Share
By Kevork Kechichian, Executive Vice President, Solutions Engineering, Arm
Silicon technologies define the billions of consumer technology products we use today. And they continue to evolve, particularly in response to the ongoing AI revolution, where computing demands have never been greater. This is leading the industry to embrace new silicon design approaches—featuring custom silicon, compute subsystems (CSS) and chiplets—that are set to define the next decade of technology innovation.
The Rise of Mobile Chipsets
The relentless push for silicon performance started with Moore’s Law, which predicted that the number of transistors would double approximately every two years. This led to the development of increasingly powerful chipsets and continuous miniaturization of transistors and other components.
The need for more performance also fueled the rise of mobile chips adopted by the world’s first cell phones, like the hugely successful Arm-powered Nokia G110 GSM of the mid-1990s. Semiconductor companies took notice and developed chipsets that were increasingly optimized for mobile devices.
As these mobile chipsets evolved, there were continuous improvements in performance and power efficiency, alongside a range of new features. Ultimately, the emergence of smartphones drove the development of the powerful and sophisticated mobile chips used today, commonly referred to as Systems-on-Chips (SoC).
These SoCs are highly flexible and incorporate all, or most, electronic components onto a single chip, such as CPUs, GPUs, modems, image signal processors and even AI accelerators. This enables the wide range of applications and services on today’s smartphones and other mobile devices.
Efficient AI is the catalyst of the dynamics driving the move to custom silicon, as the technology industry strives to move faster and differentiate designs.
Mobile SoCs proliferated to other consumer devices and technology markets, including Internet of Things (IoT) and embedded systems, data centers and cloud computing, and automotive applications. For example, PC and laptop devices are increasingly adopting mobile SoCs due to the focus on slimmer hardware designs and consumer demand for multiday battery life.
Efficient Computing for AI
More computing functionalities and components led to a continuous industry-wide push for greater performance that still exists today. However, as Moore’s Law reaches its limit due to physical design as well as technical and economic factors, there is a renewed focus on efficient computing in modern SoC designs.
Efficiency is particularly relevant for today’s AI workloads that continue to grow in size and complexity, bringing rising energy usage and costs that are not sustainable in the long term, both for the planet and businesses. The drive for efficient computing in AI covers all technology markets, from cloud data centers to the edge—and the billions of devices used worldwide today.
Industry-Wide Adoption of Custom Silicon
Efficient AI is the catalyst of the dynamics driving the move to custom silicon, as the technology industry strives to move faster and differentiate designs. Almost everyone involved in the computing industry is exploring and investing in custom silicon, particularly the leading cloud hyperscalers.
AWS’ Graviton4, Microsoft’s Azure Cobalt and Google Cloud’s Axion CPU have custom silicon solutions designed to accelerate cloud workloads, including AI, and reduce power consumption. All of these are built on Arm (NASDAQ: ARM) technology.
Custom silicon is not just for big technology companies. Faraday Technology, an emerging fabless technology provider, is developing its own 64-core custom silicon SoC for data centers and advanced 5G networks, which is supported by Arm and Intel Foundry Services (IFS).
CSS and Chiplets
A vital enabler for custom silicon is CSS, which is configured for specific markets, like data centers, cloud computing and consumer technology markets now and automotive applications in the near future. Custom silicon solutions then further optimize the CSS for specific use cases and applications for greater performance, lower power consumption and a quicker time to market.
CSS can be used as a foundation for building chiplets, which allow the stacking, partition and interconnection of multiple semiconductor dies. This packaging technique is appealing to the industry as it will lower design costs, support interesting new silicon designs, work across different vendors and speed up silicon innovation to keep pace with the rapid growth of AI. Arm is already working with technology partners to accelerate the emerging chiplet market through common frameworks and industry standards.
Ongoing Silicon Evolution
Thanks to the modern silicon evolution, today’s technologies are more powerful, efficient and connected. Fundamentally, this means better user experiences. In the age of AI, silicon designs continue to evolve through innovative new approaches designed to manage increasingly complex computing workloads and deliver new, flexible configurations.
With a nearly 35-year heritage in SoC designs, Arm has been at the forefront of this silicon evolution, from the very first mobile chipsets to new custom silicon designs. Through further silicon innovation, Arm and technology partners from the semiconductor industry and beyond will continue to play a pivotal role in defining the future of computing.