Experts from the automotive software and hardware industry met last month in a technical gathering to discuss the latest trends and developments in multi-core processing architecture for automated driving and the car of the future, making liberal use of buzzwords like real-time, high-performance, deterministic, partitioning, safety, and security.
At the Embedded Multi-Core Conference (EMCC) in Munich, Germany, the subject of discussion was the high-performance multi-core processor, which will serve as the basis for many electronic control units (ECUs). The ECU is any embedded system in the car that controls the electrical systems or subsystems; examples of ECUs include the engine control module, the surround-view park-assist system, and the infotainment head unit.
Among the speakers during the two-day event was Jon Taylor from ARM, who offered in his keynote an impressive overview of both current and future automotive trends. While autonomous driving prototypes at present are too expensive and the legal framework surrounding them may be an obstacle, the first mass-production of L4 vehicles could start by 2025. With more than 200 sensors predicted to be in the car in 2020, intelligence—in the form of custom accelerators and machine-learning processors—is moving to the edge. ARM, for example, has integrated Nvidia´s Deep Learning Accelerator software and framework in the company’s Project Trillium into its own initiatives on artificial intelligence (AI). Yet in some cases, centralized computing will be more appropriate and hybrid architectures will dominate. All told, power, scalability, packaging, and cost pressure will determine the processing capabilities of the six vehicle domains: ADAS/autonomous, cabin, body, e-powertrain, chassis, and connectivity.
As Jürgen Becker from the Karlsruhe Institute of Technology said at the event, the semiconductor industry must be evolutionary, not disruptive. Quantum computing is still far away from becoming a commercial solution, so research into the design and creation of advanced system-on-chips (SoCs) will be the key to keep silicon alive. For instance, chips from the mobile phone market integrating central processing units (CPUs), graphics processing units (GPUs), digital signal processors (DSPs), AI accelerators, and modems are increasingly popular in infotainment. However, on issues relating to safe decision-making, autonomous vehicles will make use of high-performance heterogeneous architectures, including both application processors and hard real-time processors. Even classic microcontrollers units (MCUs) are becoming multi-core, and experts like Gerhard Wirrer from Infineon highlighted during the conference the need to develop complex abstraction layers to help customers get the most out of every core.
Kai Lampka from Elektrobit also spoke about evolution—from classic ECUs to real-time and performance oriented multi-core architectures. Engineers must be ready to implement applications on a software-centric, hardware-independent approach. Legacy code from different sources is being integrated into a single SoC, with both safety-aware and safety-agnostic stacks running side by side. Microkernel-based hypervisors can exploit hardware virtualization to provide temporal and spatial separation while minimizing security attacks and safeguarding against failure surface. But several challenges remain, including how to orchestrate parallel execution, control access to resources and shared peripherals, and predict memory worst-case execution time, among other things.
Notably, software consolidation will lead to a reduction in development costs and integration efforts, and hypervisors are opening the door to open-source platforms like AUTOSAR and GENIVI. Even so, competition in the software domain is growing, and it is not clear if automotive players can align and find a standardized solution for autonomous driving. Overall, the tide of increasing complexity and the industry´s fear of open source have a big impact on security and safety certifications, which are critical but whose intended objectives may be conflicting: while synchronization with other vehicles and infrastructures enables safety, security benefits from isolation.