Semiconductor industry sales regularly go through their ups and downs, and chipmakers are always looking for the next killer application that will stimulate new growth. They might just have found such a boon in the automotive industry: computer vision technology that will eventually enable autonomous vehicles.
An autonomous vehicle uses several cameras, along with a range of additional sensors, to process the information in its surroundings and make driving decisions in real time based on the sensor data. Such systems rely heavily on computer vision technology to extract and process the information.
These algorithms are based on deep learning technology. Deep learning is essentially an evolution of neural networks, which by their nature are computationally intensive. This is where graphics processor units (GPUs) come in. Deep learning algorithms map nicely onto GPUs for acceleration purposes. Studies have shown that GPUs accelerate these algorithms by as much as 33x compared to CPU-only systems.
So how does the growth of computer vision in the automotive market translate into GPU sales? To get a generalidea of the potential opportunity, we could build a market model based on car sales. There are more than 1.2 billion cars on roadways worldwide with about 88 million new cars sold annually. In 2014, the top luxury brands sold about 5.9 million units worldwide. Luxury cars are the main segment that will utilize some sort of computer vision system going forward, and Tractica believes that computer vision capabilities will be fairly pervasive in luxury cars within the next several years.
If we assume that you’ll need one GPU chip per car, that gives you 5.9 million chip sales. Multiply that by, let’s say, six systems in the future car and that gives you 35.4 million chips. As the systems get sophisticated and the number of sensors increases, the number of GPU chips per system will increase. Now the numbers are ramping up at a rapid pace and are looking pretty good, especially if you are a GPU chip company.
Granted, computer vision technology and autonomous vehicles are both in their early stages, but we are already seeing demonstrations from Audi, BMW, and others. Daimler, for example, recently licensed its self-driving truck in Nevada and Google has been testing its self-driving car for some time. It is just a matter of time before the computer takes the wheel, and such a transition should bode very well for semiconductor companies.