ARM introduces two new AI processors

It is fair to say that ARM’s brand is not as recognizable as that of other chip maker giants like Qualcomm or Intel, at least for the average consumer. But the way the company designs its chips is integrated in practically every central processing unit (CPU) present in today’s mobile devices (smartphones and tablets), cameras, and even Internet of Things (IoT) ready gadgets. 

The last few years have since seen ARM shift into developing graphics processor units (GPUs), and just this week, the firm has revealed that it is now looking to offer its manufacturing partners machine learning, as well as dedicated detection chips, by way of its Project Trillium. The overall aim of the project, of course, is to try to establish ARM’s machine learning processors as the default standard for the machine learning platform for wireless and IoT.

As a direct result of the Trillium initiative, ARM is unveiling both a machine learning chip that should handle general artificial intelligence tasks, and a new generation object detection processor that is designed to recognize human faces and human actions, in videos that are rendered in high resolution quality (even equalling full HD) and running at 60 frames per second. Those familiar with object detection processors may remember that ARM had developed its first gen chip for Hive’s smart security camera.

What is intriguing about these two new AI processors is that ARM had basically developed them using a new approach. It turns out that the company’s existing designs were not that power efficient to begin with, so the firm went to work coming up with an entirely fresh design. The result is two new chips that not only deliver in terms of efficiency, but allows for better performance overall. Although the new machine learning processors are at 3 teraops per watt, they can still provide 4.6 teraops in terms of mobile performance. And ARM is claiming that the 4.6 teraops figure is just the beginning. 

In the near future, the company believes that quite a number of original equipment manufacturers (OEMs) will be utilizing both the machine learning and object detection processors together. The object detection processor could do the initial processing, figuring out the human faces and objects in a given picture, while the machine learning chip will then handle the actual face, object, or image recognition.

For more information about ARM’s new chips, read the official press release now.