AMD bakes its Ryzen chips on 3 nm, while Intel’s latest Cure Ultra chips are built on 1.8 nm or 18 Angstrom. What exactly do these nanometers mean, and what do they really say about a chip’s performance?
Intel’s third-generation Core Ultra processors, or Panther Lake, are manufactured on Intel’s own 18A baking process. 18 stands for 18 Angstrom or 1.8 nanometers. AMD has its Ryzen CPUs built in TSMC’s factories, on 3 nm. In the near future, AMD will embrace 2 nm at TSMC.
There’s a lot of back-and-forth between Intel and TSMC regarding these nanometer designations. What Intel calls 18A, for example, would be identical to what TSMC describes as 5 nm. The distinction is important: the size of a production node has a direct impact on chip performance. Now that Intel is taking a place at the technological forefront again with Panther Lake, and AMD has emerged as a formidable competitor, we are investigating what these nanometers actually mean.
About Nodes
The collective name for the technology used to make a chip is a ‘node’. Nodes are distinguished from each other by a naming convention in nanometers. 7 nm, 5 nm, 3 nm… are therefore production nodes. They give an indication of the accuracy of the production process in a factory. For example, Intel has chip factories that produce chips on a 4 nm node, which means that these chips are manufactured with a collection of modern technology that lives under the name ‘4 nm’.
A common misconception is that the nanometers refer to the size of the transistors on a chip. Today, this is not the case. Until 1997, there was a direct relationship between the size of a transistor and the naming of a node. Specifically, a node corresponded to the length of the transistor gate produced with the technology.
The 350 nm node in 1995 got its name because of chips with transistors with a gate length of 350 nm that rolled off the 350 nm node production line. Around the turn of the century, this relationship was lost. New technologies for building microtransistors led to higher transistor density on microchips, making gate length no longer the relevant factor.
Fewer Players, more Differences
Until about 2013, there was consensus in processor land about what a specific node should entail. Because it takes an average of almost fifteen years to get technology from a scientific paper to an operational chip factory, there was until then an overarching industry body that developed roadmaps for processor nodes over the long term. The International Technology Roadmap for Semiconductors, ITRS for short, set goals for each node regarding the number of transistors per chip and the average size of components on such a chip.
ITRS was abolished in 2017, although there is a successor with the International Roadmap for Devices and Systems (IRDS). Its importance is smaller. Logically: where chips used to be built in factories of a dozen manufacturers, today there are only three companies that can set up production lines with the most modern technology: Intel, Samsung and TSMC.
Hijacked by Marketing
If the nanometer designation no longer corresponds to a specific component on a chip, and it is also not an umbrella term under which conventions in terms of size and technology are hidden, what are concepts such as nm and 3 nm still worth?
The nanometer designation for nodes has now been taken over by the marketing department.
Technically, very little. The nanometer designation for nodes has today been taken over by the marketing department and no longer directly relates to components of a microchip. If Intel goes from 4 nm to 3 nm or even 1.8 nm, you cannot deduce from this what exactly is happening in the production process.
Indicative Term
Today, the node size is more of an indication that a chip maker has equipped its factories with new technology, which is more precise at various levels and allows for higher transistor density. How much more precise is not entirely clear. Historically, a new node should be accompanied by a doubling of the number of transistors on a given area, but even this rule of thumb is little more than a rough indication today. When a node shrinks from size X to size Y, the manufacturer mainly indicates that a big leap forward has been made in its own production process.
What that production process is exactly differs more than ever between different factories. Intel builds its chips differently than TSMC and Samsung. Different technologies related to a scale change are introduced at different times (FinFET, RibbonFET, Gate-last…). This ensures that a chip designer today must work closely with his manufacturer of choice.
Chip designs must be optimized more than ever not only for the node used, but for the implementation of that node by TSMC, Intel or Samsung. Anyone wanting to have a design created for Samsung, for example, made in a TSMC factory on a similar node, can expect a small year of redesigns.
Comparing
Looking at one manufacturer, the naming of a node is still valuable. 3 nm is better than 5 nm, and the improvements are large enough for the chip builder to speak of a new node. As a rule, this means more compact components on the chip, which ensures better performance. We will explain exactly why this is later in the article. In any case, you can undoubtedly say that Intel’s 1.8 nm Panther Lake chips are structurally superior to the 3 nm Arrow Lake processors.
3 nm is better than 5 nm. Surprise?
Making a comparison between manufacturers has become more difficult. AMD’s 3 nm Ryzen chips are baked by TSMC. A smaller node makes for better chips, but node names are the property of the marketing department. Is 3 nm at TSMC therefore really smaller than, for example, 4 nm at Intel?
The Numbers behind the Node
To investigate such a claim, we need to look for more objective parameters. Specifically, it concerns the Contacted Gate Pitch and the Minimum Metal Pitch. The Gate Pitch is roughly the minimum distance between transistors on a chip, while the Metal Pitch is the minimum distance between the interconnects that connect transistors to logical circuits.
A smaller distance between transistors implies a higher transistor density, a smaller distance between the interconnects means that you can also connect the extra transistors to more complex circuits.
For clarification, we look at the 3 nm and 4 nm nodes from Intel and TSMC. TSMC’s N4 node has a gate pitch of 51 nm and a metal pitch of 28 nm. Intel 4 has similar figures with 50 nm and 30 nm. For the TSMC N3E node, the gate pitch drops to 48 nm and the metal pitch to 23 nm. For the Intel 3 node, we remain stuck at 50 nm and 30 nm.
This shows that what Intel calls Intel 3 and Intel 4 is close to TSMC N3E and TSMC 4. The difference between 4 nm and 3 nm is only small in both cases, which implies that the node reduction from 4 nm to 3 nm was also partly driven by marketing.
For Intel 18A and TSMC N2, the comparison is more difficult to make. TSMC would achieve a higher density with N2 than Intel 18A (238 MTx/mm² vs 313 MTx/mm² – million transistors per square millimeter). Intel 18A could then deliver more performance. It seems that both nodes will be evenly matched, even though Intel suggests with the 18A designation that it has a lead.
Smaller, but why Better?
Finally, we come to the heart of the matter: why is smaller better? A higher density (of which Contacted Gate Pitch and Minimum Metal Pitch are good objective parameters) is generally associated with smaller transistors. A smaller transistor has a smaller distance between source and drain. This means that a lower voltage is needed to switch the transistor from 0 to 1 and vice versa.
A chip with an identical design, baked on a smaller node, will therefore draw less power than its predecessor. On the one hand, this has a direct impact on the battery life of laptops, but on the other hand it also enables better performance. After all, the lower consumption also reduces the heat generated.
Because chips on a smaller node inherently run cooler than identical chips built on a larger node, they can handle slightly higher frequencies at the same TDP (thermal design point = maximum power consumption). The architecture (changes to chip design that do not relate to the node) also has an impact, but it is smaller.
Drawing Conclusions
Therefore, you can state that Intel Panther Lake (Intel 18A) is inherently better than Meteor Lake (Intel 4), without knowing the details of the node: smaller is physically always better. The chips are both baked in Intel factories and are therefore comparable.
For the same reason, you can say little about Panther Lake versus AMD chips based on baked on the TSMC 2N node. 1.8 nm seems smaller than 2 nm, but as we explain in detail above, that is not always true in processor land.
This article originally appeared on August 29, 2019. It has been updated with the most recently available information.