Intel Announces Intel 14A (1.4 nm) and Intel 3T Foundry Nodes, Launches World's First Systems Foundry Designed for the AI Era

[btarunr]

Intel Corp. today launched Intel Foundry as a more sustainable systems foundry business designed for the AI era and announced an expanded process roadmap designed to establish leadership into the latter part of this decade. The company also highlighted customer momentum and support from ecosystem partners - including Synopsys, Cadence, Siemens and Ansys - who outlined their readiness to accelerate Intel Foundry customers' chip designs with tools, design flows and IP portfolios validated for Intel's advanced packaging and Intel 18A process technologies.

The announcements were made at Intel's first foundry event, Intel Foundry Direct Connect, where the company gathered customers, ecosystem companies and leaders from across the industry. Among the participants and speakers were U.S. Secretary of Commerce Gina Raimondo, Arm CEO Rene Haas, Microsoft CEO Satya Nadella, OpenAI CEO Sam Altman and others.



"AI is profoundly transforming the world and how we think about technology and the silicon that powers it," said Intel CEO Pat Gelsinger. "This is creating an unprecedented opportunity for the world's most innovative chip designers and for Intel Foundry, the world's first systems foundry for the AI era. Together, we can create new markets and revolutionize how the world uses technology to improve people's lives."

Process Roadmap Expands Beyond 5N4Y
Intel's extended process technology roadmap adds Intel 14A to the company's leading-edge node plan, in addition to several specialized node evolutions. Intel also affirmed that its ambitious five-nodes-in-four-years (5N4Y) process roadmap remains on track and will deliver the industry's first backside power solution. Company leaders expect Intel will regain process leadership with Intel 18A in 2025.

The new roadmap includes evolutions for Intel 3, Intel 18A and Intel 14A process technologies. It includes Intel 3-T, which is optimized with through-silicon vias for 3D advanced packaging designs and will soon reach manufacturing readiness. Also highlighted are mature process nodes, including new 12 nanometer nodes expected through the joint development with UMC announced last month. These evolutions are designed to enable customers to develop and deliver products tailored to their specific needs. Intel Foundry plans a new node every two years and node evolutions along the way, giving customers a path to continuously evolve their offerings on Intel's leading process technology.

Intel also announced the addition of Intel Foundry FCBGA 2D+ to its comprehensive suite of ASAT offerings, which already include FCBGA 2D, EMIB, Foveros and Foveros Direct.

Microsoft Design on Intel 18A Headlines Customer Momentum
Customers are supporting Intel's long-term systems foundry approach. During Pat Gelsinger's keynote, Microsoft Chairman and CEO Satya Nadella stated that Microsoft has chosen a chip design it plans to produce on the Intel 18A process.

"We are in the midst of a very exciting platform shift that will fundamentally transform productivity for every individual organization and the entire industry," Nadella said. "To achieve this vision, we need a reliable supply of the most advanced, high-performance and high-quality semiconductors. That's why we are so excited to work with Intel Foundry, and why we have chosen a chip design that we plan to produce on Intel 18A process."

Intel Foundry has design wins across foundry process generations, including Intel 18A, Intel 16 and Intel 3, along with significant customer volume on Intel Foundry ASAT capabilities, including advanced packaging.

In total, across wafer and advanced packaging, Intel Foundry's expected lifetime deal value is greater than $15 billion.

IP and EDA Vendors Declare Readiness for Intel Process and Packaging Designs
Intellectual property and electronic design automation (EDA) partners Synopsys, Cadence, Siemens, Ansys, Lorentz and Keysight disclosed tool qualification and IP readiness to enable foundry customers to accelerate advanced chip designs on Intel 18A, which offers the foundry industry's first backside power solution. These companies also affirmed EDA and IP enablement across Intel node families.

At the same time, several vendors announced plans to collaborate on assembly technology and design flows for Intel's embedded multi-die interconnect bridge (EMIB) 2.5D packaging technology. These EDA solutions will ensure faster development and delivery of advanced packaging solutions for foundry customers.
Intel also unveiled an "Emerging Business Initiative" that showcases a collaboration with Arm to provide cutting-edge foundry services for Arm-based system-on-chips (SoCs). This initiative presents an important opportunity for Arm and Intel to support startups in developing Arm-based technology and offering essential IP, manufacturing support and financial assistance to foster innovation and growth.

Systems Approach Differentiates Intel Foundry in the AI Era
Intel's systems foundry approach offers full-stack optimization from the factory network to software. Intel and its ecosystem empower customers to innovate across the entire system through continuous technology improvements, reference designs and new standards.
Stuart Pann, senior vice president of Intel Foundry at Intel said, "We are offering a world-class foundry, delivered from a resilient, more sustainable and secure source of supply, and complemented by unparalleled systems of chips capabilities. Bringing these strengths together gives customers everything they need to engineer and deliver solutions for the most demanding applications."

Global, Resilient, More Sustainable and Trusted Systems Foundry
Resilient supply chains must also be increasingly sustainable, and today Intel shared its goal of becoming the industry's most sustainable foundry. In 2023, preliminary estimates show that Intel used 99% renewable electricity in its factories worldwide. Today, the company redoubled its commitment to achieving 100% renewable electricity worldwide, net-positive water and zero waste to landfills by 2030. Intel also reinforced its commitment to net-zero Scope 1 and Scope 2 GHG emissions by 2040 and net-zero upstream Scope 3 emissions by 2050.

View at TechPowerUp Main Site

 

[john_]

If 18A and 20A are 5nm, I doubt 14A means 1.4nm.

 

[ncrs]

That infographic looks familiar

(TPU article from 2018)

Hopefully Intel will be able to deliver on their plans and stay competitive. TSMC becoming a monopoly isn't good for us consumers.

 

[3valatzy]

If 18A and 20A are 5nm, I doubt 14A means 1.4nm.

Yup.

TSMC 10 nm = 52 MTr / mm2
Intel 10 nm = 100.76 MTr / mm2
TSMC 7 nm = 114 MTr / mm2
TSMC 6 nm = 114 MTr / mm2
Intel 4 nm = 123 MTr / mm2
TSMC 5 nm = 138 MTr / mm2
TSMC 4 nm = 144 MTr / mm2
TSMC 3 nm = 224 MTr / mm2
TSMC 2 nm = 258 MTr / mm2

With 100% density being TSMC N2. Compared to N2, Intel 18A appears ~ 75% as dense, and Intel 3 appears 45% as dense.

https://www.reddit.com/r/intel/comments/15ctryl

Intel 3 nm = 133 MTr / mm2 ?
Intel 2 nm = ~150 MTr / mm2 ?
Intel 1.8 nm = 193 MTr / mm2 ?

TSMC becoming a monopoly isn't good for us consumers.

Don't forget SMIC, UMC, GlobalFoundries, Samsung and Intel.

 

[ModEl4]

I wonder what chip design Microsoft has chosen to produce on the Intel 18A process.

 

[phints]

Imagine if Intel just released a new hedt CPU on a new lithography rather than talking about it for years? What ever happened to even Intel 4 let alone Intel 18A?

If Zen 5 will use TSMC 4nm does Intel even stand a chance on a recovery in Q4 2024?

 

[3valatzy]

Imagine if Intel just released a new hedt CPU on a new lithography rather than talking about it for years?

I suspect in that case AMD would be forced to offer something better than a 16-core.

What ever happened to even Intel 4 let alone Intel 18A?

Some say that ASML delivers the needed EUV machines only to TSMC? If that's the case, then it's more like denied access to Intel of edge technologies.
The other version is that Intel lacks engineering know-how to deliver anything beyond its 10nm/7 process.

If Zen 5 will use TSMC 4nm does Intel even stand a chance on a recovery in Q4 2024?

Except that Intel has the larger pie of the market share, despite the very high competitiveness of Ryzen. That's only the DIY market and mostly in Germany which heavily favours AMD's products for historical reasons, they have a former AMD arm in Dresden?

Intel/AMD x86 computer CPU market share 2024 | Statista

In the first quarter of 2024, 63 percent of x86 computer processor or CPU tests recorded were from Intel processors, while 36 percent were AMD processors.

www.statista.com

 

[ncrs]

Don't forget SMIC, UMC, GlobalFoundries, Samsung and Intel.

I am not forgetting them, but TSMC is dominating the highest end manufacturing of semiconductors. Intel is using TSMC for GPUs and now parts of CPUs as well.
Samsung has not been able to score a huge client in recent years as far as I know. The last one was NVIDIA's consumer versions of Ampere - the datacenter version was made by TSMC.

 

[Denver]

Yup.

TSMC 10 nm = 52 MTr / mm2
Intel 10 nm = 100.76 MTr / mm2
TSMC 7 nm = 114 MTr / mm2
TSMC 6 nm = 114 MTr / mm2
Intel 4 nm = 123 MTr / mm2
TSMC 5 nm = 138 MTr / mm2
TSMC 4 nm = 144 MTr / mm2
TSMC 3 nm = 224 MTr / mm2
TSMC 2 nm = 258 MTr / mm2


https://www.reddit.com/r/intel/comments/15ctryl

Intel 3 nm = 133 MTr / mm2 ?
Intel 2 nm = ~150 MTr / mm2 ?
Intel 1.8 nm = 193 MTr / mm2 ?



Don't forget SMIC, UMC, GlobalFoundries, Samsung and Intel.

From what I've researched, the 13900k (intel 7) has 25.9B transistors, so if the die is 258mm², that's 100mT/mm².

Intel's process density isn't bad, the same goes for Samsung; the problem is that in terms of efficiency, both are crawling at the feet of TSMC

 

[redeye]

That infographic looks familiar

(TPU article from 2018)

Hopefully Intel will be able to deliver on their plans and stay competitive. TSMC becoming a monopoly isn't good for us consumers.

how is TSMC monopoly when they’ll make chips that Intel won’t make…?

 

[john_]

If Zen 5 will use TSMC 4nm does Intel even stand a chance on a recovery in Q4 2024?

There are rumors of AMD going with 3nm for Zen 5c. If AMD is going to use both TSMC processes, 4nm AND 3nm, then maybe their inside info is that Intel's future nodes are doing better than the past and AMD needs all the advantages TSMC has to offer.

 

[FoulOnWhite]

big IF Intel start to do better than TSMC, AMD will be screwed as they are at the mercy of TSMC. I never write off Intel. Still if i had the cash right now i would be buying into an AMD rig instead of upgrading my Intel one.

 

[ncrs]

how is TSMC monopoly when they’ll make chips that Intel won’t make…?

Without competition from Intel (or Samsung, or any other foundry) TSMC will be able to maximize pricing for their top nodes, and that will increase the cost to the end consumer.

 

[DavidC1]

Yup.

TSMC 10 nm = 52 MTr / mm2
Intel 10 nm = 100.76 MTr / mm2
TSMC 7 nm = 114 MTr / mm2
TSMC 6 nm = 114 MTr / mm2
Intel 4 nm = 123 MTr / mm2
TSMC 5 nm = 138 MTr / mm2
TSMC 4 nm = 144 MTr / mm2
TSMC 3 nm = 224 MTr / mm2
TSMC 2 nm = 258 MTr / mm2

The issue with that analysis is that it says Intel 4 is only 22% increase in density over Intel 7 when Meteorlake shows 40% increase in density.

Meteor Lake Die Shot and Architecture Analysis – Why Is Intel 4 Only A 40% Area Reduction Versus Intel 7?

Die shot analysis shows 40% area reduction on identical structures from Intel 7 to Intel 4. Analysis on Redwood Cove and Crestmont architecture, Foveros Omni, Intel's use of TSMC N3 in the GPU, and SOC/IO chiplets

www.semianalysis.com

It's at an even greater 52% when you analyze the E core Crestmont. Analyzing individual blocks shows gains as much as 60% gains. L2 cache size has slightly decreased while capacity increased from 1.25 to 2MB, a 60% increase in density.

If 18A and 20A are 5nm, I doubt 14A means 1.4nm.

Neither are TSMC's N3 3nm. After 32nm, meanings of process has been thrown out the window. Actually, deviation started back in 65-45nm.

 

[Noyand]

Imagine if Intel just released a new hedt CPU on a new lithography rather than talking about it for years? What ever happened to even Intel 4 let alone Intel 18A?

If Zen 5 will use TSMC 4nm does Intel even stand a chance on a recovery in Q4 2024?

Intel 4 is already found on commercial products :

It seems that people are still confused about Intel new naming scheme : it's a TSMC scale, intel 7nm is supposed to compete against TSMC 4nm hence they called it intel 4. ( And no, ADL/RPL being so inefficient isn't because intel 10nm doesn't in fact compete with TSMC 7nm, the arch is just inefficient vs AMD. Intel nodes were never comparable to the competition at equal number: TSMC 10nm was competing with Intel 14nm. They didn't care when they had the lead, but now they changed their naming to not look bad. That's a pure marketing strat. )

 

[Eternit]

What is Intel 16? Is it renamed 22nm?

 

[3valatzy]

The issue with that analysis is that it says Intel 4 is only 22% increase in density over Intel 7 when Meteorlake shows 40% increase in density.

Meteor Lake Die Shot and Architecture Analysis – Why Is Intel 4 Only A 40% Area Reduction Versus Intel 7?

Die shot analysis shows 40% area reduction on identical structures from Intel 7 to Intel 4. Analysis on Redwood Cove and Crestmont architecture, Foveros Omni, Intel's use of TSMC N3 in the GPU, and SOC/IO chiplets

www.semianalysis.com

It's at an even greater 52% when you analyze the E core Crestmont. Analyzing individual blocks shows gains as much as 60% gains. L2 cache size has slightly decreased while capacity increased from 1.25 to 2MB, a 60% increase in density.

I know. That's because someone has put semiaccurate data in the used source for reference (wiki), and I automatically used them without modifying to the more accurate values. Thanks for the correction.
This should be better:

TSMC 10 nm = 52 MTr / mm2
Intel 10/7 nm = 100.76 MTr / mm2
TSMC 7 nm = 114 MTr / mm2
TSMC 6 nm = 114 MTr / mm2
TSMC 5 nm = 150 MTr / mm2
Intel 4 nm = 150 MTr / mm2
TSMC 4 nm = 155 MTr / mm2
TSMC 3 nm = 224 MTr / mm2
TSMC 2 nm = 258 MTr / mm2

Intel 3 nm = 160 MTr / mm2 ?
Intel 2 nm = ~180 MTr / mm2 ?
Intel 1.8 nm = 193 MTr / mm2 ?

 

[DavidC1]

I know. That's because someone has put semiaccurate data in the used source for reference (wiki), and I automatically used them without modifying to the more accurate values. Thanks for the correction.
This should be better:

TSMC 10 nm = 52 MTr / mm2
Intel 10/7 nm = 100.76 MTr / mm2
TSMC 7 nm = 114 MTr / mm2
TSMC 6 nm = 114 MTr / mm2
TSMC 5 nm = 150 MTr / mm2
Intel 4 nm = 150 MTr / mm2
TSMC 4 nm = 155 MTr / mm2
TSMC 3 nm = 224 MTr / mm2
TSMC 2 nm = 258 MTr / mm2

Intel 3 nm = 160 MTr / mm2 ?
Intel 2 nm = ~180 MTr / mm2 ?
Intel 1.8 nm = 193 MTr / mm2 ?

Those are theoretical values(both for Intel and other foundries).

No matter how accurate an analysis(or analyst) might be, unless they have insider info, they can't get it correct. Also, implementations are different. The 100 MTr/mm2 for Intel 10nm is also a theoretical value likely only reachable by certain sections of the chip such as graphics and their E cores. Their P cores never lived up to the post-22nm density claims with both 14nm and 10nm being 2x as dense rather than being 2.7x. Atom on the other hand got 64% reduction with 14nm Airmont, which is a 2.7x density gain and Tremont on 10nm is significantly smaller than Goldmont Plus at about 0.8-0.9mm2 or so despite the big changes.

Actually according to Intel, Intel 4 can reach 2x the density over Intel 7. In reality it doesn't, partly because full libraries won't be available until Intel 3. Same will be with 20A vs 18A. While 20A might sound impressive, it'll underperform(in perf and density) like Intel 4 and it'll be 18A that's good. There's a reason why the volume variants are coming only with Intel 3 and 18A. With 20A, it is a small subsection with current rumors saying it'll only come with 6+8 core for -S desktop Arrowlake and rest being on N3.

Another thing about Intel's 10nm implementations is that the high-performing versions such as the 10nm SF and the Intel 7 uses slightly larger pitch over the preceding variations. Intel 7 is slightly larger than 10nm SF, and 10nm SF over 10nm. Despite that, the die size did not change. However if they were to aim for max density, 10nm SF and Intel 7 would likely be a density reduction over 10nm. The E core on Alder/Raptor is at 1.5-1.7mm2 which is larger than expected and is a deviation and likely a combo of density reduction on Intel 7 plus needing to be stuck on larger cells due to being on the same process variant as the P core. Same seems true of Meteorlake with E core on that chip being 1mm2 when a near straight shrink should be smaller.

True picture is always much, much more complex than on the surface. Likely even if Intel reaches or even beats theoretical density on 18A over TSMC's N2, TSMC is still far more attractive to use because of the broad ecosystem support and reputation.

 

[Minus Infinity]

If 18A and 20A are 5nm, I doubt 14A means 1.4nm.

It's means equivalent to competing nodes in feature size/density etc without resorting to the bogus nm metrics thrown around by TSMC and Scamsung.

 

[Dr. Dro]

Yup.

TSMC 10 nm = 52 MTr / mm2
Intel 10 nm = 100.76 MTr / mm2
TSMC 7 nm = 114 MTr / mm2
TSMC 6 nm = 114 MTr / mm2
Intel 4 nm = 123 MTr / mm2
TSMC 5 nm = 138 MTr / mm2
TSMC 4 nm = 144 MTr / mm2
TSMC 3 nm = 224 MTr / mm2
TSMC 2 nm = 258 MTr / mm2


https://www.reddit.com/r/intel/comments/15ctryl

Intel 3 nm = 133 MTr / mm2 ?
Intel 2 nm = ~150 MTr / mm2 ?
Intel 1.8 nm = 193 MTr / mm2 ?



Don't forget SMIC, UMC, GlobalFoundries, Samsung and Intel.

"nm" has largely been a marketing term for a very long time now, for example, the lithographies of Ivy Bridge/Haswell and Broadwell/Skylake:

 

[Scrizz]

The title should be changed. The press release never mentioned anything about 1.4nm.

Intel Launches World’s First Systems Foundry Designed for the AI Era

www.intc.com

If 18A and 20A are 5nm, I doubt 14A means 1.4nm.

Intel never mentioned 1.4nm in the press release. That was added by TPU

 

[Dr. Dro]

The title should be changed. The press release never mentioned anything about 1.4nm.

Intel Launches World’s First Systems Foundry Designed for the AI Era

www.intc.com

Intel never mentioned 1.4nm in the press release. That was added by TPU

It's not incorrect, 14 angstrom translates to 1.4 nm. It is a simple divide nanometer by 10 formula. However, it remains that it's a marketing term, transistor and fin pitch sizes are obviously nowhere near close to the 10 nm realm, let alone 7, 5 or 3. The use of nanometers in lithography has been inaccurate for a very long time, it's simply a marketing quirk nowadays.

 

[londiste]

Another thing about Intel's 10nm implementations is that the high-performing versions such as the 10nm SF and the Intel 7 uses slightly larger pitch over the preceding variations. Intel 7 is slightly larger than 10nm SF, and 10nm SF over 10nm. Despite that, the die size did not change. However if they were to aim for max density, 10nm SF and Intel 7 would likely be a density reduction over 10nm. The E core on Alder/Raptor is at 1.5-1.7mm2 which is larger than expected and is a deviation and likely a combo of density reduction on Intel 7 plus needing to be stuck on larger cells due to being on the same process variant as the P core. Same seems true of Meteorlake with E core on that chip being 1mm2 when a near straight shrink should be smaller.

The same applies to TSMC as well. The quoted numbers are high density variants - mobile, power efficiency - but high performance variations are generally a good third less dense by default. Intel's processes are generally comparable to latter. Yes, there are chip design level things to get performance (clock speeds mostly) further up but that applies to both.

Plus, there are things that do not scale any more. Memory is the most talked about one and one of the reasons AMD has been trying to get cache off to chiplets, be it 3DCache or MCD-s of RDNA3 GPUs.

 

[N/A]

i get the feeling that 14A would be the new 14NM that lasted for a decade 5-11th gen provided they can even manage to get it off the ground.

 

[DavidC1]

The same applies to TSMC as well. The quoted numbers are high density variants - mobile, power efficiency - but high performance variations are generally a good third less dense by default. Intel's processes are generally comparable to latter. Yes, there are chip design level things to get performance (clock speeds mostly) further up but that applies to both.

Plus, there are things that do not scale any more. Memory is the most talked about one and one of the reasons AMD has been trying to get cache off to chiplets, be it 3DCache or MCD-s of RDNA3 GPUs.

You probably know this already but when they talk about such numbers, they clarify what kind of process it is. If they achieve 85% of theoretical that's damn good. Oftentimes it falls far short, not surprising to be under 50% of the number. So it'd be 0.5 x HP process and 0.5 x HD process.

We might want to differentiate between cache and actual memory such as RAM. RAM chips have been stuck under 1xnm(10-19nm) sizes for a decade now. The reason is because the simple structure of RAM means it's easier to scale, thus it hit tool limits faster, while things like SRAM in caches and logic are more complicated so it's further away from that. RAM cell sizes are fraction of SRAM, not just because of the 1T+1C vs 6T. Seriously, they are friggin' small! eDRAM, which is a variant for logic processes is few times smaller than SRAM, is few times larger than RAM cells.

SRAM is more complex, but it's still very regular structures easier to optimize, hence it hit the limit faster. Such limits disproportionately benefit companies that are behind, since they are further away from the limit. TSMC is the leader, so they hit it first. Intel was seriously behind, so Intel 4 is a 60% density gain for SRAM which is substantial. We'll see how Intel does once they become the leading edge.