Ethernet the way forward, and not exclusively in high-end applications
MADISON, Wis. — When they consider the cost of ADAS/AV (autonomous vehicles), many observers assume that the computing power required for AI processing is going to be the costliest element.
“Not so,” according to Alexander E. Tan, vice president and general manager of Automotive Ethernet Solutions at NXP. He predicts that in 2025 in-vehicle networking will cost more than computing, considering the huge volume of data generated by an AV loaded with sensors and how quickly that data must be distributed.
The automotive industry already uses a variety of in-vehicle connectivity technologies. They range from CAN, FlexRay and LIN to MOST and LVDS.
Many are custom-built interfaces, some tailored for specific automotive applications. They have proven robust and reliable. But once a new generation of auto tech, with AVs and ADAS, starts sending and routing megabits or gigabits of onboard data, no current automotive connectivity interface suffices.
That’s where Ethernet comes in.
System architects at car OEMs have already identified Ethernet as the key in-vehicle backbone network, NXP’s Tan explained. They do not look at the Ethernet as the high-speed, high-quality connectivity exclusive to high-end cars. Rather they look at it as a framework that scales down for high-volume vehicles, he added. “The car industry is absorbing Ethernet like a sponge.”
NXP Semiconductors announced Monday the acquisition of a high-speed automotive Ethernet IP company called OmniPHY based in San Jose. OmniPHY, with 100 employees, has a sizable development team in India.
The move will accelerate the rollout of NXP’s own PHY chips and switches based on1000BASE-T1. More important, the OmniPHY deal sets up NXP as the imminent driving force behind a plethora of new, very high-speed connectivity solutions currently brewing. They include Multi-Gig Automotive Ethernet standard, automotive MIPI and HDBase-T — many pitched by players eager to move into automotive turf.
NXP’s Tan was careful not to favor one high-speed connectivity initiative over others during the interview. But he said he strongly believes in the emerging IEEE Multi-Gig Automotive Ethernet PHY standard and MIPI. The MIPI Alliance, started by developing interface specifications for the mobile industry, is addressing point-to-point high-speed data interface specs in the automotive industry. Tan also confirmed that NXP is a member of HDBase-T.
Asked for a prognosis of these different activities, Tan said, as a leading automotive chip supplier, “Our responsibility is to offer a tool box of really good solutions” to the automotive industry.
From the frontline of the connectivity market, Tan has witnessed the automotive industry’s shift from a very conservative outlook, slow to adopt new technology, to an eagerness to change and accelerate cycle time.
We asked Tan what prompted system architects at car OEMs to take this 180-degree turn, how automotive Ethernet is different from those advanced high-speed Ethernet whose use is already proliferating in data centers, and what he sees lie ahead for the automotive industry.
Here’s an excerpt of that conversation.
Absorbing Ethernet like a sponge
EE Times: You say the automotive industry is “absorbing Ethernet like a sponge.” Please explain, including when the turnaround started.
Alexander Tan: As a precursor to the rollout of BroadR-Reach (2011), the auto industry was already beginning to get interested in Ethernet in 2008 to 2010. As with any new technology, though, carmakers were initially tentative. They picked a few models, built testers, sourced key parts and first built a telecom gateway system based on Ethernet.
The original idea was to use Ethernet to upload software onto on-board diagnostic systems. Carmakers realized if they did it via CAN bus it would have taken them two days.
A lot has changed since then. The emergence of new players such as Tesla and Uber in the auto industry prompted carmakers to think differently, learn to identify new technologies.
Until then, car OEMs depended on the purpose-built connectivity solutions to do tasks inside a vehicle. But car OEM architects quickly realized the need for a backbone network for a vehicle.
EE Times: Which factors really drove carmakers to embrace Ethernet?
Tan: There are many, but to name a few:
EE Times: How did an entertainment system become a part of the equation?
Tan: Car OEMs realized that consumers’ iPads or iPhones, for example, could quickly outdate in-vehicle entertainment systems. Their challenge was how not to make the interior of their car look so out of step with the technology already available today. Who would want to keep green LED in a dashboard? Consumers want higher resolution screens that can display better images. I wouldn’t say that the entertainment system did it, but it resulted in a path car OEMs have taken, by driving them to adopt a much faster pace of change.
EE Times: Ethernet has been broadly embraced by the computer industry, and high-speed, multi-gigabit Ethernet technologies are getting used by data centers (10/25/100 GbE) and enterprise infrastructure (2.5/5/10 GbE). What would it take to reuse those advanced connectivity technologies in automotive?
Tan: The IEEE 802.1 Standards Committee has already developed the Audio Video Bridging / Time Sensitive Networking (AVB/TSN) set of standards to help existing networks carry time-sensitive data. The move has helped Ethernet move into the industrial market, and subsequently open the door to automotive.
But to move Ethernet into the automotive environment, you must consider:
Things like the weight of cables, power consumption and EMI (electromagnetic inference) of Ethernet aren’t something you need to worry about in data centers. In a car, you don’t have the luxury of putting up with so much power consumed by servers, or cabling that is not flexible.
Have you seen what’s inside an autonomous vehicle?
EE Times: Yes, I have. Pretty much all the guts are inside a trunk.
Datacenter in a trunk
Tan: Right. If you open a trunk, what you see are: a couple of servers, switches/routers — maybe from Cisco — and a huge cable harness. This makes it literally a “data center in a trunk.”
EE Times: True.
Tan: Today, most of these AVs — with a data center in the trunk — are designed for proof of concept. The main concern of every carmaker testing AVs on the road is: “How can I get enough miles on improving my algorithms?”
EE Times: So, the cost of each testing vehicle is no object?
Tan: Right. That’s OK, as far as they remain testing vehicles. The real question, though, is how OEMs would be able to shrink the data center in a trunk in an autonomous vehicle — at a cost we can afford.
EE Times: How must does such testing cost?
Tan: The cost of electronics in the very first generation of a testing autonomous vehicle is said to cost twice as much as the body of a vehicle that housed autonomous functions. Unless the cost of electronics goes down, a company like Uber — planning to use AVs as a car service — would have to have its AV run for 3,000 to 4,000 miles just to recoup the cost of electronics inside.
The cost of electronics inside an AV should be an adder, or a fraction of the cost of a vehicle rather than a multiplier.
EE Times: So, what are car OEMs looking for?
Tan: Aside from specific physical-layer requirements, OEMs want to make sure Ethernet-based in-vehicle networking architecture offers scalability. While it should make OEMs’ development tasks reasonable, the same framework should be able to address a huge range of vehicle models — from Volkswagen to Audi.
EE Times: That said, isn’t the cost of AVs coming down?
Tan: Thus far, much of the cost reduction is mainly coming from second-generation sensors. As the volume of data increases, a network gets more expensive.
With the right algorithms, more processing and compression are expected to be applied to a growing volume of data collected by sensors. But we’re not there yet.
The volume of data still needs to be routed to the right locations inside a vehicle via flexible cables. By 2025, we predict that the network inside an autonomous vehicle will cost more than computing.
Tool box needed
EE Times: Of course, Ethernet isn’t the only game in town. Many players are looking for an opening to enter the in-vehicle network market today.
Tan: There will be a demand for a high-speed connectivity when moving a lot of data to one destination via one direction. MIPI, for example, is a lot like LVDS in terms of sending a lot of data in a single direction.
There also needs to be a high-speed interface to route data to multiple locations. Then, there will be a demand for bidirectional networks, sending and receiving data at the same time. Ethernet will be the backbone of a vehicle. I see a lot of technology innovations are happening today. New ideas are coming out from everywhere. This is a sea change. And I think it is a good thing that a lot of people want to get involved in [various industry groups].
As a chip vendor, I think we are responsible for developing a tool box of really good solutions.
EE Times: What’s ahead?
Tan: While developing all these new, very high-speed interfaces for ADAS/AVs, we see further challenges in security, safety and reliability. They still need to be addressed.
EE Times: When did you join NXP?
Tan: A year ago. I worked at National Semiconductors, Texas Instruments and most recently at Marvell where I worked on gigabit Ethernet efforts for automotive.
EE Times: So, you are an automotive connectivity guru?
Tan: “Guru” is a strong word, but I have been in the connectivity business for a long time. And undoubtedly the automotive market is getting hot.
When I was a kid, I grew up thinking that the future is in cybernetics. But look where we are now. Little did I know that future was going to be inside vehicles.
— Junko Yoshida, Global Co-Editor-In-Chief, AspenCore Media, Chief International Correspondent, EE Times