Arm revealed its plans for next year at its annual Customer Technology Days. It also provides a comprehensive look at the “core modular elements” you can expect in Arm-socs from various smartphone and tablet manufacturers, among others, in the coming years. Naturally, AI plays a major role in this, and of course the CPU and GPU cores are becoming a bit more powerful and economical. However, Arm will also expand its services; Going forward, the design company will supply nearly ready designs for 3nm production of its core, where previously it only provided the basic elements for it.
CSS
The term CSS has nothing to do with web page layout, but has been used for some time in the server branch of Arm. Arm has been supplying Neoverse Compute Subsystems, or CSS, for this server market for some time. These are almost complete and optimized designs that can be used in slide design. For the customer segment, Arm has so far only offered so-called RTL designs, which only provide a schematic representation of the gates and transistors used. With the move to CSS, Arm provides full builds, known as GDSII, for its core. Such a design is actually a whole design of a manufacturer like TSMC or Samsung: the so-called adhesive tape.
The Customer CSS feature, as Arm calls this initiative, is multifaceted. Soc manufacturers, such as MediaTek, Qualcomm, and Samsung, can integrate Arm designs into their products much faster. This means lower implementation costs and a shorter period between design and production. Furthermore, Arm has already made improvements for its customers. GDSII designs are fully optimized for the “target node”, so you get the most performance possible with the least possible power.
The target node for this first generation CSS for Client is TSMC’s 3nm node and Samsung’s 3nm node. Customers can therefore choose the foundry where their products will be manufactured. The previous version of CSS for Client, known as Total Compute Solutions since 2021, has been optimized for 4nm chips over the past two years. With the reduction of node and improvements of Arm itself, which of course knows its designs better than anyone else, the clock speed of the new generation has also been increased. The fastest core, the X-core, has been given a target clock speed of “more than 3.6 GHz.” In practice, the clock speed was found to be 3.8 GHz, while previous designs were improved to 3.3 GHz.
To be clear, Arm does not provide a complete community. It doesn’t do this with different TCO variables and it doesn’t do it with CSS. However, it provides all the individual building blocks, just as before. However, core elements of the CPU and GPU are now optionally available as off-the-shelf files that can be sent to TSMC or Samsung for 3nm production. Building the rest of the SoC still needs to be done as usual, although the other parts are somewhat less important than the cores. There is a reference platform and Arm can also oversee the development of other components of the SoC if desired. We’ll look at those cores and other components in the following pages.