Qualcomm is bringing machine learning technology to its upcoming Snapdragon 820 mobile processor to protect Android-based smartphones against malware.
Company officials argue that the sheer volume of mobile malware makes it difficult for traditional signature-based anti-virus solutions to defend against zero-day and other dangerous software. Also, the malware can morph and evolve through changes in its binary code after it's in the wild, making them even more difficult to identify and neutralize.
What the chip maker wants to do with its new anti-malware technology—Snapdragon Smart Protect, announced Aug. 31—is to use machine-learning behavior analysis capabilities to detect and classify application behavior that is suspicious or unusual. The goal is to be able to analyze the malicious software before its code is changed.
At the same time, Qualcomm is putting in application programming interfaces (APIs) that can be used by both smartphone makers and security software vendors to bring real-time machine-learning analysis to their devices and software. The OEMs and software makers, which learn the reasons why specific behaviors were deemed to be possibly malicious, can share the information with the device user, who can then take whatever actions are needed to protect the device.
Qualcomm already has some mobile security software makers onboard, including Avast, AVG and Lookout, which are bringing the Snapdragon Smart Project to their offerings, according to company officials.
Machine learning—also known as deep learning, the idea that with the right technology and right algorithms, machines can learn from their experience and adapt their behavior—is a hot topic in the tech industry, where everyone from IBM to Google to Nvidia is looking to incorporate it into their offerings, from systems to software to search engines. The development of neural networks will be crucial to efforts from speech recognition to natural language processing to self-driving cars.
However, most machine-learning efforts involve large systems and a lot of processing power. Qualcomm officials through their project are looking to bring it to mobile devices.
The foundation of Snapdragon Smart Project is the company's Zeroth technology, an optimized architecture leveraging both hardware and software that is designed to take advantage of the processors' heterogeneous compute capabilities and deep-learning algorithms. It's part of Qualcomm's Haven Security Solutions. Company engineers have designed the technology specifically for the mobile space and to take advantage of hardware and software features in the Snapdragon 820.
"Another advantage of Snapdragon Smart Protect is the fact that it was designed to operate primarily on a device, rather than in the cloud," officials said in a statement. "Snapdragon Smart Protect performs observation and analysis on the device in real time. So it doesn’t need to be connected, providing always-on monitoring of application behavior."
The Snapdragon 820 system-on-a-chip (SoC) is expected to be released this fall, and smartphones offering the processor—and its Snapdragon Smart Protect technology—in the first half of 2016.
The Snapdragon 820 is an important product for Qualcomm, which is the world's top supplier of mobile processors. The company took a hit earlier this year when Samsung officials decided not to use the Snapdragon 810 in its Galaxy S6 smartphone reportedly over concerns about overheating. In an effort to accelerate the release of the Snapdragon 810, officials had decided to use one of ARM's SoC designs. After Samsung's rejection, Qualcomm officials opted to return to using its own custom 64-bit ARM-based CPU design for the Snapdragon 820.
Over the past several weeks, Qualcomm has released details about features that will be in the Snapdraon 820, such as the next generation of the company's Adreno GPU and Spectra camera image signal processor (ISP) for improved graphics and image quality, and Qualcomm's Hexagon digital signal processor (DSP), where tasks that can run faster or more efficiently than on the CPU can be offloaded.