AV processor battle heats up, while the future of EVs hinges on wide bandgap technology

Junko Yoshida, chief international correspondent, EE Times

recent memory, no topic has generated as much interest and enthusiasm among
investors, entrepreneurs and engineers as highly automated vehicles (AV).
Seemingly, almost everyone in the high-tech industry wants to climb onto the
self-driven bandwagon — including robotic scientists, AI specialists, chip designers
and sensor developers.

significant, the promise of AVs has energized traditional automakers around the

Kato, senior executive director at Denso Corporation noted in his keynote
speech at International solid-state circuits Conference (ISSCC) this year that
the automotive industry is experiencing a “once-in-a-century transformation” —
driven by advancements in connected, efficient and automated driving.

the high interest, the progress of EV has not matched the predictions made several
years ago. The number of pure-electric and plug-in hybrid
cars represented barely 1 percent of the 17 million cars and light trucks
sold in the United States last year. Meanwhile the promise of automated
driving – billed as “a lot safer than human driving” – is facing intense
scrutiny after a fatality caused by a self-driving Uber in Tempe, Arizona.

Slow progress due to the charging infrastructure & battery concerns for EV. (Source: IHS Automotive)

it can be easy to dismiss the promise of EVs and AVs as marketing hype few in the
tech, auto and semiconductor industries can afford to sit on the sideline as a wave
of AV innovations unfolds before their eyes. A revolution is happening in big
data, deep learning, AI processing on the edge and perception technologies
enabled by advanced sensors. Similarly, on the EV front, companies like Infineon,
Wolfspeed and Rohm are among many who see a pot of gold in the promise of Wide
Bandgap (WBG) semiconductor technologies. They believe it will
help downsize automotive inverters by increasing power density.

Where EV and AV

AVs and Connected Cars aren’t being developed in silos. Tier one OEMs are
integrating the best of today’s technology innovations by developing electric
cars infused with automated driving and connectivity.

EV’s foremost virtue is a lot fewer moving parts. Its three major components
are battery, inverter and electric motor. In contrast, an internal combustion
engine contains thousands of tiny pieces that must be maintained. From an
engineering point of view, the EV matches AV technology perfectly because an AV
needs more electrical brainpower to manage vision, sensor fusion, mapping and
path-planning functions while processing an ever-growing volume of software.

marks a radical transition – from mechanically driven cars to software-driven
vehicles – for the auto industry.  It opens the door for designers to
develop apps that make EVs run more efficiently, for example.  

Semiconductors launched earlier this year what it calls “Greenbox,” a platform
for car OEMs to develop new hybrid and EV apps. 

NXP Greenbox platform

GreenBox supports the development of HEV
and motor control applications. (Source: NXP Semiconductors)

Cornyn, vice president and general manager for NXP’s vehicle dynamics and
safety product line, said designers can use Greenbox to create applications
that boost overall energy efficiency in route-planning. One example is an
HEV facing a long uphill grade. Cornyn noted that a piece of software using specific
route knowledge – obtained by the AV – can enhance the HEV’s battery management
up the hill.

Rapid AV

Fueling investment and progress in AV
development are new entrants in the field, especially tech companies like

Juliussen, director of research for infotainment and advanced driver assistance
systems (ADAS) for automotive at IHS Markit, said, “Waymo has been quietly
taking the lead [in the driverless market]. They are ahead of everyone else on
the field.”

a safety report published last year, Waymo explained its self-driving
software and hardware, and how it tests vehicles. Citing the report,
Juliussen noted that Waymo’s difference from competing robocars is “designing
their own sensor systems from the software point of view.”

eight years of work on driving software, Waymo has learned to “see what’s
around the car far better than others,” he said. Waymo’s ability to tightly
couple software mimics the Apple approach, Juliussen observed. This is
something traditional carmakers, who lack software prowess of their own, have difficulty

In the
United States, the AV is not a question of if, or even when. The first domino
toppled when the state of Arizona in January granted Waymo a permit to operate
as a transportation network company. Early in February, Waymo confirmed its
plan to start charging customers for robo-taxi rides in 2018.

the Uber accident in Arizona in March, Uber, Toyota, Volvo and Nvidia all
announced temporary test-drive suspensions for their vehicles on public roads.
But Waymo and GM appear to be moving forward.


Among all car OEMs, Toyota, one of the most
conservative Japanese car companies, made the best argument earlier this year
at the Consumer Electronics Show for why the society needs highly automated

than getting stuck in the perennial argument over who gets to drive the car
(machine or person), Toyota steered its AV vision along a different road. Their
idea is “e-Palette,” an automated vehicle for on-demand shopping and
distribution, on-demand food trucks and hospital shuttles — a veritable
on-demand city. 

Toyota ePalette

Toyota’s e-Palette demo at CES 2018 (Photo: EE Times)

course, Toyota Motors CEO Akio Toyoda isn’t alone in defining his company’s
challenge as changing from car-making to mobile service. By partnering with
Amazon, DiDi, Uber, Pizza Hut and Mazda, Toyota, at least, advanced the idea
that AVs and EVs aren’t just robo-taxis. They can create a “mobility
platform” open to other developers.

Baidu factor

In the global world of EVs and AVs, China is emerging
as a big factor. Many analysts believe that Baidu, styling itself as “the
Google of China,” could put a dent on the market. The company has shown an
ability to harvest big data in China. Its commitment to Apollo, an open
platform for highly automated vehicles, reinforces its case. 

on the success of the original Apollo platform announced last July, Baidu has
been moving to update the application. Apollo 2.0 will unite all four of the
Apollo platform’s modules — cloud services, software, reference hardware
and vehicle platforms. Baidu claims that Apollo — based on Baidu’s Duer
OS — can now autonomously guide a vehicle through basic urban
environments, even at night.

has amassed more than 90 partners, including support from chip vendors of
computing platforms such as Nvidia, Intel, NXP and Renesas. The Apollo platform
is designed to lower the bar for car OEMs. Reportedly, 200 car OEMs are in
China today. Apollo would enable them to enter the AV market, much as Google’s
Android created a huge community of non-Apple smartphone vendors in China.

have big data, we provide open development platform and we [China] have volume
production capabilities,” said Qi Lu, Baidu’s vice chairman, at the Consumer
Electronics Show. More important, he said, China provides “a policy friendly to
AV [development and testing].” To discourage a wave of fledgling Chinese car
OEMs fragmenting AV development, the Beijing government has officially endorsed
Apollo, Lu said.

released in March “Apollo Scape,” billed as the largest open-source training
dataset for autonomous vehicle perception, to advance the company’s big data business
and promote collaboration with tech and automakers in the West. Apollo is also
joining an Industry Consortium, the Berkeley DeepDrive, an alliance of a group
of companies, including Ford, GM and Nvidia that are investigating technologies
in computer vision and machine learning for automotive applications. This
consortium will have access to the Apollo Scape dataset.

Magney, founder and principal at VSI Labs, described this dataset as “a
tremendous benefit to any company or entity that is developing a camera based
AI-perception system.”

explained, “Not only is the dataset gigantic, but it is also very diverse,
containing images and labels from many complex environments, weather, and
traffic conditions. Of course, the larger and more diverse the dataset is, the
more capable/useful it is to train a very capable AI perception system.” He
added, “Furthermore, this dataset not only has labeled objects, it has
pixel-by-pixel semantic segmentation, which means each individual pixel has a
class label. This is much more useful for training perception systems but is
also much more laborious / time consuming to create.”

Baidu has invested lots of time and money into this dataset. So why give it
away for free? “We believe Baidu is putting massive investments into enabling
every company to develop self-driving,” Magney said. “This is because Baidu
believes the more adoption of self-driving, means there is more money to be
made from licensing the Baidu Mapping and Localization assets among other

AV processor

The hottest story among chip companies is the
question of who will rule the AV processing platform. Just as Intel dominated
the PC platform and Qualcomm rules the mobile front, every chip company sees
fresh opportunity in the nascent AV processor market.

far, Intel (Mobileye) and Nvidia are dominant AV platform vendors.  Intel has
its Go automated driving platform, consisting of Mobileye-designed EyeQ5 for
vision, another EyeQ 5 for fusion and planning (the chip will take sensory data
from radars and lidars), Intel’s low-power Atom SoC for trajectory validation
and issuance, and other hardware including I/O and Ethernet connectivity.

is pushing its Drive platform, combining deep learning, sensor fusion, and
surround vision. One of the SoCs Nvidia developed for Drive is Xavier, a
complete system-on-chip. It integrates a GPU architecture called Volta, a
custom 8 core CPU architecture, and a new computer-vision accelerator.

Renesas Electronics disclosed earlier this year that it is “ready to intersect”
with the demand for highly automated vehicles in volume, by rolling out its
next-generation R-CAR SoC. Although the Japanese company will not start
sampling until 2019, the new SoC, the company claimed, will “double
deep-learning performance efficiency compared to the Intel/Mobileye’s upcoming
EyeQ5 SoC.” According to Intel, it will provide 24 TOPS at 10 watts.

in AV processor race, however, aren’t limited to incumbents. Startups have been
springing up, all claiming breakthroughs in the realm of deep-learning

(pronounced “Think-Eye”), for example, last year unveiled at Hot
Chips details of its high-performance processor, Graph Streaming Processor
(GSP), billed as a “next-generation computing architecture.” Asked to
distinguish GSP from GPUs and DSPs, ThinCI cited GSP’s abilities in
direct-graph processing, on-chip task-graph management and execution, and task
parallelism. “We believe our GSP can beat the computing engine in any of those
processors,” said Dinakar Munagala, ThinCI’s CEO.

a large Japanese tier-one OEM and a key investor in ThinCI, last year revealed
the launch of a new subsidiary to design and develop semiconductor IP cores for
key components in automated driving. The architecture of a new chip is being
jointly developed with ThinCI, Denso announced. Denso calls it a Data Flow
Processor (DFP) and said it is “very different from CPU or GPU.”

EVs’ future

It’s no overstatement that both public and
private institutions around the world are eager to accelerate the evolution of
Wide Bandgap (WBG) tech into commercial power electronics applications. Many
believe that the future of EVs rides on WBG.

Because WBG materials such as silicon carbide (SiC) and gallium nitride (GaN)
can greatly improve power-conversion efficiency.

bandgaps allow WBG materials to withstand far higher voltages and temperatures
than silicon. This promises greater durability and reliability at higher
voltages and frequencies, enabling higher performance for less energy.

in applying WBG technology in an EV are clear. Increasing energy efficiency
reduces the cost of battery packs, shortens charging times and improves driving

automotive industry already includes EV manufacturers who are early adopters of
SiC-based inverters. “We strongly suspect that Tesla is using SiC based
inverters in some of their Model 3 vehicles for example. Several
announcements in Japan including Toyota have been done regarding SiC based
inverters for EVs by 2020 even though it might be for small volume production,” said
Hong Lin, technology and market analyst, at Yole Développement. “The
market has started, we might now enter the hockey-stick adoption rate.”

On the
other hand, SiC adoption in EVs’ powertrain appears to have a long way to go.
Yole listed four major hurdles for SiC adoption as “cost, reliability,
integration and supply chain.”

biggest challenge lies in economy of scale and SiC wafer price. To respond
to automotive’s main requirements (cost and reliability), “high
quality 6-inch SiC wafers are mandatory,” said Mattin Grao, Technology &
Market Analyst at Yole. “Only few suppliers, such as CREE, SiCrystal and II-VI,
are available to offer these substrate sizes and quality. This has caused a
shortage in 2017.”

important, “As SiC devices will always be more expensive compared to a Silicon
device, the carmakers need to get the value out of the system,” Grao pointed
out. Automakers need to understand their ROI, determining whether they can
absorb the price difference for SiC devices.
There are some hopeful signs, especially for
potentially improving the SiC supply chain.

and Cree, for example, recently agreed on a long-term agreement for the
provision of SiC wafers. Yole sees Infineon securing with Cree a 6-inch SiC
wafer supply as critical. “Getting access to high quality SiC substrates is
definitely key in this industry,” said Grao.

however, cautioned that not every automotive OEM, at this point, is sold on SiC
over silicon. “Not all of OEMs have been able to prove SiC performance in their
demonstrators,” said Grao.

Until more carmakers – beyond Toyota, Tesla, BYD
– are convinced to include SiC technology in a next-generation product line,
its market will remain unpredictable.