Artificial Intelligence (AI) is a big deal  in data centres today with companies such as NVIDIA, Intel, Google, Amazon, Microsoft, Baidu and Alibaba all investing heavily. Client devices such as smartphones, however, are just beginning their AI journey. Unlike data centres, smartphones are constrained by computing power, power consumption, memory footprint and network bandwidth and cannot run full-fledged training and inference machine learning models. This is beginning to change as client-optimised AI frameworks such as TensorFlow Lite, Caffe, Caffe 2, ONNX and Android Neural Networks APIs become available to developers.

The smartphone industry has recognised the importance of AI and started applying AI in areas including imaging, security and voice. Currently AI in smartphone processors is driven by Apple, HiSilicon, Qualcomm and Samsung LSI. Arm, CEVA, MediaTek, UNISOC (Spreadtrum & RDA), Cadence, Synopsys, Imagination, Cambircon and other stakeholders are also trying to claim their stake in this fast growing area.

Strategy Analytics has started tracking the smartphone AI chip market share and found that AI chip shipments grew three-fold in Q1 2018. Apple A11 Bionic was the top-selling AI chip in Q1 2018, by our estimates. The current smartphone AI chip market is divided in two camps primarily – dedicated AI cores (Apple, HiSilicon) and repurposed AI cores (Qualcomm and Samsung LSI). We believe that the market will consolidate around dedicate AI core (integrated in AP) approach. Key smartphone AI chips include Apple A11 Bionic, HiSilicon Kirin 970, Qualcomm Snapdragon 845 / 835 and Samsung LSI Exynos 9810.

Yesterday Samsung announced its new Galaxy Note 9 flagship smartphone, which featured Qualcomm Snapdragon 845 / Exynos 9810 processor. We note that the Samsung Galaxy S9 and S9 Plus also use the Snapdragon 845 / Exynos 9810 processor. Samsung included AI features such as scene detection and flaw detection to alert users to take significantly better photos. Strategy Analytics reviewed Snapdragon 845 announcement and its AI features back in January 2018. Qualcomm currently uses its CPU, Adreno GPU and Hexagon GPU combination to enable AI use cases such as image recognition and object detection. We note that Huawei and LG also implemented AI camera enhancements in their recent flagship devices.

While AI continues to add value to smartphones, some key questions remain.

•  Will AI positively influence the smartphone chip pricing, which is pressured by increased competition and lack of differentiation? Recent earnings call comments from TSMC suggest AI functionality is likely to add to die area and thus pricing. If this is the case we can see a healthy smartphone chip pricing trend in the coming years.
•  Will AI be limited to flagships and premier tier only? That seems to be the case, at least in the near term. However, new neural IP solutions from Arm, CEVA and others could bring AI to lower price tiers.
•  Which AI architecture will win in smartphones? Arm, the market leader in smartphone CPU and GPU IPs, is playing catch up in the AI IP although the company enables AI use cases with its existing CPU and GPU IPs. The battle for AI chip leadership will be watched closely as new architectures such as MIPS and RISC-V also appeared to be gathering momentum in non-smartphone areas such as data centers and automotive.
•  Will AI expand to other smartphone component areas such as RF, PMIC and modem? Which company is well positioned to capture this growth?

Strategy Analytics believes that it is early days for AI in smartphones. AI will ignite the architecture innovation in smartphone APs in the coming days. Strategy Analytics will be watching this space closely.