Since 22nm it has already rained a lot, but this is the crucial moment when Intel has started to implement FinFET transistors which will soon see their end, but it is precisely the starting point to mark the inflection and what is future. FinFET as a transistor technology improved the so-called “Area Gate” with a very advanced 3D structure for the time and as such, the scaling of the total area per transistor was improved.
This defeated the current measures at that time and marked a different nomenclature and a goal leap where Intel decided to change the name of its nodes and thus 22nm for 300mm wafers was born.
What happened next ? Many know the story: TSMC came, saw and won, at least momentarily, which by incorporating a more advanced lithographic process implies that they can somehow do better marketing than Intel, at least until ‘now.
The problem that we have dealt with from time to time is that there is no unification of the criteria for naming transistors as such, because each company benefits from the improvements that it implements and chooses a way of determine the benefit.
Comparatively, an Intel Pitch Gate with the same nomenclative number has nothing to do with TSMC or Samsung and vice versa, that is to say that the 10 nm of some does not technically correspond to those of the competition, we choose the constructor that we choose. From a marketing standpoint, the smaller number implies a smaller size of the transistor and that would result in a better surface area, but that is not true in the vast majority of cases.
New names: Intel 7, Intel 4, Intel 3 and Intel 20A
Intel’s current 10nm is a bit more advanced than TSMC’s 7nm and a step ahead of Samsung’s 8nm. .
Therefore, the blue giant had to tidy up and organize its technical characteristics with the fixed nomenclatures, since they did not work in the general public. Therefore, and respecting the current 10nm to which the “SuperFin” tag has been added, the problem is that this initially created controversy as it looked like 10nm + would carry that name and it really isn’t.
This is due to the alignment that we mentioned above and which leaves the way for the future with 4 key names that we will explain below with more data in hand: Intel 7, Intel 4, Intel 3 and Intel 20A , where those of Santa Clara completely gave up the slogan “nm”.
Intel 7
The evolution of 10 nm called first 10 nm + then 10 nm SuperFin as we explained just above will finally be called Intel 7. All this mess comes from the problems, the delays and the announcements that the blues made at the over the years and which finally take shape with this new name already adapted to the next decade.
What are its main improvements? The most important thing is the performance, where Intel assures us that we can see an increase per watt which could be up to 15% compared to the current 10nm, although they also affect that it could be 10% in the worst case. case.
Is it really a lot or a little? Well, being an update of the previous 10nm and seeing the jump is more than correct as these percentages are normally what is achieved in a new node, so calling it Intel 7 is in our opinion more than correct.
These are still FinFET transistors, but there are some key optimizations that we didn’t know about before like better resistance, better power control as well as its delivery. We’ve seen this in Alder Lake processors, where Intel’s efficiency has dramatically improved clock by clock.
Intel 4
Let’s come back to our position, since if the 10 nm SuperFin are now Intel 7, the old 7 nm are now called Intel 4. What will be the jump here? Considering this will be the company’s first node to use the UVU as an engraving technology, expectations are really high, and for good reason. The blue giant speaks of a 20% gain in performance per watt, which if we take into account that in principle up to 12 layers will be used in each wafer is a very relevant fact.
Because? Well, it’s simple. Fewer layers mean easier wafer creation, lower costs and higher performance.
Intel will optimize the number of layers to achieve power reduction that will be better in performance as we get closer to processors that have greater power limitations.
In other words, processors with a lower TDP could improve performance by 20% on this node, although we don’t know by how much. When will it arrive on the market and with what architectures? Well, that will be sometime in 2023, maybe it will debut before the middle of the year with Meteor Lake for desktop and by the end of the year it will be the same for Granite. Rapids in data centers and server.
Intel 3
This lithographic process also comes with some controversy as the company did not say 100% whether it was the old 7nm + node or the one called 7nm ++ at the time. From what little we know, it’s more likely to be the latter, as Intel claims there will still be 18% more performance per watt.
Also, the jump in power reduction is bigger and the performance is more scalable because the processor requires less voltage or is limited to it, so again we could see a little bigger performance gap.
Intel 3 as such will be the end of FinFET transistor technology and technically the forerunner of the biggest leap in the company’s history. For this, this Intel 3 has a higher surface density based on the growth of HP, which offers even better resistance, new materials that will improve the interconnections of the layers and with this it will be possible to interconnect more interposers.
EUV technology still has a lot to say, to the point that the company said the jump will be bigger than in the previous standard that we saw, that is, there is a bigger improvement than in the previous nodes. The architecture that will bring this node to life will be Arrow Lake at the end of 2023 if all goes well or at the beginning of 2024 at the latest.
Intel 20A
This is the biggest change as a concept and novelty that Intel has implemented in its history, as it encompasses a series of far-reaching improvements. The designation A refers to the Angström unit of measure, mainly because the company wants to leave behind the nanometer as such.
It will arrive sometime in 2024, possibly in the first half of the year, although there are already rumors that it could be by the end of the year due to all that we are seeing with the delays in. all companies. Likewise, the main improvement is that we say goodbye to FinFET and we will say hello to RibbonFET, Intel’s implementation of technology GAA o Gate-All-Around which we have already dealt with exclusively in its corresponding article.
The second improvement is the appeal PowerVia, which is intended for electrical consumption as well as for its implementation in the transistor. FinFET was fed from the top of the transistor through the same system that carried the signal, requiring near absolute precision and constant innovation in the materials used with each lithographic jump.
Intel 3 is the limit and Intel 20A what it will do with this technology PowerVia It’s simple: separate in a new transistor diagram the signal path and the power supply, which will now be produced by the lower part of each of them. You don’t have to be very smart to see the benefits that each transistor’s FinFET structure couldn’t deliver: better efficiency, lower consumption, better signal, more stable power, better control in gates, more signal noise. lower, better internal latencies, not to mention the lowest failure rate per slice.
As they do? Basically a layer is added under the transistors on the back of the board where the power cables for each unit are created. Intel is so convinced that the results will be good that they might even adapt it to FinFET with some resources devoted to it.
And they’re not even sure they can implement it, but in the words of the blue giant, they are hoping to at least try it. In any case, we are talking about a knot that would likely happen in 2025, in the end, although it would already go into production in 2024 where, regardless of that, it should bring the basic architecture to life Nova Lake with Panther Cove and Darkmont as performance and efficiency microarchitectures respectively.
Table of Contents
