It’s a prediction that Jack Weast, Principal Engineer and Chief Architect at Intel, has heard many times before: the age of Moore’s Law, which says that chip processing power will double every year, is coming to an end. The observation has more or less held true since the invention of the integrated circuit in 1971, when the gap between transistors on a chip was 10,000 nanometres. On today’s chips, the gap is 14 nanometres, ergo you can squeeze on many more transistors.
But could it really be that the ‘Law’ has run its course? Not really, says Weast. Whilst it’s true that in 2015 the company announced that 10 nanometre technology products will not arrive in 2016 as planned, product and process optimisation continues unabated. “Moore’s Law will continue,” says Weast, “but its cadence may be just a bit slower than it has been in the past. Put differently, it’s just adjusting for the complexities of the small scale and highly complex process nodes our industry now works with.” Progress in the chip industry still moves at an incredible rate. Intel even has a joke about it – if cars had developed at the same rate, they’d be regularly achieving fuel economy of two million miles to the gallon.
If cars had developed at the same rate, they’d be regularly achieving fuel economy of two million miles to the gallon
Tongue in cheek though it is, it’s certainly true that the automotive industry moves far more slowly – where chip power doubles every two years, all-new models from OEMs only arrive once every five to seven years. Does this present any disconnect for Intel?
“It’s fun,” laughs Weast. “Our current automotive roadmap was established around five years ago, and we already have parts in millions of cars out on the road today. It’s been a fascinating learning experience. But we also see opportunities to help automotive companies accelerate their design lifecycles – we’re no stranger to the five-year cycle, when looking at 10 nanometre chips for example, but what we’d like to do is help the automotive industry figure out how to accelerate their timeframe, and bring it down to four, perhaps even three years.”
New architectures
The magic word here is convergence. Simplified chip architectures, fewer electronic control units (ECUs) for features like advanced driver assistance systems (ADAS), reduced attack surfaces and common microarchitecture from R&D through to production – all these things, says Weast, could have a significant impact in reducing time to market, and accelerate the emergence of highly-automated and autonomous cars: “We’re very interested in partnering with OEMs on this. The world is changing, and it’s going to be increasingly difficult for a five-year design lifecycle to produce competitive products that consumers really want.”
In addition, Intel will look to provide OEMs with the option of not having to lock down their final design architecture for software until much later in the design process. The selection of an operating system, or the applications and capabilities of a product; these are things that no longer have to be set in stone. As Weast puts it, they can be completed, and designed later, and this could mean a far quicker move from R&D to production.
A system needs to be able to recover or continue operating if an ECU goes down. And that’s going to be an interesting challenge
Most OEMs have already hailed the arrival of automated driving, but few have nailed their colours to the wall when it comes to fully autonomous vehicles. By contrast, Intel is unequivocal – the day of the self-driving car is coming, and with it, a wealth of challenges and opportunities for the industry. The emergence of 5G, for example, could radically change the connected car’s capabilities and usages. This will require more computing power in datacentres, says Weast, to support the hundreds of millions of connected cars.
This is good news for Intel. If there’s computation involved, says Weast, Intel wants to be able to offer the best solution on the market – and the connected, self-driving car will have to perform levels of computation previously unheard of in the automotive industry. Once drivers aren’t driving, what they’ll do inside the car will change. Entertainment and increased work productivity are two areas where Weast predicts considerable demand, and as far as he’s concerned, the industry can’t even imagine just how much computing this will require. As such, it’s not just about autonomous driving: “it’s the entire spectrum of how the automobile and the automotive market is going to change, and deliver incredible new experiences that we can’t imagine today.”
The day of the self-driving car is coming, and with it, a wealth of challenges and opportunities for the industry
Fully autonomous driving could arrive earlier than 2030, he suggests, but with one caveat: “When people make predictions about autonomous cars, there’s an assumption we’re talking about all the time, on any road, under any conditions. This is an extremely high bar.” The reality, he says, will be more constrained. Moving into the next decade, Weast predicts deployment of fully autonomous vehicles in certain situations – for example, a 20-block downtown area designated as off-limits to human drivers.
Unbreakable
Security and safety will present the biggest challenges, and Weast predicts a move from reliability-based architectures to what he terms ‘availability-based architectures’. In a reliability model, a single chip is used, and efforts centre around reducing the failure rate of that part to as low a level as possible. Availability-based models assume that at some point, a mechanism will fail. As such, the mechanism should be built in a way that, when it does fail, it can continue operating.
“You see this in aerospace, where planes may have multiple redundant systems,” explains Weast. “Think about ADAS systems – when it fails, there’s a human take-over mechanism. If your adaptive cruise control ECU dies, your car could ask you take over. But if there’s no steering wheel or brake pedal, how do you design this availability? A system needs to be able to recover or continue operating if an ECU goes down. And that’s going to be an interesting challenge.” Weast compares it to building a ‘check engine light’ for the autonomous car – something that indicates a failure, but doesn’t stop the car from driving.
There’s another design challenge for OEMs – how can a product be kept competitive over its lifetime? “Almost every other product we have, like our phones or even TVs, can receive silent, forced software updates. The notion of the car you drive off the lot being the same car you’re driving five years later is going to change, partly because of the emphasis on software-driven services and capabilities.”
It’s going to be increasingly difficult for a five-year design lifecycle to produce competitive products
Once again, this raises the question of design and architecture. OEMs can no longer afford to design a car that meets initial specifications, with no additional headroom. Not only would this extra headroom be needed to introduce new features for customers, but it would be needed to counteract emerging security threats.
Convergence remains central, he concludes. “For us, it’s about reimagining the way we do things in the automotive industry,” he says. “Our theme around convergence says, ‘hey – we see pain points, we see struggles’. The consolidation of ECUs, convergence on a common microarchitecture, or convergence on safety and security architectures, these things can result in the simplification of an already very complex product. And where things are simpler, they can be done more easily, more cost-effectively, and more quickly.”
Moore’s Law may well be starting to slow a little, but if Intel and competitors get it right, the chip industry could hold the key to accelerating and transforming automotive design and production.
This article appeared in the Q2 2016 issue of Automotive Megatrends Magazine. Follow this link to download the full issue.
