Intel’s CPUs are going to get smaller and smaller, and to help potential buyers understand just what that means, it’s launched a new metric: The angstrom. In 2024, Intel will officially leave nanometer-scale process nodes behind and will begin to classify them with angstroms. The first chip, launched in the second half of that year (if all goes to plan), will be classed as using an Intel 20A node.
What is Intel 20A? It’s essentially 2nm, with 10a per nm, although it’s technically using a 5nm process by existing naming standards. It’s all a bit confusing, but the idea is to bring Intel’s process names to parity with its competitors and to give Intel greater scope for more gradual reductions in size, without delving into decimal places.
All processors, whether they’re central, graphical, or otherwise, are built using a specific architectural design. That, in turn, is built atop a semiconductor manufacturing process, more commonly known as a process node, or node. The nanometer naming convention was initially used to measure the length of a transistor gate, and that typically correlated exactly with half the distance between individual transistors. In contemporary computing, there is far less correlation there, making the measurement more arbitrary.
Typically, smaller node sizes mean that there is a greater density of transistors on a processor, which in turn leads to faster processors. It’s not quite that cut and try, though, with Intel remaining on its 14nm process node on desktop for over five years, but it enhanced performance considerably over that time. AMD, on the other hand, moved from smaller process node to smaller process node, and saw incredible performance gains because of it, alongside other enhancements.
Due to the arbitrary nature of process-node naming at different manufacturers, and their differing process-node designs, however, the measurement today isn’t particularly indicative of anything. Smaller is better when comparing different process nodes from the same manufacturer, but it’s equally possible that a smaller node from one company may perform worse than that of a competitor with a larger process. This is most obvious in the case of Intel, where its 10nm process node features a transistor density almost twice that of TSMC’s 10nm process node, despite both bearing the same 10nm name.
With its latest roadmap, Intel didn’t just announce angstrom as a new naming convention for its chips in 2024, it also removed all mention of nanometers entirely. Its upcoming nodes will now be known by a simple number, with no qualifiers. These numbers are purely marketing spin, but they are designed to sit in line with Intel’s competitors that are around similar performance.
Intel took a lot of flak for its many years of struggling to produce 10nm process nodes at volume, while in conjunction, AMD was heavily praised for the performance gains made using TSMC 7nm process nodes. Other fabricators, like Samsung, have moved well beyond 10nm too, even if Intel processes and processors remain hotly competitive on almost all fronts. It’s losing the marketing war, and offering products built on comparatively performing but larger process nodes wasn’t good for its image. So it’s changing it.
Intel 20A is a rebranding of what was Intel’s 5nm process and will debut in 2024. It will be followed by Intel 18A in 2025, which would previously have been called its 5nm+ node.
The rebranding will begin much sooner, though. Intel’s upcoming Alder Lake processors, set to debut in late 2021, will use Intel’s newly renamed Intel 7 process node. This was previously set to be called Intel’s 10nm Enhanced Super Fin, or 10ESF process. Switching to a simple Intel 7 name is much more streamlined. That will be followed by Intel 4 and Intel 3 processes in the coming years, before the Intel 20A process comes to the fore.
Under the new branding scheme, the 20A node would have been called Intel 1, but will instead be the first entry in what Intel calls the new “angstrom era” of process fabrication, where the process nodes aren’t just smaller again — the transistors they’re built with will be vastly different. Intel’s own 20A design will feature multi-gate RibbonFET transistors. It may be that other manufacturers move to similar design changes around that time, but they will certainly be working with increasingly more dense and smaller processes. Intel’s new Angstrom naming convention gives it greater room to 18A and beyond in the years that follow.