Positioning: Qualcomm Ventures Beyond...

A few days ago, Qualcomm paper-launched their "Zeroth" processor. Designed to emulate the neural networks of the brain, Qualcomm says this product will improve machine learning and computer perception problems. But are they straying too far from their core competency in wireless and digital communication?

Neural networks and "feedforward" systems have been a growing buzzword lately. IBM has been at it for a while now with their SyNAPSE hardware and plenty of other software companies have added it as a tool for data analysis. The basic mechanism is to create a system that models biological neurons and can learn over time. Typically, a human will train the system by telling it which behavior is "right" and "wrong," although there are some tasks which are done without any human intervention. It's just like training your dog, only now your dog can answer things like "How many different types of objects are there in this picture?"

Why is neural networking a big deal?  Because despite their talent for arithmetic, typical computers don't "think" the way that we do, and computers have to expend a lot of energy to emulate easy "human" tasks. If you want a great example, look up how much computing power and camera work goes into painting the first down line for NFL games on TV.

Yeah, that one line takes 7 computers...

Qualcomm's press release seems to indicate that they want this product to become part of their fabulously successful SoCs. The company claims that it will help users train their devices for more human-like behavior. And while that sounds great on paper, I can't help but think that this is a great idea with horrible business implications. Qualcomm rode the smartphone wave due to their wireless communication products which made their offerings more appealing than any competitor's. Meanwhile, the Zeroth initiative aims to allocate space on a chip (in chip manufacturing, space=$$) dedicated to a task which would typically be performed in a centralized server. In other words, Qualcomm wants to integrate a device which will help users care less about Qualcomm's differentiating feature - the wireless radio. Finally, these learning devices should generally be able to leverage data from lots of users to improve their learning. Learning from only the local user might take longer to generate good results.

How will this Zeroeth product fare for Qualcomm? We'll have to wait and find out. Certainly I'd love to see more accurate individualized learning  in my smartphone, but I do have my reservations on this particular strategy. On the other hand, the company has been on a tear lately, introducing new products in all kinds of spaces. The next few years will either redefine Qualcomm as a broad silicon IP company, or solidify them as the big fish in mobile SoCs.