Whether connecting with friends and family, managing work tasks, or indulging in entertainment, these devices have redefined how billions worldwide live, work, and play.
In the age of AI, smartphone experiences are becoming ever more intelligent, immersive, and dynamic. Consumers are already taking advantage of generative AI workloads such as image generation and text summarization, and OEMs and developers are pushing to deliver the most compelling AI use cases. This evolution is driving unprecedented computing demands, with the rise of generative AI ushering in new possibilities for the end-user. However, meeting these demands requires ongoing hardware and software innovation, paving the way for the next era of smartphone capability.
At the core of this innovation lies the mobile system-on-chip (SoC) and underlying component technologies that are helping to define the performance of the devices, as well as enabling a range of transformative user experiences.
Central to these AI processing capabilities is the SoC architecture, which continuously adds new features and instructions to process current and future AI workloads. These new features and instructions often directly correspond with the ongoing evolution of applications that are now faster and more intelligent.
Developers are constantly looking at ways to enhance the user experience for the most popular media apps. One example of an SoC advancement that has been created to support developers is SVE2, an architectural instruction set that features across many of today’s flagship smartphones. It incorporates vector instructions to accelerate key workloads on mobile devices, such as video and image processing which leads to better quality photos and improved on-demand video through applications like Facebook, Instagram and YouTube, among others.
Looking ahead, architectural features such as Scalable Matrix Extension (SME) will help to facilitate and accelerate a variety of new generative AI use cases on the smartphone, from virtual chatbots to virtual assistants. This is achieved through SME being able to manage multiple matrix and vector operations that are crucial for advanced AI processing.
On-device generative AI
Thanks to these SoC advancements, many generative AI workloads are now a common part of the modern smartphone experience, from text generation to advanced camera effects and filters.
Most of these AI inference workloads can be processed at the edge – on the device. This is largely thanks to the computing capabilities of today’s smartphones, as well as the large language models (LLMs) behind generative AI becoming smaller and more efficient. Meta’s new Llama 3.2 LLMs are great examples, as they enable fundamental text-based generative AI workloads that can run on the smartphone device.
For the industry, the more AI being processed at the edge, the more power that is saved from data travelling to and from the cloud, leading to energy and cost savings. It also means less latency and far faster, more responsive user experiences. In fact, running the new Llama 3.2 3B LLM on Arm-powered mobile devices leads to a 5x improvement in prompt processing and 3x improvement in token generation.
While the range of generative AI workloads being processed on the smartphone is already impressive, we are still scratching the surface of its potential. The future is likely to see new AI-based use cases running entirely on the smartphone device, including group chat and voice note summarization features and real-time voice assistants, to name a few.
The importance of IPC performance
In terms of key performance metrics for smartphones, there is a renewed focus on balancing performance and efficiency, which is best represented by Instructions per Cycle, or IPC for short. IPC is a critical metric for evaluating CPU performance, measuring the number of instructions a CPU can execute per clock cycle.
To better understand IPC, think about this analogy of a factory with robots. A factory with a high IPC will have advanced robots performing multiple tasks per assembly line step, making the factory capable of manufacturing more products in a set amount of time. Having more capable robots means fewer are needed to achieve the desired output, with efficiency savings as a result. However, less capable robots in the factory that can only perform one task per step will require a greater quantity, resulting in a low IPC and an inefficient process.
Now, applying this analogy to smartphones, it is clear that IPC performance is crucial, especially with battery life and thermal management being key factors. Better IPC means less energy is used to process computing tasks, which translates into longer battery life, less overheating and a better overall user experience. All of the most common real-world smartphone use cases, such as gaming, web browsing, video streaming and multitasking, require a strong IPC performance to ensure smoother, longer-lasting experiences.
The driving force behind smartphones
Mobile SoCs serve as the driving force behind the world's smartphones, empowering billions with seamless access to essential daily experiences. As AI and advanced computing evolves at an unprecedented pace, the underlying hardware and software of mobile SoCs must rise to meet these new demands.
A bold focus on cutting-edge AI capabilities, combined with a visionary balance of performance and efficiency, will be key to unlocking the future of mobile computing. This will usher in smoother, longer-lasting, more immersive, and profoundly intelligent experiences that redefine how people interact with technology.
Author details: Chris Bergey leads Arm’s client line of business where he is responsible for defining the compute platform that shapes user experiences in the smartphone, the Metaverse, gaming, and laptop/tablet markets.
