Automotive Ethernet

27 May 2020

Automotive Ethernet

Automotive networking is perhaps the latest application area for Ethernet. But Ethernet in some form has been going for nearly 50 years. Let's start with a little history.

History

Ethernet really started in 1971, although it was called AlohaNet. It was the first wireless packet data network. As you might guess from the name, it was created in Hawaii. The University of Hawaii had its main time-sharing computer system on the Oahu campus, but there were outposts of the University on other islands. It was satellite-based. I won't go into all the details, except for one. The system used a shared radio channel. When a system wanted to transmit, it would wait until the channel seemed to be free, and then it would transmit to the satellite. It would also listen to the satellite downlink. Due to delays, it was quite possible that the channel only seemed to be available when transmission was initiated, and that the packet would collide with another packet from another transmitter. The first level of retransmission was to notice that this had happened, wait a little, and then try again. There was a usual network protocol for handling end-to-end acknowledgment, packets that were damaged in other ways, and so on.

When Robert Metcalfe and David Boggs at Xerox PARC designed the original Ethernet in the early 1970s, some of the design was based on AlohaNet, which Metcalfe had studied as part of his PhD. Despite the name, Ethernet didn't run in the ether like AlohaNet, it ran over coaxial cable. But it had some of the same features: transmit if the channel seems to be free, listen for collisions, retransmit automatically. All the Alto computers at PARC were networked with the 3Mb/s original Ethernet.

The first commercial Ethernet was the 10Mb/s so-called DIX Ethernet, launched in 1980. DIX stood for DEC-Intel-Xerox, the three companies that created the standard: a company that built computers and needed a local-area-network technology, a semiconductor company that could build the chips to bring the cost down, and the original company that developed the idea (and was the only company with any experience running such a network). The "type" of networks went by the unwieldy initials CSMA/CD, for carrier-sense multiple-access collision detect. The 10Mb/s Ethernet also ran over a single shared coaxial cable. This soon became IEEE standard 802.3.

Gradually all other network technologies, such as Apollo's token ring, the Cambridge Ring, and others, were superseded by Ethernet. The shared coaxial cable was replaced with point-to-point twisted pairs (standard telephone wiring) in the 1990s, with switches and routers to get traffic to its destination. Today, nearly 50 years later, Ethernet's triumph over other wired (or fiber) networking technologies is pretty much total. Cloud data centers connect their servers to the top-of-the-rack router with Ethernet, the top-of-the-rack routers are connected within the data center with Ethernet. Data centers are connected to each other with Ethernet.

Automotive Networking

But in automobiles there were other wired standards, notably CAN bus (which I'm always disappointed doesn't stand for car-area-network but for controller-area-network). CAN bus worked well when very little data was being moved around inside a vehicle: moving the seats, adjusting the mirror, tuning the radio, sonar for reversing. But the move towards ADAS (advanced driver assist systems) and autonomous driving has changed everything. Now the typical car architecture is to have a powerful central computer, and connect it up to feeds from video cameras, radar, and lidar. These require a lot of bandwidth, more than CAN bus (and other technologies like LIN or FlexRay) can provide. When you need a wired network with a lot of bandwidth, there is really only one choice since there is the opportunity to leverage the entire ecosystem of chips, software, test equipment, network monitoring, and more.

Automotive Ethernet book Another advantage is that Ethernet runs over unshielded twisted pair. This means that it can significantly reduce the weight and complexity of the wiring harness. I have heard that these harnesses can reach 50Kg (100 lbs) which is a lot of weight affecting mileage, not to mention a lot of cost in manufacturing it. Since the car is literally assembled around the harness, if the harness fails, the car is usually scrapped — it is cheaper to buy a new car than attempt to replace it.

Bob Metcalfe, the inventor at PARC of the 3Mb/s original Ethernet (and the Metcalfe of Metcalfe's Law) said in the foreword to Automotive Ethernet: The Definitive Guide:

As I have been saying for 40 years: “If it’s networking you need, Ethernet is the answer; what is the question?” In this case, we might consider two related questions. “What took the automotive world so long? And how will it now use Ethernet going forward?”

So cars are going to contain Automotive Ethernet.

Automotive Ethernet

The above diagram shows that architecture with devices spread around the car and linked back to the head unit and the switch by Automotive Ethernet. In addition to the Ethernet capabilities required in data centers, Automotive Ethernet also needs to support Time-Sensitive Networking (TSN) and the IEEE 1722 Audio-Video Bridging (AVB) Transport Protocol. These technologies allow the network to provide quality-of-service (QoS) guarantees on latency and synchronization.

Doing an Automotive Ethernet Design

There are three parts to an Ethernet interface, the PHY, the MAC, and the controller. Cadence has Design IP (DIP) available for all three parts.

The PHY is the analog cell that actually connects the chip to the physical medium, in the case of automotive Ethernet, this is the twisted-pair copper, also known as 100Base-T1.

The MAC is the media-access controller that handles all the other aspects of the interface beyond the PHY. Compliant with IEEE Standard 802.3, the Cadence IP for Gigabit Ethernet MAC is highly customizable with support for an integrated 1000BASE-X PCS, a high-performance DMA with advanced AXI offloading features and descriptor caching, QoS, 1588 and TSN/AVB features to support any application. It supports a host of other features including VLAN, TCP/IP offload, and remote network monitoring (RMON).

The Cadence 10G/2.5G/1G Multi-Speed Ethernet Controller IP for Automotive Applications is a highly customizable soft controller IP, which is compliant with the IEEE 802.3 standard. To support a range of Ethernet applications, the Controller IP for Automotive features integrated 1000BASE-X and USXGMII PCS modules, a high-performance DMA with advanced AXI offloading capabilities and descriptor caching, QoS, and IEEE 1588. The Time-Sensitive Networking/Audio-Video Bridging (TSN/AVB) functionality enables unified Ethernet communication of critical data without traffic congestion in shared networks. Furthermore, the Controller IP for Automotive supports a host of other features including IEEE 802.3az Energy-Efficient Ethernet (EEE), VLAN, TCP/IP offload, and remote network monitoring (RMON).

For analyzing the wires that make up the network, use Sigrity SystemSI Automotive Ethernet Channel Simulation. This lets you analyze the ECU-to-ECU communication performance via the physical Ethernet channel with Sigrity SystemSI technology for automated chip-to-chip signal integrity analysis. This lets you test different combinations of IP, different cable lengths, shielding vs non-shielding, aging, and compliance checks for automotive Ethernet 100Base-T1 PHY.

For analyzing the transmitters and receivers, Automotive Ethernet Compliance Checks for 100Base-T1 BroadR-Reach PHY involve provided IBIS and AMI models for the IP. For a deeper dive into how this is done, see my post AMI and IBIS: Who Put the Eye in AMI? Using Sigrity SystemSI Simulation can characterize output droop, power spectral density, jitter, clock frequency, distortion, and return loss.