The smartwatch is a wearables product that has grown and evolved rapidly since the first devices came to market in 2013. Initially, it was introduced as an accessory to the smartphone, but smartwatches have been increasingly used as standalone, autonomous communications devices. It is now a rapidly growing market, with over 68 million units of smartwatches sold in 2020, more than twice the sales of 2017.
This growth in sales is matched by the growing expectations from consumers. Health and fitness tracking still represents the most compelling use case for smartwatches. However, use cases are expanding across navigation (including graphics rich navigation guidance), location monitoring, mobile video calls, music, contactless payments and acting as a ‘digital key’ to buildings and cars. The expanding use cases require IP and solutions that provide the low power consumption and low area needed for the smartwatch form factor. At the same time, they need to support a wide range of ‘always-on’, high-performance use cases.
Due to the expanding use cases on smartwatches, devices have adopted some new technologies and features that support AI and machine learning (ML) workloads. Smartwatches are making use of the range of sensors to collect health and medical data from users, such as ECG, electrodermal, oxygen, pressure, vibration, and movement. Moreover, audio sensors are being used to provide intelligent personal assistants through keyword spotting and speech-to-text conversion. Due to the highly personal nature of the data that is being collected, it is likely that much of the processing needs to happen on the device. This means more ML processing power on smartwatches.
How to move ML processing to the ‘edge’ (on the smartwatch device) is a topic that Arm will be exploring at DevSummit. There will be a technical panel session featuring speakers from Qualcomm, Ambiq and B-Secur. The panel session considers the role of different hardware, operating system and software factors that are enabling intelligence for smartwatch use cases. It will also reflect on the technologies that are already providing the compute and ML capabilities needed for the smartwatches of the future. Many of these technologies are based on Arm IP and solutions.
When looking at Arm IP in smartwatches, the Arm Cortex-M range of CPUs is a good place to start. Cortex-M CPUs are being increasingly used as sensor hubs for ‘always-on’ processing in smartwatch devices. Traditionally, Arm’s smartwatch partners have used Cortex-M4 or even Cortex-M0 for this task, which consists of interfacing with the different sensors that a smartwatch uses. These include accelerometers, gyroscopes, GPS, oxygen, skin temperature and heart rate sensors.
However, one of the big Arm developments for the smartwatch market has been the introduction of Helium – Arm’s vector processing technology – in Cortex-M55, and the new Ethos-U uNPU product line, including the Ethos-U55 uNPU. Due to the growing importance of ML workloads and the fact Helium improves signal processing and ML performance by up to 10x, the Cortex-M55 is of special interest to the smartwatch market. Moreover, the Ethos-U55 also plays an integral role in smartwatches, improved processing of dedicated health use-cases, for example. The uNPU also provides improved power efficiency for audio processing, as well as image-processing applications. Image processing is likely to be a growing trend among smartwatches with more camera-enabled devices coming to market.
Moving some of the processing from the main Cortex-A subsystem to a pair of low-power IPs in M55 and U55 is important. This significantly improves battery life and saves power, while continuing to improve the ML capabilities of smartwatch devices. Indeed, a smartwatch runs in ambient or ‘always-on’ mode most of the time. This is where Cortex-M55 and Ethos-U55 come into their own, providing sensor processing, ML analysis and driving the ‘always-on’ display when the user has a quick glance at their watch.
However, future premium smartwatches still need more powerful Cortex-A CPUs and even Mali GPUs for the wide range of new use cases and more complex compute workloads on the device. This has been demonstrated through Samsung Electronics announcing its new Exynos W920, which is the first ever 5-nanometer (nm) processor for the wearables market. Not only does the processor adopt the Cortex-M55 for ambient compute to reduce display power consumption, but the Exynos W920 packs in two Arm Cortex-A55 CPU cores for high performance and power efficient applications processing. It also integrates the sub-premium Arm Mali-G68 GPU (which was mentioned in this blog about the Arm Mali-G78 GPU) to help boost CPU performance by 20 percent and add 10 times better graphics performance compared to the predecessor. Cortex-A55 helps to provide faster launch times for the wide range of new applications that are now available on smartwatches. Meanwhile, Mali-G68 enables a more interactive 3D graphical user interface (GUI) on the device.
As Harry Cho, vice president of System LSI marketing at Samsung Electronics, said:
“With the Exynos W920, future wearables are able to run applications with visually appealing user interfaces and more responsive user experiences while keeping you connected on the go with fast LTE.”
The Exynos W920 supports a new unified wearable platform after Samsung collaborated with Google Wear OS, and has been applied first to the newly released Galaxy Watch 4 model.
The adoption of different Arm IP that fulfill different tasks and workloads highlights the importance of a system-wide solution approach to smartwatch SoC design. In May 2021, Arm demonstrated how our new Total Compute solutions take a holistic system-wide optimization approach across hardware IP, physical IP, software, tools, and standards. This provides the widest choice to Arm’s partners to meet diverse use cases and cost points across all consumer device market segments.
For wearables, Total Compute solutions consisting of ‘LITTLE’ Cortex-A CPU configurations, Cortex-M CPUs, Ethos u-NPUs and entry-level or mainstream Mali GPUs. These are all aligned together as part of an SoC design could be the most appropriate for the market. These ‘efficiency Total Compute solutions’ offer ultra-scalability to achieve best-in-class cost efficiency – a perfect fit for the wearables segment that requires performance in a design which is power and area efficient. Moreover, they also offer a performance boost for the range of AI and ML workloads emerging on smartwatches.
Through the Total Compute solutions, partners can execute specialized AI workloads for various use cases in different power and silicon cost constraints. The different sets of IP provide specialized and AI compute capabilities. For example, the Mali GPUs offer mixed precision capability for image enhancement. Then, there are also the Cortex-M55, Ethos-U55, and even the Arm Ethos-U65 NPU that are all specialized for ‘always-on’ ML use cases.
Moreover, the application ecosystem on smartwatches continues to grow. Through the ‘developer access’ element of Total Compute, this means it has never been easier for developers to develop, debug, optimise, and port applications on Arm-based devices. Through the Total Compute solutions, developers benefit from this ease of use, but also gain quicker access to greater performance across the solution stack so they can make improvements to their applications.
Smartwatches are an exciting growing market, with devices continuing to evolve as more use cases develop. This is leading to demand for greater processor power to enable AI and ML workloads and security tasks at the ‘edge’. At Arm, we are offering IP and solutions that address this increasing demand for ‘always-on’, high-performance use cases. At the same time, we are enabling a low power consumption, low socket area design for the smartwatch form factor. We are continuing to work with partners to address some of the SoC design challenges associated with the smartwatch market, particularly more processor power within the limitations of a smartwatch form factor.
At the DevSummit technical panel session ‘How to move ML processing to the ‘edge’’, Arm and the panel of smartwatch partners focus on ML processing on the device and how we are working to address this challenge by considering some of the key questions about the topic. Visit the Arm DevSummit website for further information.