Prepare for the next next jump.
Qualcomm Technologies, Inc. (QTI) has been researching, developing and shipping artificial intelligence acceleration on their Snapdragon processors for the past decade. The latest artificial intelligence technology in QTI is born from smartphones. It leverages its expertise and leadership in performance and performance/watts, then scales it up and rolls it out into other Qualcomm products, from smartphones to cars and data centers. Qualcomm will soon be hosting the annual Snapdragon Summit in the beautiful city of Maui, Hawaii, and we expect this event will be peppered with new potential for AI. After all, Qualcomm AI Research recently promoted eightfirst AI‘, so there are good reasons to expect something exciting to come.
To understand the directions Qualcomm might take, we thought it might make sense to review what Qualcomm has already developed and see what we can glean from its approach. While mobile users of Snapdragon-based devices are accustomed to a premium experience, they are probably unaware of the number of such practical and fun features that are enabled by AI.
Let’s start at the beginning, which, given recent Qualcomm AI engines, was relatively modest. In 2007, Qualcomm decided that adding a digital signal processor (DSP) to its mobile application chip could significantly increase performance and support modem functionality. Introducing the Snapdragon S1, Qualcomm has included a DSP to add modem support for 2G and 3G connections directly on the same chip that powers the operating system and user app. Then in 2012, the Snapdragon S4 Pro included another digital signal processor that provided support for digital audio, with up to four microphones and five speakers. While this may be overkill for a mobile phone, Qualcomm engineers have gained experience and a deeper understanding of DSP and realize that the concept can be greatly expanded.
In 2015, Qualcomm took a huge step forward, announcing that the new Snapdragon 820 would include Qualcomm’s first dedicated mobile AI engine to support 4G communications, imaging, voice, and sensor operations. Users didn’t know they were “working on AI”. They only knew that their photos were better when the AI was at their best.
The journey to true on-device AI processing is starting to build momentum, especially with the addition of a tensor accelerator to the Qualcomm AI Engine in the Snapdragon 855 in 2018, along with 5G, which in itself requires significant AI processing both at the base stations and on the device. Then with 865 a year later, Qualcomm expanded its on-device AI use cases to include AI Imaging, AI Video, AI Speech, and an Always-on sensing hub, now powered by two stress outputs.
Finally, in 2021, 7 . was releasedThe tenth The new generation Qualcomm AI Engine has been announced in the Snapdragon 8 Gen 1 mobile platform. Qualcomm has introduced a new Hexagon processor architecture called the Embedded AI Accelerator Architecture. It has added support for AI gaming, antenna shifting, and always-sensing camera use cases enabled by Qualcomm AI Engine. In 2022, Qualcomm integrated the various components of its software into the Qualcomm AI Stack, laying the foundation for the industry’s software development kits (SDKs).
While Qualcomm will be quiet as usual ahead of the upcoming Snapdragon Summit, we can speculate on what we might hear in November. First, we believe AI will emerge as a pervasive topic at this event, with AI making its mark with great use cases such as real-time language translation in audio. Xiaomi has already demonstrated this ability, and we can expect others to follow suit. We wouldn’t be surprised to see more tensor cores and technologies that can improve support for switch networks for natural language processing and the next big thing in AI.
Artificial intelligence is becoming ubiquitous and a bit transparent. As a new method of programming, AI unlocks solutions to previously unsolvable problems and new capabilities. Transformation is incredibly impressive on mobile devices. Qualcomm AI Research has begun enabling on-device AI solutions within the power, size, and cost constraints that every mobile phone manufacturer must live with.
It will be an interesting trip.
Disclosures: This article expresses the opinions of the authors, and should not be taken as advice to buy from or invest in the companies mentioned. Cambrian AI Research improves to have as many, if not most, semiconductor companies as our clients, including Blaize, Cerebras, D-Matrix, Esperanto, FuriosaAI, Graphcore, GML, IBM, Intel, Mythic, NVIDIA, Qualcomm Technologies, Si-Five, SiMa.ai, Synopsys, and Tenstorrent. We do not have investment positions in any of the companies mentioned in this article and do not plan to start any of them in the near future. For more information, please visit our website at https://cambrian-AI.com.