Ten years ago, I would often say that the car is becoming a "smartphone on wheels." Since today's cars generate more data than ever before, today the phrase "server on wheels" is probably more fitting. It tells us something about the enormous real-time processing power required to run
increasingly connected and autonomous cars.
But unlike smartphones, most of this data needs to be used and processed in the vehicle. To do this, automakers are looking to edge computing architecture to bring data capture, control and storage in-vehicle.
With this in mind, we can begin to imagine the car as an "edge data center on wheels."
Why are automakers rejecting the cloud in favor of edge data centers?
There are certain factors relevant to driving that mean transferring data to the cloud is impractical – and even dangerous. Edge data centers process time-sensitive data at the point of origin, enabling faster delivery to the end device that needs it.
The biggest issue is latency. If you consider that a self-driving car generates roughly 1 GB of data per second, this gives you some idea of how much information would need to be processed and returned for a quick trip to the grocery store.
Of course, we haven't reached the era of autonomous vehicles being the norm. But even for the average connected car, transferring data elsewhere doesn't work when it needs to make critical decisions that could mean the difference between getting stuck in a traffic jam and arriving to your destination on time. Or, in the case of EV battery information, running out off power on a highway.
And when it comes to self-driving cars, latency on the road is life or death.
There is also a big cost involved in sending all that data to the cloud. Edge data centers in cars could work out cheaper as they cost less to deploy and would already have the infrastructure to be housed inside.
The cloud nonetheless still has a role to play. Data that is not as time-critical could be fed to the cloud for processing and analysis later. In this way, edge data centers offer an efficient hybrid solution to the significant latency and cost challenges that connected, and eventually autonomous, cars will present.
Increasingly intelligent cars require increasingly powerful processers. We are talking magnitude tera operations per second (TOPS) of about two TOPS for Level 2 autonomy and 4,000+TOPS for Level 5.
Throw the demands of an EV system into the mix – where every milliwatt of power needs to be optimized to save energy – and the requirements for data processing swell further.
There are various applications, too, where data needs to be handled differently. Immersive in-vehicle infotainment systems necessitate lightning-fast performance. Advanced driver assistance systems (ADAS) require sensors that can identify a range of road conditions. As access to 5G widens, new solutions will be needed to make the most of improved connectivity across the board.
Technology such as radar that enables your average parking sensor -- and lidar, which can be used to predict if a pedestrian might cross a road, also have different handling requirements from each other.
To be processed locally, all this needs to be scaled down to a single SoC (system on a chip).
The chip will then need to be able to cope with all the challenges of being inside a moving vehicle -- with zero room for error. Reliability, safety, and anti-defect design are therefore paramount in the development of chip design and manufacturing for automotive applications.
All of these considerations must be taken into account across various stages of chip fabrication, from the process technology and memory design all the way to final testing, to ensure that right portfolio of solutions are created to power the connected cars of today -- and tomorrow.