With government targets for carbon emissions and consumers becoming more aware of the benefits of EV’s, both in terms of cost and environmental impact, the automotive industry is at a tipping point. This change also comes at a time where advanced driver-assistance systems (ADAS) and autonomous driving are moving ever closer.
As we get ever closer to Level 3 autonomous driving and beyond, allowing drivers to take their eyes off the road and their mind off the task of driving, there is an ever-growing need for a wealth of sensors being added to vehicles to replace the driver’s eyes on the road. Increased numbers of cameras, exterior LIDAR and even interior sensors to monitor occupant safety and attention generate a wealth of data that all needs to be processed. This whole process is very challenging for one thing: power consumption.
With every watt of power used affecting the range of the vehicle, how can automakers deliver the features of the car of the future without impacting the range of that car?
One simple way to do this is with improving efficiencies in the systems themselves. By developing lower powered sensors that can still collect data we can make an immediate saving in power, but all the data collected needs to go somewhere and with more data comes an increasing burden on the System-on-Chips (SoCs) to process it. High level concept cars for autonomous driving and cutting edge ADAS systems rely on a desktop grade GPU solution that is water cooled to be able to brute force their way through these dense and complicated calculations. But, when you are limited in your battery capacity, a 350w GPU is not going to cut it.
One of the best ways around the issue is to design from the ground up with the car as a platform in mind. EV’s have experienced an explosion of displays and rich information in the cockpit. Rather than trying to make something else work, it is much better to have higher density silicon to allow for much more efficient computation whilst also reducing the silicon area in the vehicle. With dedicated automotive aware design, comes the ability to embed functional safety built in from the ground up. As an example, the recent XS-series of Imagination GPUs for automotive feature ISO 26262 compliance. What this means is that a single GPU can actually fault check versus the need for a pure brute force duplication of for redundancy.
We have the first piece of the power saving puzzle with a highly efficient GPU. GPUs are a great solution as they provide high programmability, but as the ADAS algorithms have developed, we have seen that there are certain aspects that work far more efficiently with dedicated hardware such as neural network accelerators (NNAs). This creates a complete functionally safe solution, for processing our ADAS workload.
SoC’s are not just important in the ADAS computing powerhouses and are spread across EV’s controlling a myriad of other features. How else can we bring the power requirement down to give drivers maximum miles on the road without having to turn everything off to limp home? In any modern vehicle there are already hundreds of electronic control units (ECUs) distributed throughout the car, from electric windows to indicators and a myriad of other seemingly simple operations.
Each of these control units need its own core to control its actions multiplying the power requirement throughout the car, one or two extra ECU’s is no measurable impact, but the scale with which they are being adding adds up to a lot of wattage. - Partial consolidation of these ECU requires a consideration to the tasks being brought together and how they can remain safe and isolated.
Hardware virtualization allows several workloads to run concurrently on a single processing element for example, dials and clusters, reverse cameras, and driver monitoring. By isolating each of these workloads to a single core, multiple operating systems and functions can run simultaneously and independently. Coupled with functionally safe standards like ISO 26262 to self-check for faults in each core, its easy to see how the number of ECUs can be drastically reduced. Keeping things moving forward.
EV’s aren’t going anywhere, and neither is the industries push for a safer more autonomous driving future. From ADAS to door locks to pixels, every watt matters , and the glue that will bind EVs and autonomy they go forward is the IP vendors and silicon manufacturers who continue to develop elegant custom, low power solutions across the range of features in a vehicle.
Author details: Jamie Broome, Senior Director of Automotive Product Management, Imagination