Intel unveils roadmap to 2025: end of the nanometer war and frontal attack on TSMC

Intel unveils roadmap to 2025

The Intel Accelerated 2021 webcast revealed the company's plans until 2025 to respond to the competition and return to the leadership of the processor industry from an innovation standpoint as well. The change of direction was already evident in the massive investments in Arizona, destined to significantly strengthen an internal production capacity that until recently seemed even destined to decrease.

Today, on the other hand, an Intel is envisaged in open challenge with TSMC, the largest manufacturer of silicon wafers, a scenario made possible by the combination of factors such as the financing of the new US administration in the internal production of transistors, the need to face increasingly fierce competition and the enormous growth of the market for PCs of all types and sizes.

Hence news in transistors, a whole new nomenclature that gets rid of nanometers, a 20 billion investment in Arizona and an agreement with the mobile chip giant Qualcomm which will be a customer of Intel foundries starting in 2024 But the evolution of technologies, which will include clear leaps for Foveros technology and thus for stacked chips, will benefit Intel CPUs such as those of the Meteor Lake series. And not only.

The evolution of transistors: from FinFET to RibbonFET

Change of nomenclature, but also of forward speed Once the production problems have been solved, Intel has finally decided to leverage its huge finances to relaunch and to do it in style. The most important innovations concern the arrival of the new second generation Foveros packaging, the first step forward on a path designed to bring to maturity a vital technology for the evolution of chips, and investments in real transistors, from evolution of FinFET technology until the arrival of the new RibbonFET transistors. But there is no need to underestimate the change in nomenclature, designed to emphasize the qualities of Intel silicon, also from the point of view of Intel Foundry Services and therefore of potential new customers for the company's foundries.

D ' on the other hand, the measurement of the transistor gate tells only a part of the truth and does not necessarily correspond to efficiency and a higher density per square millimeter. The reduction of the production process is important but, as we know well, the 10 nanometer FinFET transistors used in the 11th generation Intel Core processors are no more backward than the 7 nanometer ones of TSMC, although the name seems to indicate just that. Hence the decision of Intel to put aside the classic naming based on size, to give the new production processes more captivating, and easily recognizable names, such as Intel 7, Intel 4, Intel 3 and Intel A20. In the latter case, the reference to the dimensions of the transistors will return, but the A refers to the ångström, a measurement that corresponds to a tenth of a nanometer. We are therefore talking about 2 nanometers for the chips that should go into production in 2024, use newly developed transistors and, for this reason too, be used in Qualcomm chips.

Among other things, Intel's plans already include transistors Intel A18, and therefore from 18 ångström, perhaps to get ahead of TSMC, already working to design its 2 nanometer transistors, in terms of measurement therefore similar to the aforementioned Intel A20. But, we said it, the numbers don't tell everything about the chips. With the transition to three-dimensional transistors, the density has freed itself from the simple measurement of the gates, so much so that a 10-nanometer Enhanced SuperFin chip can have more transistors than a chip of the same size based on TSMC's 7-nanometer production process. And it is something that Intel wanted to emphasize by renaming it Intel 7, with a move that reaffirms its intention to launch a direct challenge to TSMC.

So this is how Intel's processors roadmap will be divided up to 2025, with the new nomenclatures.

Intel 7

Intel doesn't want its new transistors to be underestimated The 10 nanometer Enhanced SuperFin manufacturing process, now called Intel 7, is the one used for the upcoming processors Alder Lake hybrids, arriving at the end of the year together with the new LGA1700 socket, and according to Intel is able to guarantee an increase in performance per watt included in a range ranging from 10% to 15% compared to 10 nm SuperFin transistors. This is thanks to optimizations of the FinFET transistor so good that Intel can also use the technology for the Sapphire Rapids datacenter processors that will be produced starting from the first quarter of 2022.

Intel 4

Intel 4, previously intended to be the 7-nanometer manufacturing process, will use EUV lithography which, combined with FinFET transistors, will allow it to condense between 200 and 250 million transistors per square millimeter. It should therefore be greater than 5 nanometers TSMC, stopped at about 170 million transistors per square millimeter, as expressed by the choice of number 4 to underline a superiority that we hope to see also in practice.

On the other hand, as we said, the transistors will still be the Intel FinFET ones developed 10 years ago. But between optimizations and the new production process they should guarantee a further 20% increase in terms of the ratio between watts and performance. Production will begin in 2022 and the first products are expected to arrive in 2023, starting with Intel Meteor Lake high-performance desktop and notebook processors which, followed by Granite Rapids datacenter CPUs, will also leverage next-generation Foveros packaging. They could therefore make a clear leap in terms of efficiency.

Intel 3

Intel 3 will also focus on the optimization of FinFET transistors, but combined with an increase in the resolution of EUV lithography functional to a production process which, as the name suggests, should be comparable to TSMC's 3-nanometer one, at least in terms of transistor density per square millimeter. The promise, in this case, is that of an increase in the ratio between consumption and performance of 18%, consistent with the probable use, given the start of production in 2023, with the Intel desktop and notebook processors, as yet unnamed, in arrival in 2024.

Intel 20A

With Intel A20, Intel transistors promise an epochal leap forward The arrival of Intel 20A transistors, whose entry into production is scheduled for 2024, it will be very important to both future Intel processors and Intel Foundry Services' appeal to customers. But, although things can change dramatically in 3 years, they seem to have already convinced Qualcomm that it will take advantage of the introduction of a completely new technology. After 13 years of honored career, FinFET transistors will finally be replaced by RibbonFETs that promise really great things.

First of all we talk about gates on all sides, with the promise of a switching speed higher than that of the current gates with the same activation current, and a smaller footprint thanks to the use of vertical fins that will also allow to produce transistors of variable width. But there is more and it is PowerVia technology that promises to solve a big bottleneck problem that has always created problems in the field of processors. Intel's solution was to separate the data interconnection from the electrical components, with the latter placed at the base of the transistors and therefore in direct contact with the power supply, with all the advantages of the case in terms of signal intensity and cleanliness. electric.

Intel 18A

The Intel roadmap also includes 2025, the year of the Intel 18A production process which presumably will transform into real processors in the course of 2026 by further refining the transistors RibbonFET and reducing the size of the production process by an additional 10%. This is thanks to the new High NA EUV lithography with which it aims to make its foundries even more attractive to potential customers. It is no coincidence that to do so he decided to collaborate with ASML, a leader in extreme ultraviolet lithography which has scanners that can cost up to 120 million dollars each.

Foveros: the transformation of processors

Even the evolution of Foveros packaging promises significant leaps forward for the CPUs of the near future The packaging technology has nothing to do with the packages that contain the processors, but it concerns the way in which the different components of the chips, in this case stacked on top of each other in order to obtain advantages in terms of data transmission, size and efficiency. We have already seen the first incarnation of Foveros at work with LakeField processors, with CPU, GPU and memory stacked on the same die, but we will also see it used in the tiles of Intel Meteor Lake desktop and notebook processors which, arriving in 2023 , will take advantage of the second generation Foveros improvements in terms of feed margins and bump pitch, reduced to just 36 microns.

Foveros Omni

We are talking about an interconnect speed ever closer to the performance of monolithic chips that will further improve with the real leap forward in technology, promised by Foveros Omni. On the other hand, this third generation, which should be available from 2023 and therefore characterize the Intel chips arriving in 2023, will also introduce the possibility of stacking layers based on different production processes, offering almost absolute flexibility both in the development phase and in the energy management of a chip installed in any system. All enriched by the die-to-die interconnection that will allow you to check the operation of every single component of the chip, in order to avoid that a possible faulty layer can ruin the entire stack.

Foveros Omni and Foveros Direct aim for monolithic chip performance combined with total flexibility in chip development

Foveros Direct

Along with Foveros Omni will also come Foveros Direct which will implement direct interconnection from copper to copper and a much higher overall density, with a bump pitch below 10 microns which should guarantee exponential advantages in terms of lower resistance and increased efficiency. It will therefore be another fundamental weapon in the arsenal of an Intel that will undoubtedly have to face similar solutions developed by the competition, but which has undoubtedly returned to move towards the future with a much more decisive attitude, as is expected from a company that has made the history of CPUs and still wants to do it.

Intel has in fact this triple soul that sees it as a protagonist not only in the consumer field, but also in the server market with the evolution of the EMIB interconnection and in as the only foundry truly capable of competing with TSMC at least in terms of future technologies such as the High NA EUV ultra-high resolution lithography.

First, however, there will be the leap towards hybrid core processors, with the Intel Alder Lake series which will probably be among the protagonists of the Intel Innovation event scheduled for October 27th and 28th, in San Francisco.

Have you noticed any errors?