(June 27, 2017)
The future of smart, self-driving cars means massive change for their electronics suppliers, from advanced internal networks, to displays that meet critical safety requirements, to 5G communications with the road ahead, and even drivers’ licenses for AI drivers.
A high-end car will contain more than $6,000 worth of electronics in five years, driving a $160 billion automotive electronics market in 2022, as almost all the electronics in a car undergo massive change, predicts IHS Markit principal analyst for Automotive Electronics Luca De Ambroggi.
IHS Markit notes that more of the value of automotive electronics is moving to software and semiconductors, with semiconductors for the automotive market seeing more than seven percent CAGR through 2022, outpacing the 4.5 percent growth of automotive electronic systems in general, and the 2.4 percent growth in vehicle units over that period, as the sector moves to full solutions that include significant software value.
The auto makers are starting to rethink their electronics entirely to integrate more intelligence. They’re looking at replacing the standard flat architecture based on a controller area network (CAN) bus with a more advanced network to handle the increase in complexity and data sharing, says De Ambroggi. “All the electronics systems will need to meet stronger safety standards, starting with security, since a car that is not secure is not safe,” he notes. That includes even the display, which used to just control the radio and connect a phone, but now will communicate driving-related information, so will have to meet critical safety requirements too.
AI needs a common platform, and a drivers’ license
“Artificial intelligence will be the enabler of fully autonomous (L5) vehicles that take over the driving, but it will take some years more to meet the requirements for performance, safety and cost, and today´s silicon technology is not yet good enough,” suggests De Ambroggi, noting the need for more efficient new architectures and reduced power usage. However, the automotive market alone is likely not large enough to support the development expense required, so it may require more general purpose semiconductor components that can be customized for the automotive market.
This in-car AI system will need to go beyond recognizing objects, to using that information to predict behavior, as when it sees a pedestrian distracted by his smart phone and judges he may walk into the street. Even chip-level speech recognition, which is relatively mature in automotive infotainment, will need to improve significantly for driver-assistant applications. “One of the challenges for AI in safety critical applications is how will we know that the deep learning is good enough? How can we certify a virtual brain? We’ll need to create a standard drivers´ license for the machine, to certify that it is smart enough to drive,” he suggests.
On the sensor side, beyond RADAR and camera, LIDAR (Light Detection and Ranging) will become a “must have,” with demand for ~35 million units by 2025, but some 15 different technologies are competing for a piece of the business. Input from these sensors will need to be linked for accuracy and redundancy. IHS projects the bill of materials for sensor fusion for advanced driver-assistance systems will double in value by 2025, although the cost for basic surround-view sensor fusion for parking applications will fall to half, as it becomes a commodity.
More big potential change from 5G connectivity in the cars of the future
The tidal wave of data as autonomous vehicles produce up to 4 terabytes of data per day by 2020 likely will require 5G communication and cloud storage, notes Katherine Winter, VP of Intel’s Automated Driving Group. She notes that Intel is powering the “brain” in hundreds of autonomous test vehicles today and is working to provide an autonomous driving end-to-end solution that encompasses the car, connectivity and the cloud. The collected information will be pushed to the data center by 5G for learning and algorithm training, as well as high-definition map access that will help the car navigate in a safe and efficient way an provide real-time updates to the vehicle on the road. While Intel is driving many innovations in the field, Winter stresses the importance of partnerships, and the opportunities in greater ecosystem surrounding the autonomous vehicle. “No one company can do it alone,” she says, pointing to Intel’s work with BMW and Mobileye.
Besides all the sensors and on-board intelligence, the smart car of the future will also depend heavily on next-generation 5G connectivity, concurs Nakul Duggal, VP Product Management at Qualcomm Technologies, Inc., on the connectivity technology requirements for autonomous driving. “5G technology is set to transform the automotive industry and redefine how we think about automobiles as well as how we use them. By 2035, we expect 5G technology will enable over $2.4 trillion in economic output across the broader automotive sector,” he says.
Figure 1: 5G moves the needle on automotive intelligence.
Source: Qualcomm Technologies, Inc.
While advancements in radar, LIDAR and camera systems are encouraging, these sensors are limited by their line-of-sight. Direct Cellular V2X (C-V2X) and its evolution to 5G will complement the capabilities of these sensors by providing 360-degree non-line-of sight awareness, extending a vehicle’s ability to “see” further down the road – even at blind intersections or in bad weather conditions. By complementing other sensors, C-V2X provides a higher level of predictability by conveying location, speed, direction and even intent, which the other sensors can only estimate.
Ideally, the smart car will know very precisely where it is within a 3D High-Definition (HD) map that’s a rich model of the real world, updated in real time, by C-V2X communications. Building accurate 3D HD maps require decimeter-level position accuracy all the time and under all conditions. And it will necessitate new types of high speed wireless connectivity, designed from scratch, to transfer 3D HD mapping data between cars, between cars and the network, and between cars and the smart city infrastructure, to enable awareness of hazardous conditions, from potholes in the road and construction zones to a traffic accident around the bend ahead. The natural evolution of 5G technology may mean that new cars will increasingly have modems that can handle CV2X, while the many distributed small cells required by 5G could be co-located with roadside units across the infrastructure, he notes.
First published in EE Times; written with contributions by Gary Frank.