etilage

[etilage]

Operating systems for autonomous cars

The operating system is crucial part of the technology stack for ADAS functions in automobiles. Especially since we talk about a systems that define the safety of the passengers. It´s a matter of life and death, so no failures are acceptable.

A modern car has a lot of different ECU´s (Electronic Control Units), controlling various functions from the car, from the automated sunroof to the control of the electric drive motor of EVs. In order to address the increasing amount of functions and ECU´s in cars, there is a trend towards domain ECU´s, which cover all functionalities of one domain, e.g. infotainment or ADAS. In the future this trend will further intensify, which means that we can expect the different domain ECU´s to merge into one very powerful processing center that will handle different virtual domains.

When we talk about operating systems in cars, we need to be aware that there are various. They can even run on top of each other on the same hardware. A car with multiple ECU´s is quite a complex system. Especially interesting are the operating systems for the ADAS domain. Since they handle safety critical functions we need real-time operating systems (RTOS), which are intended to serve real-time applications that need to meet certain processing time requirements. Apple CarPlay and Android Automotive are used in cars, but are definitely not real-time operating systems, since their strict focus on infotainment does not require that.

For you orientation, here is a list of most prominent RTOS´s for the ADAS domain:

QNX Neutrino

WindRiver VxWorks

Green Hills Integrity®

Nvidia DriveTM OS

Mentor Nucleus® OS (Siemens)

RTLinux

[etilage]

The battle for the brains of autonomous cars

When cars move towards higher levels of autonomy, more processing power will be needed. A rule of thumb says, that going from one autonomy level to the next (see SAE´s automation level definitions) increases the need for processing power ten-fold. According to IHS, the automotive semiconductor market was worth $41.8 billion in 2018 and will reach $65.5 billion by 2025. The fastest growing segments are electronics for Advanced Driver Assist Systems (ADAS) and Hybrid/Electric Vehicles. Chipmakers will play a crucial role in the battle for the autonomous car technology stack. Especially because they see the Internet of Things as the next big growth markets, the automotive industry as one of the most interesting ones among those.

According to Gwennap, a leading analyst in this area, automotive semiconductors can be split in 4 categories:

Microcontroller for traditional features (e.g. ABS, ESP) - an area that is dominated by incumbent players like Renesas, Infineon or NXP

Wireless modem chips connecting cars to the internet - produced by companies like Qualcomm or Intel

Chips for cameras and sensors - e.g. from Mobileye, which was acquired by Intel for about $15 billion in 2017

Processing chips for AI - an area in which Nvidia, among others, is very active

There is no chip to rule them all, one chip to find them, one chip to bring them all, and in the darkness bind them. No one company will be able to offer the whole range, which makes this a very dynamic and interesting playing field. Especially processors from the latter two categories will shape the advent of autonomous cars, which is the main reason why we have seen various acquisitions in this area. Renesas just recently (03/2019) completed its acquisition of IDT (Integrated Device Technology) to enhance its capabilities in this emerging field. Interesting to observe is once again Tesla´s strategy, which is very different from traditional OEM´s. Tesla decided to develop their own AI chip (FSD Chip; Full Self-Driving Chip). According to Elon Musk it allows them to “run the neural network at a fundamental, bare metal level”, which improves performance significantly. This reminds of Apple´s strategy to develop their own chip for their iPhone and Google´s development of TPU (Tensor Processing Unit), a custom-designed machine learning ASIC (Application Specific Integrated Circuit). While Tesla´s strategy might me risky, it could give them a decisive competitive edge. Traditional OEM´s prefer to work with companies like Nvidia or Intel, which benefit from scale advantages as an independent party with access to multiple OEM´s. Carmakers tend to count on suppliers in these type of situations in order to minimize their own R&D investments, focus on their core competences and stay open for any potential technological advancements.

Source: Tesla

Only future will tell which strategy is the winning one. It will continue to be an exciting area. One of many in the upcoming years when it comes to defining the leadership roles in autonomous driving technology. The chips are just one piece of the technology stack. A lot more battles will be fought when it comes to the operating systems, middleware, application specific software, sensors etc. Various players try to build open platforms with the objective to maneuver them into a strong position of controlling parts of the technology stack while leaving the system open for innovation for contributions from other parties.