Page 44 - EE Times Europe Magazine - June 2025

44 EE|Times EUROPE

AV Safety Is Multifaceted—Even at the Hardware Level

LiDAR digital twin platform co-developed by Valeo and Applied Intuition. On the right: virtual reality in the simulation (resembling a real
camera feed). On the left: virtual LiDAR point in this simulation (resembling a real LiDAR point cloud). (Source: Valeo)

expected. We’ve developed proprietary safety mechanisms, built into digital twin isn’t just a tool for efficiency; it’s becoming fundamental
our architecture, to ensure the neural network is deployed correctly to how we validate safety in the software-defined vehicle era,” he said.
and operates without such critical errors. This validation has to happen “It allows us to explore millions of edge cases, stress-test our systems
in real time, and it’s central to making perception-based AI models against real-world uncertainties, and ensure that safety-critical behav-
viable for automotive use.” ior is not only designed but continuously verified and improved. That’s
Sulkes was referring to the Hailo-8 AI accelerator, which has Hailo’s a game-changer for automotive safety.”
lauded memory protection scheme for the neural network core, allow-
ing real-time detection of memory errors or data corruption during CYBERSECURITY: SUBSET OF AV SAFETY
neural network execution, ensuring that any deviation from expected The main obstacle to increasing the level of autonomy and safety, and
results is caught instantly. And like any other piece of processor hard- making it available to consumers, is cost. Almost all of these new func-
ware, there are logic protection mechanisms that monitor the integrity tions—automated lane changes, pedestrian recognition, adaptive cruise
of processing operations within the core. control, and the like—are powered by AI models, which must work
reliably, in real time, and within the strict thermal and power budgets
WHAT DOES SOFTWARE DO FOR SAFETY? of a vehicle. Furthermore, the integrity and safety of these AI-driven
While component-level safety remains essential for autonomous systems are critically dependent on robust cybersecurity. As Synopsys’s
driving, the software-defined approach is what has enabled the current DiGiuseppe noted, “A malicious cybersecurity attack—whether it’s
level of autonomy to be achieved. “It [software] brings a lot to the through system hijacking or tampering with safety mechanisms—can
table quite evidently,” said Harald Barth, product marketing manager directly affect the safety operation of the car.”
at Valeo. “Software enables us to uphold our safety standards across Industry players are taking various approaches as they build trust
every phase, from design and validation to deployment and continuous and embed complementary modules within the hardware setup. “Each
operation. It is what makes monitoring real-time. It’s this software- OEM may have its own interpretation of what secure means, depend-
defined layer that makes our hardware safer, more resilient, and truly ing on regional regulations or internal policies, so ensuring security
road-ready.” isn’t a one-size-fits-all approach,” Hailo’s Sulkes said. “We adjust and
Innoviz’s Hofstetter offered an example: “When your optics enhance our security mechanisms to match the specific demands of
engineers and perception developers are in the same room, you can each project.”
solve problems collaboratively instead of working around them. The He added that Hailo’s role, “as the AI accelerator guys, is to work
perception teams might point out that a small change in scan pattern alongside the main system-on-chip. That means security has to be
could make object detection far more reliable, or the optics team might built into how we communicate with that host. We use secure boot,
tweak the hardware to simplify downstream processing. It’s that kind authenticated communication, and other standard cybersecurity prac-
of synergy that improves performance and accelerates development tices, and automakers often build on top of them.”
timelines. That’s what we did in our program with BMW. We didn’t Progression through the levels of autonomy is multifaceted, and dis-
just supply the LiDAR hardware; we also developed the perception cussions over safety and security must not cease. “You have to embed
software.” both safety and security measures from Day 1; you can’t treat them as
Barth described the key role of digital twins in automotive safety, add-ons,” Hofstetter said. “There’s no magic powder you can sprinkle at
notably in the context of Valeo’s recent partnership with AWS and the end to make a product secure or safe. It takes coordinated planning,
Applied Intuition to create a cloud-native validation pipeline that design, and testing across teams from the very beginning.” ■
enhances test coverage and enables secure over-the-air updates to con-
tinuously improve system safety throughout the vehicle’s lifecycle. “A Saumitra Jagdale is a contributing writer for EE Times Europe.

JUNE 2025 | www.eetimes.eu