Multi-core Processors in Automotive ADAS Systems
by Embedded Office (comments: 0)
From advanced sensors to artificial intelligence, motor vehicles of all types are rapidly becoming hosts to the latest electronics technology. In this post, we examine the reasons why the latest automotive systems need multi-core processors.
Modern automotive systems and multi-core processors
Over recent years, the transport manufacturing industry has seen a wide proliferation of electronics know-how and hardware expertise. Some notable examples include improvements to electrical and electronic power systems, as well as refined telematics for insurance costing and route optimisation. Today, even though motor vehicles feature more electronics than ever, there is more in the pipeline. Control circuits and advanced driver-assistance systems (ADAS) are set to become part of standard vehicle specifications, for instance, instead of being costly additional options.
Collectively, these advanced new features all require the latest-generation hardware to provide the necessary processing power and support the high levels of specification.
Currently, the main challenges posed by the take-up of electronics systems in the automotive manufacturing industry are those connected with safety and security. The reliability of self-driving cars is one topical example; recent media reports have focussed on their performance and reliability. There have been accidents and a collision with a pedestrian.
Given the demands of the open road, the often random hazards and – especially – the challenges presented by night vision, safety-critical software applications and neural networks require enhanced computing power. To meet these needs, the latest multi-core processor hardware enables designers to meets the necessary standards and develop complex motoring systems that can deal with numerous complex variables in real time.
The fields of artificial intelligence and machine learning techniques have seen astonishing growth over recent years; both are set to make a significant contribution to automotive ADAS systems. Into the bargain, experts predict that neural networks will act in tandem with software applications and multi-core processors to fuel the pace of ADAS as well as other motoring advancements.
Specifically, some of the exciting new developments that will make a major contribution to in-car technology and the driving experience over coming years include:
Cameras able to stream video at 4K resolution.
Multiple video stream integration to provide 360° monitoring.
Enhanced parking, navigation and collision detection systems featuring miniature range sensors with phased array radars and Light Detection and Ranging (LiDAR).
Custom embedded hardware, initially in high-end vehicles.
In-car Wi-Fi hotspots to provide in-vehicle infotainment (IVI).
Hypervisors to control and manage the increasing complexity and number of multi-core solutions. Vendors of automotive compatible hypervisors currently include Blackberry QNX, Wind River VxWorks, Green Hills Integrity and the embedded hypervisor from Mentor Graphics. Finally, Lynx Technologies offer a secure separation kernel hypervisor that also makes use of virtualisation techniques.
As the above technologies become more mainstream, economies of scale could well lead to further cost reductions and, consequently, wide-scale adoption.
Even automotive wiring is changing considerably. In the future, not all automotive communication will be hard wired. Wireless communication for Wi-Fi access point support and cellular links to the outside world are becoming commonplace, but there is more on the way. Vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I) will improve ADAS safety and performance, while also providing self-driving cars with more comprehensive and accurate levels of real-time environmental data.
Collectively referred to as V2X (vehicle to everything), V2V and V2I testing is currently in progress in selected locations around the world. Such projects aim to develop and implement a smart-city approach, facilitating information from cross traffic detection and smart stoplights to finding parking spaces or tracking vehicles.
Advanced IVI and ADAS work better with improved driving displays. Consequently, organic LED screens (OLEDs) are making their way into prototype cars, with characteristically curved faces that give improved views of instrumentation. Head-up displays (HUDs) are also becoming more widespread and richer in features, as they are now capable of displaying more driving information in a larger space.
Safety and security in a single system
Security affects all aspects of the automotive environment, from manufacturing through to secure networked updates. Coordinating and supporting this infrastructure can be tricky in the modern automotive sector, just as it often is in the wider electronics industry. Specifically, developers and engineers face an ongoing priority task to combine safety and security in single systems.
Hypervisors became the preferred solution to address the above requirements and to allow the effective management of system assemblies and components when, by design, safety-critical and security-critical functions have to co-exist with IVI and other non-critical functionality. In other words, hypervisors allow the partitioning of virtual machines (VMs) and, similarly, the division of safety and security certifications. As a result, non-critical components do not have to undergo the higher level, time-consuming and costly accreditation processes that are necessary with critical systems.
Type-1 hypervisors have fast become an established norm. In contrast, electronics industry bosses and experts believe that type-0 hypervisors (such as LynxSecure) work better to meet the need to separate domains with differing levels of security, e.g. with ADAS and IVI. The difference between type-1 and type-0 hypervisors centres on their processor hardware. Type-0 devices require virtualisation support from peripherals as the hypervisor codebase is minimal and, therefore, less vulnerable to hacker attacks.
Having reviewed why multi-core architectures are becoming more popular in the automotive industries to fulfil the demand for processing power and functionality, one should be aware that hypervisors have limitations. Although multi-core processor technology offers security with hypervisors, it does not necessarily always deliver systems that are faster than their single-core predecessors. When designing a system for top performance, therefore, it is imperative to take an overall view of the system under development, starting at the design stage. The aim is to achieve the most appropriate architecture for the system concerned, i.e. with optimal distribution of the systems on the individual cores.
In this sense, there is neither a definitive "right architecture" nor a distribution of threads that suits every process, application and system. Consequently, it is useful to consider the following points to attain the best balance possible:
Additional cores tend to give progressively diminishing returns in performance.
Existing single core application architecture is not automatically compatible with multi-core hardware. For this reason, a new analysis of the data flow is necessary.
Wherever possible, it is best to minimise the number of positions where core synchronisation occurs.
To illustrate the above, industry engineers quote examples where single core applications were converted to multi-core equivalents but with a resulting decrease in performance. Such cases typically required consequent redistributions of the associated sub-systems. The ensuing tasks sometimes proved less than straightforward and needed a redesign to reduce the level of integrated functionality.
Finally, recent trends suggest that the same safety and security-related car system components that are becoming more common in IVI tend to have only limited third-party support. Such a lack of continuity might affect long-term reliability of the associated software, illustrating the need for robust embedded systems and processors. It looks likely, therefore, that the segmented approach that is made possible by virtualisation and hypervisors will gain ever-increasing importance.
In this article, we have considered some emerging motor vehicle technologies and seen how in-car electronics system designers, developers and engineers seek to combine safety and security with other essential design and performance considerations. In embedded systems for the automotive industry, hypervisors enable the partitioning of differing in-car systems with varying levels of security. Type-0 hypervisors tend to be more secure and less prone to hacking; their codebase is minimal, as the peripherals of the associated system(s) necessarily provide virtualisation support. Consequently, embedded multi-core processors look set to become widely specified for ADAS and other in-vehicle systems including improved displays, communication advances and smart-city information networking.
You are interested in a free talk to an expert...
Arrange an expert meeting
We offer a free and non-committal interview with one of our experts. We can get to know each other, answer your open questions and also discuss the first requirements of your project.