Apple unveiled its A11 Bionic SoC (System-On-Chip) with the launch of its new iPhone. The A11 chip has an embedded AI (Artificial Intelligence) unit that Apple calls a Neural Engine.

At IFA 2017 earlier in the month, Huawei showcased its latest Kirin 970 smartphone SoC with what it calls a Neural Processing Unit (NPU), for AI processing. While Huawei was the first to unveil a SoC with dedicated AI segment, Apple became the first to launch a smartphone with a specific use-case in its FaceID function.

Qualcomm the largest smartphone SoC player has taken a slightly different approach for AI computation in smartphones—adapt existing technology to utilize its strengths. Instead of dedicated silicon space on-board, it has developed a Neural Processing Engine, which is an SDK (Software Development Kit) that allows developers to optimize apps to run different AI applications on the Snapdragon 600 and 800 series processors. Developers can choose the core [processor core] of choice relative to the power performance profile they want for the application.

Need for on-board AI in Smartphone

The smartphone SoC segment is moving away from simply adding a faster CPU and more cores in the chip. Chipmakers are integrating new technologies into their SoCs to differentiate them in the market. For example, in the initial stages of dual cameras in smartphones, the device needed to have separate ISPs for processing the images. Now most of the mid and high-end SoCs have in-built support for the same. The integration of AI in chipsets sets a new industry standard.

Over the years Cloud AI has seen broad application in multiple areas. On-board AI will complement Cloud AI with improved latency, stability and privacy for users. While Cloud AI was limited in its capabilities by the user’s interaction, on-board AI can leverage the large set of data collected by sensors on the device in a real-time, scenario-specific and personalized way. The evolution of smartphone SoCs with enhanced processing power and integrated AI will make devices more cognizant of user needs, delivering personalized services and making smartphones even smarter.

Apart from being faster and more efficient, on-board AI can also effectively counter the increasing security threat smartphones are facing by:

• Real-time malware detection
• Recognizing user behavior to identify if the phone is being misused
• Spam detection over emails and other apps

As was discussed at the Apple keynote event, the FaceIDs will be stored and processed using on-board neural engine to protect and secure user data from any threat.

What is next and what to expect?

Both Huawei and Apple announced that the AI platforms will be open for application developers, Qualcomm also recently released the SDK for developers. We expect existing smartphone apps like Facebook, Snapchat and others to become smarter by leveraging the technology. The biggest advancement would be in the way user interacts with virtual assistants, as it become less dependent on a cloud back-end. Virtual assistants will become smarter by analyzing and learning user behavior, thereby uniquely serving each user according to their need. This could potentially help virtual assistants take the leap and become a main-stream medium of interaction between the user and device.

Augmented Reality (AR) is another major application that will benefit from on-board implementation of AI.

Another potential application is to extend battery life of devices by analyzing its usage and powering down all or part of its capabilities without affecting the user experience.

As discussed earlier opening the platform for developers will result in other innovative application of this technology.

Since Apple, Huawei and Qualcomm all have taken a leap in AI technology we expect Samsung to follow next in its Exynos line of smartphone SoCs. We also expect Qualcomm to embed a dedicated AI unit in its future generation of Snapdragon SoCs. MediaTek and Spreadtrum, the other major smartphone SoC players will take some more time to follow since they cater to the mid and low end of the smartphone market that will naturally see a delay in the adoption of this technology. They will have to depend on partnership with third party AI chip providers to compete in the segment, should they decide to.