32 PART | I ITS technology enablers
From what we see today three key topics drive the growth of in-vehicle and
backend performance needs:
• Over the air software update enabling new services and features by software
after SOP;
• New updates during product lifetime require scalable HW concepts and ad-
ditional reserved performance “headroom” for future services introduced
after the start of production;
• Cybersecurity requirements to preserve operational and functional safety.
3.2 Consequences for vehicular electrical/electronic (E/E)
architectures
Resulting from these automotive innovation drivers, a dramatically increased
computing performance is required. The 100 and more ECUs that today are
provisioned in E/E vehicular architectures are hardly able to cover the increas-
ing requirements for high computing performance, which will guarantee a real-
time response on emergency, operational and functional safety, and security to
external threats and attacks. The novel autonomous vehicle E/E-architectures
must be based on a few HPC solutions that comprise specialized hardware with
graphics processing units (GPUs) and special controllers and are able to accom-
plish multiple perception tasks in real-time without fail (Kovacˇ, et al., 2019).
In order to achieve real-time vehicle perception and build a reliable model
of a vehicle’s surroundings, it is necessary to collect and process heterogeneous
sensor data, to be able to provide on-the-fly processing and analysis of video
streams, and be ready to integrate information and knowledge from various
sources and in varying condition in order to take correct decisions. It thus relies
on the data fusion and behavior prediction capabilities of deep learning models
(e.g., for computer vision), which must be low latency, low complexity, and eco-
nomic in terms of processing power needs. A feasible solution to this complex
problem seems to be the replacement of embedded automotive processors that
are incapable to provide advanced data fusion services (from data-intensive sen-
sors such as LIDARs, ultrasonic radars, and cameras), with high-performance
processors such as those used in HPC centers and their combination with spe-
cialized MCUs for the automotive example. All the issues that emerge for the
automotive industry concern the management of big data that result from the
various sensors, the onboard application of deep neural network models for the
real-time processing of such data at inference time and the offline training of
massive amounts of historical data at cloud or server level. All such solutions
strongly rely on transparent and secure communication between car-embedded
HPC and cloud-based HPC. This perspective leaves space for many collabo-
ration opportunities between the automotive and high-performance computing
industry both in terms of novel architectures, as well as in terms of hardware
and software that will efficiently cover the automotive industry needs.