Intel will manufacture chips for Qualcomm and want to “abandon” nanometers

THE Intel has a new business model. The company announced, on Monday afternoon (26), the division Intel Foundry Services, which aims to produce chips for third parties and thus compete with TSMC and Samsung. The first customers have already been revealed: qualcomm and Amazon.

Meteor Lake Wafer Chips (Image: Disclosure/Intel)

An announcement like this was expected since the beginning of the year. Intel is one of the few companies that make its own chips, but in March, the company unveiled its plan to invest $20 billion in expanding its production capacity, not necessarily to increase its own inventories, but to produce semiconductors for other companies.

This is the business model of TSMC, leader in the segment. Samsung and GlobalFoundries also make chips for third parties. There were even rumors that the latter would be bought by Intel for US$ 30 billion, but everything indicates that the negotiation, if it really existed, did not advance.

For now, details about the chips that will be produced by Intel for Qualcomm are sparse. It is only known that these units will be based on technology Intel 20A, which will be available from 2024 (you will already find out what this is).

About Amazon, the information is that Intel Foundry Services will produce chips made to measure for the servers of the Amazon Web Services (AWS) platform.

Intel 20A and the “end” of nanometers

Intel has a clear goal: to return to leadership in the semiconductor industry by 2025. The creation of Intel Foundry Services is only part of that plan. Another is a reformulation of the roadmap of the company’s technologies.

This redesign includes a new approach to manufacturing processes. Intel will no longer focus on nanometers and will instead adopt names like Intel 7 and Intel 20A.

Why? Today, the company’s most advanced processors have 10-nanometer technology that, at least to some extent, is comparable to 7-nanometer chips from TSMC or Samsung. Probably, the company developed the new nomenclature to avoid this kind of comparison, which makes its technology look outdated.

In short, it will look like this:

  • Intel 7: expected for the end of 2021 in the Alder Lake line, corresponds to 10 nanometer chips improved over the current generation, with performance per watt optimized by 10-15%;
  • Intel 4: With manufacturing expected in the second half of 2022 and sales in 2023, Intel 4 chips will have a 7-nanometer process and 20% better yield per watt compared to Intel 7 processors;
  • Intel 3: Expected in the second half of 2023, this technology will be an enhanced version of the 7-nanometer process; 18% better performance per watt is expected compared to Intel 4 chips;
  • Intel 20A: expected in 2024, this is a technology based on a new transistor architecture called RibbonFET; here, Intel starts measuring in angstrom, not nanometers (hence the letter ‘A’ in the name).
New Intel Technologies Roadmap (Image: Disclosure/Intel)

New Intel Technologies Roadmap (Image: Disclosure/Intel)

RibbonFET and PowerVia

An angstrom is equal to 0.1 nanometer, so you’d think 20A chips would be 2 nanometers (20 x 0.1). But it’s not like that: these chips will have a 5-nanometer process and, in 2025, they should be succeeded by the technology Intel 18A. This will also be based on 5 nanometers, but with an optimized process.

Keep in mind, however, that Intel intends to shift the focus away from nanometers. The reason is that, starting with Intel 20A technology, the company will combine the technologies RibbonFET and PowerVia.

RibbonFET technology will replace the FinFET standard, adopted by Intel since 2011, and promises faster switching speeds as well as higher transistor density.

In practice, what the company will do is adopt Gate-All-Around (or GAAFET) transistors, which can be more easily optimized for performance and lower power consumption. It is worth noting that TSMC and Samsung also intend to adopt this type of transistor at some point.

PowerVia technology optimizes signal transmission by eliminating the need for power routing on the front of the chip’s wafer. What does that mean?

Today, making a circuit involves adding a layer of transistors and, above that, several layers of metallic material for the communication of the different components of the chip.

With PowerVia technology, the transistor layer is in the middle, as if forming a sandwich. On one side are the communication components; on the other, those related to energy consumption. In theory, this method simplifies connections, improving power supply parameters.

Now are you going?

Pat Gelsinger, Intel CEO (Image: Disclosure/Intel)

Pat Gelsinger, Intel CEO (Image: Disclosure/Intel)

It’s early to know whether all these changes will be enough for Intel to reach the goal of returning to industry leadership by 2025, but one thing is certain: the arrival of Pat Gelsinger as CEO has brought a new perspective to the company.

Meeting the presented schedule will be one of the biggest challenges going forward, but Intel seems to have prepared for it: in addition to Pat Gelsinger himself, a veteran who contributed to the legendary 80486 chip design, the company has spent the last few months hiring (and rehiring) several semiconductor specialists.

Despite the confusing nomenclature, it may be time to give the company a vote of confidence.

Leave a Comment