Array
(
    [content] => 
    [params] => Array
        (
            [0] => /forum/threads/intel-18a-process-node-offers-25-higher-frequency-at-iso-36-lower-power-at-same-frequency-versus-intel-3-over-30-density.23047/
        )

    [addOns] => Array
        (
            [DL6/MLTP] => 13
            [Hampel/TimeZoneDebug] => 1000070
            [SV/ChangePostDate] => 2010200
            [SemiWiki/Newsletter] => 1000010
            [SemiWiki/WPMenu] => 1000010
            [SemiWiki/XPressExtend] => 1000010
            [ThemeHouse/XLink] => 1000970
            [ThemeHouse/XPress] => 1010570
            [XF] => 2021770
            [XFI] => 1050270
        )

    [wordpress] => /var/www/html
)

Intel 18A Process Node Offers 25% Higher Frequency At ISO & 36% Lower Power At Same Frequency Versus Intel 3, Over 30% Density

XYang2023

Well-known member
Intel has detailed its next-gen 18A process node, which will be the replacement for its Intel 3 node, offering improved clock & voltage scaling.

Intel 18A Process Node Offers Improved Performance, Efficiency & Density Versus Intel 3 Node
At the 2025 Symposium on VLSI Technology and Circuits, Intel further highlighted the features of its next-generation 18A process node, which will be featured on upcoming families such as the client-aimed "Panther Lake" CPUs and the server-aimed Clearwater Forest E-Core-only Xeon offerings.

Advanced CMOS Technology

“Intel 18A Platform Technology Featuring RibbonFET (GAA) and Power Via for Advanced High-Performance Computing” – Intel (Paper T1-1) An advanced Intel 18A technology featuring RibbonFET and Power Via provides over 30% density scaling and a full node of performance improvement compared to Intel 3. Intel 18A offers high-performance (HP) and high-density (HD) libraries with full-featured technology design capabilities and enhanced design ease of use.

It looks like the main highlights for Intel's 18A nodes will be RibbonFET (GAA) & PowerVia, enabling the transition into the next iteration of process technologies.
Intel 18A Process Node Offers 25% Higher Frequency At ISO & 36% Lower Power At Same Frequency Versus Intel 3, Over 30% Density 2

Starting with the 18A RibbonFET technology, Intel will be making a substantial leap from FinFET tech, offering improvements to gate electrostatics, more effective width per footprint versus FinFET, less parasitic capacitance per footprint versus FinFET, and offering improved flexibility versus FinFET.



Intel also improves upon the design flexibility of RibbonFET vs FinFET by introducing multiple ribbon widths for both 180H and 160H libraries, optimized logic power/leakage vs performance through DTCO, and specialized Ribbon Widths for SRAM that are optimized for bitcell performance, all enhancing the performance and design capabilities of next-gen chips fabricated on the 18A node.


Intel's 18A PowerVia technology will also help make improvements in the power delivery of next-gen transistors with backside power signal wires versus front-side power signal wires. These new wires are decoupled and optimized separately, allowing for:
  • - Improved Logic Density
  • - Better standard cell utilization
  • - Lower signal RC
  • - Reduced voltage droop
  • - More design flexibility
1750259021326.png


1750259049072.png



With these improvements, Intel 18A provides over 15% iso-power performance gain over Intel 3.


At the same voltage of 1.1V, Intel 18A delivers around 25% higher frequency, and it also supports low-voltage operations below 0.65V, where you get to see up to 38% power savings at the same clocks. Intel states that multiple factors of 18A help improve performance, such as:

  • - RibbonFET transistors
  • - Backside power advantage
  • - Frontside interconnect improvements
  • - Process/Design co-optimization
intel-18a-process-technology-_-vlsi_2025_4
intel-18a-process-technology-_-vlsi_2025_5
intel-18a-process-technology-_-vlsi_2025_7
intel-18a-process-technology-_-vlsi_2025_8
intel-18a-process-technology-_-vlsi_2025_9

As for density scaling, Intel 18A provides up to 39% (~30%) density improvement over Intel 3 with the backside power technology enabling an 8-10% improvement in cell utilization, which cuts worst-case IR droop by 10x. Intel 18A also supports HP Library Height of 180nm vs 240nm (Intel 3), HD Library Height of 160nm vs 210nm (Intel 3), and an M0/M2 pitch of 32/32 versus 30/42 (Intel 3).

intel-18a-process-technology-_-vlsi_2025_10
intel-18a-process-technology-_-vlsi_2025_11


Lastly, in terms of SRAM scaling, Intel 18A achieves an HCC SCRAM density improvement of 30% vs Intel 3, offering HCC 0.0230um2 and HDC 0.0210um2 SRAM. As for 18A, it just doesn't stop there as the company has additional iterations of the node which will be launching between 2026-2028. The family includes 18A-P and 18A-PT, which were recently disclosed at Direct Connect 2025. Intel also expects customers to leverage these nodes for their chip production.

 
I don't think peak frequency is guaranteed to be higher Intel 4 is a better node than Intel 7 yet Redwood Cove clocked lower than Intel 7 for peak frequency with almost the same design.
Straw hat time - what frequencies do you think Panther Lake and Nova Lake will top out at ?

Complete Silly/Wild Assed Guess for me:
Panther Lake: 5.6 GHz (Intel marketing talking about Arrow Lake performance hints they'll have a high clocked SKU)
Nova Lake: 6.3 GHz (18A will be more mature by late 2026, and they'll want to push desktop clocks)

for comparison:
Lunar Lake max turbo: 5.1 GHz (288V)
Arrow Lake max turbo: 5.7 GHz (285K) / 5.8-5.9 GHz (theoretical Arrow Lake refresh)

Feel free to laugh at me in 15 months after Nova Lake launches..
 
Straw hat time - what frequencies do you think Panther Lake and Nova Lake will top out at ?

Complete Silly/Wild Assed Guess for me:
Panther Lake: 5.6 GHz (Intel marketing talking about Arrow Lake performance hints they'll have a high clocked SKU)
Nova Lake: 6.3 GHz (18A will be more mature by late 2026, and they'll want to push desktop clocks)
PTL - 5.4-5.5 ish
NVL Desktop is N2 btw so 6 GHz it's not Intel node and there is nothing like Intel nodes when it comes to raw GHZ
for comparison:
Lunar Lake max turbo: 5.1 GHz (288V)
Arrow Lake max turbo: 5.7 GHz (285K) / 5.8-5.9 GHz (theoretical Arrow Lake refresh)

Feel free to laugh at me in 15 months after Nova Lake launches..
Laughing in Advance so I don't forget in 15 months 🤣 🤣
 
The industry should have some key metrics on this for each application.
Bitcoin (hashrate/wafer at X J/T)
CPU
mobile
 
18A is not really a replacement for Intel 3. Intel makes <%5 of its processors on Intel 3. Intel chose TSMC 3 for its "3nm" products. No external company is using Intel 3 or 4. Even employees at Intel Foundry compare Intel 7 wafer price and cost vs Intel 18A to look at financial and performance impacts since Intel 4/3 is not meaningful.

Lets see how Panther lake compares to other lakes and AMD and what its cost is. Also lets see if anyone purchases Intel 18A wafer for external foundry.

@Scotten Jones is my typical reference for comparing nodes and I think he has summaries for Intel vs TSMC.
 
18A is not really a replacement for Intel 3. Intel makes <%5 of its processors on Intel 3. Intel chose TSMC 3 for its "3nm" products. No external company is using Intel 3 or 4. Even employees at Intel Foundry compare Intel 7 wafer price and cost vs Intel 18A to look at financial and performance impacts since Intel 4/3 is not meaningful.
Timing matters as well Intel 3 is so damm late compared to competition N4 node by 2 years if it had launched same time as N4 it would have been good well it only got 2 products as of now but it's use will increase in coming years for newer products.
Lets see how Panther lake compares to other lakes and AMD and what its cost is. Also lets see if anyone purchases Intel 18A wafer for external foundry.

@Scotten Jones is my typical reference for comparing nodes and I think he has summaries for Intel vs TSMC.
Yup
 
Intel makes <%5 of its processors on Intel 3
Is this statement based on 2024 wafer volume? Then it is likely outdated a bit imho.
Only Granite Rapids & Sierra Forrest (Xeon 6 Server CPUs) were the products on Intel 3 in 2024 and both of them launched in 2Hish of 2024. They did not ramp to volume in 2024 (GNR was only shipped volume in Jan of 2025). I read Intel shipped only 100k+units of GNR in Q1'25. They are still shipping SPR & EMR predominantly in server space. [Side note - AMD's Turin shipments ramped (~200k units) much faster than GNR due to maturity of N4 but Genoa shipments dropped double digits]

Now Arrow Lake -U (ARL-U) is basically a (Meteor Lake MTL-U) ported from Intel 4 to Intel 3. This is a high-volume product for enterprise notebooks market. Some iGPUs tiles for Panther Lake are also on Intel 3 and these SKUs are supposed to be the high-volume SKUs of Panther Lake line up.

My expectation is Intel 3 will scale in volume this year (2025) through next year as base tiles is expected to be Intel 3 based for both Clearwater Forrest and Diamond Rapids. Also, more they shift to Xeon 6 (Intel 3), more revenue for Intel Products and Foundry.

Edited to remove the shipment volume % of SPR & EMR as that may be inaccurate, just based on my memory but SPR is largest in volume & EMR is second largest. I am certain of that.
 
Last edited:
I don't think peak frequency is guaranteed to be higher Intel 4 is a better node than Intel 7 yet Redwood Cove clocked lower than Intel 7 for peak frequency with almost the same design.
I was told by someone very credible that Panther Lake ES2 samples clocks (@4.8 Ghz) higher than ARL & RPL ES2 at that point of time during the development cycle. But we can't be sure of anything until the product details come out officially.
 
I was told by someone very credible that Panther Lake ES2 samples clocks (@4.8 Ghz) higher than ARL & RPL ES2 at that point of time during the development cycle. But we can't be sure of anything until the product details come out officially.
This is actually very, very good news. Intel 4 (in MTL) clocked significantly lower than Intel 7 out of the gate. Having 18A properly outperform (ISO-design, e.g. Intel high perf -> Intel high perf design) the node it’s replacing (TSMC N3) is critical for Foundry, Products, and the future Intel’s margins.
 
I was told by someone very credible that Panther Lake ES2 samples clocks (@4.8 Ghz) higher than ARL & RPL ES2 at that point of time during the development cycle. But we can't be sure of anything until the product details come out officially.
Is clock really that important for a laptop chip?

Also, on X, some rumors suggest arm based cpus will have massive lead on IPC. to quote, "What’s alarming is that Intel’s CPU team still seems unable to design an architecture that can effectively counter ARM. Rumors say even the next-gen PTL P-cores show no substantial progress"
 
Is clock really that important for a laptop chip?

Also, on X, some rumors suggest arm based cpus will have massive lead on IPC. to quote, "What’s alarming is that Intel’s CPU team still seems unable to design an architecture that can effectively counter ARM. Rumors say even the next-gen PTL P-cores show no substantial progress"
I am not a semiconductor guy, but I have some basic understanding of things so please take this with that in mind.

MTL/Redwood Cove P core sucked (at least portrayed as sucked by the reviewers) because it regressed in clock frequency compared to Raptor Cove in RPL. Now just going by common sense, if Intel is really getting stuck on improving IPC (Instruction per Clock) as being complained here then improving clock frequency will definitely improve performance. The real question is "is the clock improving iso power". If that is the case, then it is important for notebooks too (perf/watt is improving so does battery life).

I follow this guy in X & I learnt a ton following these guys in X - so I respect that. But his comments are based on Geek Bench CPU scores which are not always representative of real-world performances. Also, I see some people more knowledgeable in semi compared to me complain to him about this and his inclusion of Apple chips which includes some kind of accelerator like feature in the chip called SVE. I can only infer from these things, but I am not 100% sure if he is right or wrong.
 
Can someone provide a bit more context? How would this compare to say TSMC 3nm and 2nm?

(Sorry can't shrink size on iPad)
1750269438905.png
 
Can someone provide a bit more context? How would this compare to say TSMC 3nm and 2nm?
Answer to that is complicated especially in this SemiWiki Forum.

If you ask senior members here, they will say none of the Intel nodes are as good as TSMC nodes just because Intel does not have external customers on their nodes. If we go by that metric, how do we compare nodes when Intel was at it's peak making only its own products and when TSMC was trailing but making chips for others?. IMO some nuances are lost in that thinking like Intel 18A does not have external customers because of myriads of other reasons e.g. immature PDK/IP ecosystem, unproven foundry & direct competitor nature of Intel Products team with potential Intel Foundry customers.

IMHO, based on things I see\saw and read, Intel 18A is going to be either equal or better than TSMC's N3P/N3X based products during the time 18A based products hit the shelf (late 2025). This lead (if at all exists) might be short lived when TSMC's N2 based products hit the shelf in 6-9 months after in 2H'2026 (these TSMC products are mostly from AMD that compete with Intel Products directly). As shared by @Xebec above, the lead of N2 over 18A could be just one\couple of the terms in PPPACT (Performance, Power, Area, Cost & Time to market) rather than total dominance like TSMC used to have over Intel for the last 5-8 years. All of this is speculation on my part (I like to think it is an educated guess but could be all ignorance), so take it as such.
 
Is clock really that important for a laptop chip?
Yes and no.

Yes - most user facing applications are still "lightly threaded" depending on a small # of fast cores for peak performance. Higher clock speed "lifts all boats", but of course IPC, cache/memory and efficiency matter. In the case of a single (user facing) app doing a short sprint activity, even mobile usually has enough thermal budget to allow a few cores to run at high clocks.

+Yes - for marketing. High frequency still can drive higher MSRP products for some consumers and users.

No - because the difference between 5.5 and 6.0 ghz is never going to boost application performance more than about 9% and usually less. But that may be (up to) 9% that can only be achieved via lower execution latency which clock speed can help with.
 
Is this statement based on 2024 wafer volume? Then it is likely outdated a bit imho.
Only Granite Rapids & Sierra Forrest (Xeon 6 Server CPUs) were the products on Intel 3 in 2024 and both of them launched in 2Hish of 2024. They did not ramp to volume in 2024 (GNR was only shipped volume in Jan of 2025). I read Intel shipped only 100k+units of GNR in Q1'25. They are still shipping SPR & EMR predominantly in server space. [Side note - AMD's Turin shipments ramped (~200k units) much faster than GNR due to maturity of N4 but Genoa shipments dropped double digits]

Now Arrow Lake -U (ARL-U) is basically a (Meteor Lake MTL-U) ported from Intel 4 to Intel 3. This is a high-volume product for enterprise notebooks market. Some iGPUs tiles for Panther Lake are also on Intel 3 and these SKUs are supposed to be the high-volume SKUs of Panther Lake line up.

My expectation is Intel 3 will scale in volume this year (2025) through next year as base tiles is expected to be Intel 3 based for both Clearwater Forrest and Diamond Rapids. Also, more they shift to Xeon 6 (Intel 3), more revenue for Intel Products and Foundry.

Edited to remove the shipment volume % of SPR & EMR as that may be inaccurate, just based on my memory but SPR is largest in volume & EMR is second largest. I am certain of that.
Arrow lake does not run on INtel 3, it runs on tsmc 3. Intel sells 10x+ more client processors than datacenter.

GR+SF+CF+DR is less than 10M processors in 2025

Panther lake launch is in 2026

Most Intel silicon is Intel 7 or older to this day.
 
Back
Top