4% and this year it will be at 8-11%. Analysts do not not understand EVs, they don’t know that the powertrain enables a better vehicle so their projections ate very low but if the auto industry doesn’t radically adjust their plans, some folks will gain massive share, especially China brands and Tesla. There might be severe EV shortages too. The 5-25% of output being electric in 2025, needs to be adjusted to 50%. And those that don’t rethink the entire vehicle, are not gonna retain their share.
A great article Ed. Insightful!!
You kinda hit the nail on the head here, Ed! This has been brewing for over a decade. And you didn’t even address cyber-security, the other area that is and will cause them enormous consternation if they continue to demand bleeding edge technology.
some are actually doing and they start by changing the requirement/model of the FIT calculation from a decade old model to a much recent one. the concept of overdesigning for safety cannot stand too long before it fails to meet the requirement for Level 5
This is “funny sad” because more performance and scaling downward will never increase reliability. I’ve worked in semiconductor device-level reliability for 35 years.
We are already at shockingly low expected lifetimes of 5-10 years for deep nanometer CMOS and historically, that has ONLY been monotonic with scaled dimension. This is for commercial temperature ranges and at-data-sheet-spec’ed Vdds with no ripple or excursions.
In 1960s vintage ICs, the lifetime was 10,000 years or 1 failure in 10,000 years. More performance WILL result in shorter lifespans.
We are already very close to crossing over on consumer product lifespans – perhaps the next 5 nm node or the 3 nm. Chips will become like motor oil – a consumable that must be replaced every 3 months if you want to have smooth and continuous operation. You can take the risk but you might end up with your car dead in the middle of the Mojave desert.
3 nm is probably the last node because the next is at very fundamental limits of electronics let alone CMOS. 1-2 nm is where every fundamental limit gets triggered. Atomic and molecular crystal unit cells are in this range – you can NOT have “half a lattice or half an atom”. We are also at bit-energy limits when combined with thermal noise limits.
There will NOT be better reliability even if we manage to create “atomic electronics” or electronics below the atom level (what would that even be?) – it will continue to decrease because we simply can’t control what atoms do under the stresses created by normal operation. We can only lower Vdd and logic levels so much before we enter thermal noise.
Moore’s Law’s end IS imminent. Now is the time for completely rethinking how things are designed and architected. This goes from HW to SW and the tools that create them. Turning the crank of size to get “more” is over now. High density package level integration is probably the short-term answer. But even that won’t last long as a solution.
Engineering is about to get harder than we’ve ever experienced and demand more of all of us that we’ve ever thought possible. Many will not be up to the challenge.