The Race To Autonomous Cars

To say that the race for autonomous transportation has been heating up would be a gross understatement. By now all, companies that aim to be leaders, or at least want a piece of the action have already established their presence. Some in a successful way, while others not so much.

This is not a race that can be won by sprinting alone. It is a team relay where partnerships have to be formed, batons passed, and every member has a key and complementary role to play in order to reach the goal.

Even before this race could start, we had to analyze the challenges facing the industry of autonomous transportation and plan the race accordingly. Here are the top three challenges:

Exploding Performance Requirements
Technology around us has seen tremendous advancements and transformations, especially in the last few decades. These advancements have come in the form of features and performance that have tended to make the human race increasingly dependent on technology. We live in a world where the thermostat in our home seems to understand our feelings better than some people.  The same is more so true in the world of automobiles.

I remember being super thrilled by the mere fact that my first car had power steering. I might be dating myself here, but power steering, really? Most millennials would be shocked that there was a time when humans used elbow grease to turn the steering wheel. Not so anymore. This and many more of the driverless features in automobiles are enabled by high performance SoCs. Performance not just in terms low latency and high bandwidth, but also in supporting heterogeneous architectures and cache coherency. Heterogeneity and coherency have become the cornerstones in high performance SoCs. The need for heterogeneous computing arises from the fact that various types of computing engines, ranging from CPU, DSP/Accel, to GPU, are adept at performing different tasks during the lifecycle of an autonomous vehicle. Small data vs. big data, complex data vs. simple data, structured data vs. parallel data. All require different engines to execute the functions efficiently, while working hand in hand and communicating between each other. This communication and sharing of data has latency and performance implications, which is why hardware coherency is of utmost importance for efficient operation.

Real-time processing of sensors
One of the key requirements in autonomous driving is reaction time. To ensure safety of a system, the time from the detection of a fault plus the time for the system to achieve a safe state is determined by one of the system goals. The plethora of sensors in today’s cars (camera, video, radar, LiDAR, ultrasonics), puts stringent requirements on the real-time capabilities of the SoC.

The CPU, interconnect, and other critical IPs of the SoC all need to handle the communication at deterministic low latencies for certain datapaths and messages. The other challenge imposed by the many real-time sensors is the sheer amount of data that needs to be collected, processed, and analyzed. Unless the car has a magic 8-ball, and one that actually works, it does not know which sensor to give priority to. Hence, the SoC has to handle the massive amount of data that needs to be processed before it knows where the obstacle might me.

The real challenge is actually doubled since the SoCs have to handle deterministic low latencies while simultaneously working on massive amounts of data. But what’s more important than having low latency paths in the design is to make sure the paths are available when required. There is no point in having a short cut if it is clogged by traffic or a tree branch lay right across it, making it unusable. They key is, hence, to have a dynamic, robust end-to-end QoS mechanism layered on top of the interconnect to enable these solutions during the time of need.

Ultra high safety and security
The latest and greatest driverless features come at the cost of high dependence on the electronics of the automobile. Monitoring of the surroundings and reaction to the events were all under human control, but with the advent of autonomous driving the machine becomes the fallback. All the trust we put on the SoCs puts a hard requirement on the safety and security aspects of the systems. Functional safety deals with ensuring the correct operation of the system while managing hardware failure, errors, and changes in environment. Security, on the other hand, deals with making sure the car cannot be hacked into and controlled externally.

The ISO 26262 standard, established in 2011, recommends a set of requirements to cover both systematic faults and random hardware faults. Ensuring the highest level of Automotive Safety Integrity Level (ASIL) means investing time and resources in making sure all aspects of the development process are well documented while still ensuring the cutting-edge safety and security features are implemented. The obvious cost, other than time and resources, is silicon area overhead and care should be taken in coming up with novel ideas to keep that in check.

Running the race
As I said earlier, this is a relay race, and that requires teamwork. NetSpeed Systems and Imagination Technologies have teamed up to build cutting-edge autonomous transportation SoCs of the future. The complimentary solutions that each brought to the table make it an obvious choice to join hands in this challenge to make the next-generation autonomous transportation SoC. This goes well beyond just building an SoC. It involves developing an ecosystem and helping to share this joint effort with the industry.

The first joint webinar will discuss NetSpeed and Imagination’s partnership to build Mobileye’s next-generation EyeQ5, with an 8X performance increase over the current EyeQ4. The Webinar, entitled “Alexa, can you help me build a better SoC?”, will be held on Sept. 28. Details can be found here.