中文 English

Battle Brewing Over Automotive Display Protocols

Different options and standards will take years to sort out.


Displays are multiplying in new and future automobiles. That means a lot more display data moving around the vehicle and traveling some distance between sensor and processor.

While existing protocols can handle some of the new duties, new protocols also are being developed specifically for this application.

“Automotive displays are proliferating, increasing in numbers and in pixel density, and, most importantly, are being used for mission-critical functions like instrumentation displays,” said Steve Creaney, design engineering director, IP Group at Cadence.

MIPI and VESA have existing protocols that are moving both from the mobile and laptop worlds into vehicles. But new PHY layers are needed to deal with some of the longer distances that the data must run. And at least one new PHY is entering the fray, meaning it’s not yet clear which PHY will dominate in the future.

Displays proliferate
Having a digital display in a car is a relatively new phenomenon. The center console was one of the first displays, replacing buttons and knobs for various controls. Having passenger and back-seat displays is also becoming common, especially in vehicles designed for multiple children or passengers.

In addition, the instrument cluster is being replaced with a display. “Whether it’s one big display or multi-display, it’s designed to look continuous,” said Willard Tu, senior director, automotive at Xilinx.

Even mirrors, both side and rear-view, are moving from simple reflections to e-mirrors. “You’re looking at a display built into a mirror,” said Tu. “That’s a very small display, so the resolution isn’t as much of a factor.” There also may be heads-up displays in our future.

Much of the data will originate in cameras. “You have many cameras in a vehicle — up to 10, sometimes more,” said Thierry Kouthon, technical product manager, security at Rambus. “Autonomous vehicles use even more camera feeds.”

Some of these displays will need high-quality video. “In the automotive industry, we want to build cars that have high-definition cameras, 4K cameras,” said Linley Gwennap, president and principal analyst of The Linley Group, at the Linley Spring 2021 conference.

Hezi Saar, director of product marketing for mobile, automotive and consumer at Synopsys, concurred. “4K is what people are adding for the central cluster and for the passenger display.”

The center and passenger displays likely will be touchscreens, meaning that they must carry signals in two directions. And the passenger display is likely to need high-bandwidth digital content protection (HDCP) when showing copyrighted material.

Fig. 1: Future cars will have a number of sensors that generate data both for computation and for display. Source: The MIPI Alliance

Lots of data to move
Graphic feeds will start either at a sensor or at a wireless communication portal (for streaming video, for example). From there it must travel to an SoC for processing and forwarding to one or more displays.

That makes for a lot of data moving through the vehicle. And it must do so on a wiring harness that vies to be the heaviest component in the car. Anything that can ease that data-transport burden will help to lighten the load in more ways than one.

Fig. 2: Expected bandwidth needs for 2026 production. Source: The Automotive SerDes Alliance

Fig. 2: Expected bandwidth needs for 2026 production. Source: The Automotive SerDes Alliance

The question then becomes, which protocol(s) should be used to move all of that data? The industry has moved beyond the older LVDS-based OpenLDI approach. Today, proprietary approaches predominate. But, as Ashraf Takla, president and CEO of Mixel, noted, “The automakers’ supply chain is sick and tired of having proprietary solutions for this application.”

Craig Wiley, board member of VESA and senior director of marketing at Parade Technologies, agreed. “The automotive industry doesn’t want to be held hostage to a single supplier,” he said.

There are two well-established alternatives, and they come from two different scales of display. MIPI protocols (CSI and DSI) have been used for years in small mobile displays and are migrating to automobiles. “Almost half of our members have some business in the automotive sector,” said Peter Lefkin, managing director of the MIPI Alliance. The second is VESA, with its eDP protocol, which originated in larger computer displays.

MIPI and VESA protocols can co-exist within a vehicle — MIPI for small displays, VESA for large — so they need to share a PHY in order to be efficient. “You can transmit a DSI packet and an eDP packet to two separate displays, just packeting them differently,” said Saar.

Focus on the PHY
All of the physical connections being used today and in discussion for the future are serial. “The best efficiency is achieved by getting data on a single lane — serially,” said Manuel Mota, product marketing manager for high-speed SerDes PHY, analog, and bluetooth IP at Synopsys.

DSI (Display Serial Interface) is the display-side protocol from MIPI. It runs between the processing SoC and the display. CSI (Camera Serial Interface) is the corresponding protocol between the sensors and processor. Each historically offered two PHY options: D-PHY or C-PHY.

“D-PHY has a discrete clock, and C-PHY has an embedded clock,” said Rick Wietfeldt, member of the board of directors at the MIPI Alliance and senior director of technology at Qualcomm. “D-PHY has two wires of data and two wires of clock. C-PHY has three wires of data and clock combined.” eDP, meanwhile, uses a SerDes with 8B/10B encoding for clock recovery, similar to USB and PCIe.

All of these options are considered short-reach protocols. They can go from chip to chip, but they’re not intended to run over cabling. The challenge is that sensors, SoCs, and displays in a vehicle may be far apart. “You have sensors distributed all over the car,” explained Takla. “And then you have an ECU and a long distance to go between the two — up to 15 meters.”

Fig. 3: A survey showed high willingness to use a serdes approach for display and camera data, although less willingness for other sensors. Source: The Automotive SerDes Alliance

Fig. 3: A survey showed high willingness to use a SerDes approach for display and camera data, although less willingness for other sensors. Source: The Automotive SerDes Alliance

There’s no way that the traditional PHYs could traverse that distance, which has triggering a move to long-reach SerDes-based links. Within MIPI, the result is its A-PHY, which can serve as the PHY for DSI, CSI, and eDP. The intent is that it is far more electrically robust in the presence of interference than the older PHYs.

“You have these radar stations that people drive by, and they get 80 mV peak-to-peak noise on the signal,” noted Wiley. “That really cuts into the margin of what you’re capable of transmitting across the cable.”

A-PHY supports 10 Gbps of data transfer, moving from NRZ signaling at the slower speeds to PAM-4/8/16 at high speeds. “PAM is needed for long reach,” noted Wendy Wu, director of marketing, IP group at Cadence.

“It’s just two wires, and we have a forward channel and a reverse channel on the same pair of wires,” said Wiley. “If you want to make it faster, you have four wires.”

Fig. 4: The MIPI roadmap for A-PHY shows NRZ for the lower downlink data rates, and the uplink with PAM providing greater bandwidth at the higher speeds.

A daisy-chain wiring topology is gaining support. “You start from one end, and you cover, let’s say, the right-hand side of the car with one daisy chain and the left-hand side with a separate daisy chain, all of which is driven by a chip in the front part of the car,” noted Wietfeldt.

In that daisy chain, all data may be transported to all stops, with each stop accessing only its data. Alternatively, the daisy chain can be operated more like a package-delivery route, with payload being dropped at each stop.

Wiley provided an example: “You have a video controller, an instrumentation panel with two displays on it, and a heads-up display. What will come into the A-PHY would be three streams. And then two streams get terminated in your panel, and then one more stream continues to the heads-up display.”

While A-PHY in the short term may link camera and display bridges, the goal is that in the longer term it will be integrated into the cameras, displays, and processing SoCs so that the bridges and other links can be removed. This will simplify and lower the cost of the bill of materials (BoM).

“These bridge chips are not cheap,” observed MIPI’s Wietfeldt. “They may be a couple of dollars. So it is a huge cost and BoM advantage.”

Fig. 5: In the MIPI view, bridges interconnected by proprietary long-reach links will move to A-PHY and will ultimately be integrated into the other components. Source: The MIPI Alliance

He’s not alone in pushing for this. “We’re working to standardize and move forward our A-PHY specification within IEEE,” noted MIPI Alliance’s Lefkin.

A new kid in town
There is another new PHY, however, that’s also intended to solve the long-reach problem. It has been developed by the Automotive SerDes Alliance (ASA), and it appears to compete directly with the A-PHY.

The ASA Motion Link, as it’s formally called, will handle data for both camera and display. It can handle both home-run and daisy-chain links just as the A-PHY can. It’s primarily a stand-alone PHY, under the expectation that this application doesn’t need an entire seven-layer stack. But it can operate with overlying protocols, given appropriate adaptation layers.

ASA Motion Link has a number of bandwidth options ranging from 1.3 to 13 Gbps. It uses TDD, with NRZ up to 6.4 Gbps and PAM-4 beyond that. “There’s a version of the specification out there now that has all the functionality and all the features in it,” said Claude Gauthier, director of strategic innovation at NXP. “Currently, there’s a move to add more bells and whistles.”

Future versions are likely to increase speed by allowing link aggregation, analogous to what’s called “lane bonding” with other protocols.

Both A-PHY and ASA Motion Link are asymmetric. They provide high bandwidth from the camera or to the display, while including a low-bandwidth back channel for control signaling.

The reason for the separate and independent ASA development isn’t publicly clear. In some of their statements and materials, it is positioned as the only standardized alternative to proprietary schemes, without acknowledging the existence of the MIPI/VESA solutions. And some say that, during the A-PHY definition process, the group split, with one side moving to create the new ASA group.

Some further digging revealed that the main concern seems to be that the A-PHY technology comes from one company, Valens, which contributed it to the standard. However, “MIPI A-PHY, like all MIPI specifications, is made available under royalty-free terms,” said Lefkin.

Still, the issue for ASA members appears to be that only essential patents get a license. Valens has implemented this in a way that includes non-essential patents, and licensees don’t get access to those patents. A statement from the ASA steering committee noted, “There are examples where the solution of one supplier was successfully made a standard, but there are many examples where it did not work.”

The ASA folks are more interested in a process where multiple companies contribute technology without one company dominating. The ASA effort is licensed under FRAND (free or reasonable and non-discriminatory) licensing.

MIPI’s concern is there may be royalty uncertainty prior to acquiring a license. There apparently has been some history in the automotive realm of companies refusing to license essential patents. Regardless, it’s pretty clear that the A-PHY and the ASA PHY will compete head-to-head. How that competition resolves itself is not yet evident.

Reducing bandwidth (and power)
Two tools are available for reducing bandwidth — compression and “smart” displays. With compression, the critical question is whether it can be visually lossless enough for use on safety-critical displays.

DSC (Display Stream Compression) is from VESA, but it can run atop eDP or DSI, giving a 3:1 compression ratio. A newer format, VDC-M (VESA Display Compression – M), comes from a MIPI/VESA collaboration. It provides as much as a 6:1 compression ratio.

The MIPI Alliance did a study to see whether its compression techniques could be considered visually lossless for use in safety-critical displays like the instrument cluster. It issued a white paper indicating results that support the use of compression. “VDC-M and DSC are visually lossless, which means that over the study no one was able to see any difference,” said Lefkin.

While DSC is establishing something of a legacy in non-automotive applications, which appears to be delaying VDC-M, that’s not necessarily the case in automotive. VDC-M may get earlier traction there because there is no current incumbent technology. “Automotive users were understandably reticent to use compression,” said Cadence’s Creaney. “However, there is increasing interest due to the cabling cost/weight savings available and more studies into functional safety measures for compressed video.”

The other practical reality is that many chips will provide a combo. “It’s likely that you won’t have a VDC-M without DSC,” said Wietfeldt.

ASA, meanwhile, also will have compression, although that’s still a work in progress. So far there is no compression-ratio target.

Command-mode or “smart” displays also can reduce bandwidth. Simpler “RAMless” displays must be constantly refreshed with data even when the data isn’t changing. Command-mode displays, by contrast, can hold their state. Changes can be made by commands that specify only the part of the display that needs to change.

While this is not likely to have much of an impact on the videos being watched in the back seat, it can significantly lower the rate of data transfer for the instrument cluster.

Keeping it safe and secure
Automotive applications have safety requirements that smartphones (MIPI’s traditional market) and laptops (VESA’s traditional market) don’t have. That brings a functional-safety requirement that hasn’t existed in the past for these protocols.

Vehicles also present a number of security challenges. “The display port and entertainment system are often a ‘Trojan horse’ used to get into the ECUs of the more critical parts of the system,” said Kouthon. So security is a key new requirement.

MIPI has modified both CSI and DSI to take into account not only safety considerations, but security as well. “For functional safety across an interface, you need three elements,” said Wietfeldt. “One is a CRC to protect the data. Secondly, you need some type of frame counter to warn you against transmissions that are being replayed. The third element is something that’s often called a ‘keep-alive’ counter to identify if the transmission is altered.”

ASA also includes functional-safety considerations. “ASA provides several tools (RS FEC, CRCs, Container Identifiers, data starve flag, etc.) to detect and diagnose malfunctions inside and outside the ASA-compliant domain,” said Gauthier.

For security, MIPI has leveraged DMTF.org’s SPDM (Security Protocol and Data Model) specification, which is an offshoot of the more familiar TLS protocol that works with TCP. This is end-to-end, implemented in the equivalent of their application layer. “Today, every other implementation is linked based,” noted Wietfeldt.

Also included is HDCP for materials like movies that may be copyrighted. “Typically, people use HDCP 2.2 or 2.x. which encrypts the data at full bandwidth between the source and the destination using cryptographic algorithms similar to AES,” said Kouthon.

Fig. 6: End-to-end security, not link-to-link (or bridge-to-bridge). Source: The MIPI Alliance

ASA also addresses security. “ASA offers an optional security sublayer in the Data Link Layer for authentication and encryption protection,” said Gauthier. “ASA will enable stream-based HDCP implementations that are compliant to Interface-Independent Adaption HDCP through an Application Stream Encapsulation Protocol (ASEP), an ASA-specific interface.”

The longer view
While these PHY battles play out in the short term, there are some who think the long-term view could be very different. The automotive industry is said to be in a state of turmoil. The basic goals of the future car are not in question, but how those goals are met is likely to vary widely until things settle out. And that settling out could take years.

“Everything from the PHY all the way up to the capabilities of the display itself are in massive flux right now, and will be for the next couple of years,” said David Fritz, senior director for autonomous and ADAS at Siemens EDA. “We’re working with the OEM teams to design, from a blank sheet of paper, their next-generation vehicle to intersect the model years around 2025/2026.”

That means that the current approaches, most of which leverage legacy technologies, may last only so long. And lurking behind all of this is Automotive Ethernet.

There are efforts underway to create a holistic communication scheme for the entire automobile. Given what the tea leaves are indicating today, that suggests Ethernet everywhere, with no bridging between Ethernet and non-Ethernet links.

“There is a holistic view of how data is transmitted across the vehicle,” said Fritz. “Display interfaces and anything that produces content are rapidly moving into the automotive Ethernet realm.”

Both the MIPI and ASA folks contemplate this possibility. “Will there be a push to have Ethernet manage all of these sensors and cameras?” asked Gauthier. “Will Ethernet gobble them up as network objects?”

At present, Ethernet is less competitive because it has only 1 Gbps of bandwidth. But a 25-Gbps version is in definition, so that should be sufficient to handle the camera and display needs for the foreseeable future.

Ethernet is also symmetric, providing equal bandwidth in both directions. For the camera/display application, that’s relatively inefficient because one pays for the excess upstream bandwidth without using it anywhere close to fully.

“What’s emerging in automotive now is 1 Gbps up, 1 Gbps down,” said Wietfeldt. “And that doesn’t make sense for a camera where there’s no need for doubling the interface cost by having a fully symmetric interface.”

That creates a distinction for them between Ethernet as the backbone and camera/display connections. “Whether it’s A-PHY or one of the other ones, including proprietary ones, they’re distinct from the ethernet category of connectivity,” Wietfeldt added.

That may not be the case forever, though. “The group developing the 25-Gbps version of Ethernet is looking at this asymmetric use case,” said Gauthier. That could make it more competitive with the existing solutions.

MIPI’s view is these aren’t competing technologies. “We don’t at all profess that A-PHY replaces Automotive Ethernet,” said Lefkin. “We see them as very complementary for different applications.”

MIPI has touted that its security is end-to-end, not link-by-link. That usually involves network layers 3 or 4 or higher. But Ethernet is a link-layer protocol, meaning that it has no reach up into those higher layers. And yet an end-to-end solution is possible.

Automotive Ethernet uses MACsec for security and achieves end-to-end by tagging a frame with the destination MAC address, so any intervening switches will pass the encrypted content along. Only the endpoint will terminate the security.

In this view, the MIPI/VESA/ASA approach would, over time, be supplanted by something considered closer to Ethernet – unless they made future changes that aligned them better with Ethernet. “I’d see an interim transition to DSI for power reasons,” said Fritz. “And then something that will incorporate a holistic service-oriented architecture, where the display is simply one more node that consumes information.”

ASA positions itself alongside Ethernet by describing the camera/display subsystem as being at the edge, “outside the network,” and presumably therefore not an appropriate target for an ethernet takeover. It also sees the ASA approach as being just a link, not a full network. “One of the reasons that SerDes is used instead of Ethernet is that it doesn’t have all the network overhead,” said Gauthier.

Standards bodies may end up on the sidelines for a while. “There’s going to be one dominant de facto standard, and there will be other OEMs who see this as one of the shrinking number of ways to differentiate from everyone else,” said Fritz. “I’m afraid that the standards bodies are going to have to catch up.”

So even when the near-term clash between PHYs is resolved, more change is likely down the road. The camera and display subsystems in cars are likely to remain as much in flux as the cars themselves will be for the next couple of decades. “I really think that in 2025 to 2030, the automotive industry is going to look absolutely nothing like the way it looks today,” said Fritz.

Leave a Reply

(Note: This name will be displayed publicly)