TSMC's N2 process has a major advantage over Intel's 18A: SRAM density

Intel silicon spin qubit progress
(Image credit: Intel)

Given how SRAM-intensive modern designs are, SRAM cell size and density are major characteristics of new fabrication technologies. Apparently, the SRAM density of Intel's 18A manufacturing process (1.8nm-class) is considerably lower than that of TSMC's N2 (2nm-class), and is closer to TSMC's N3, according to the ISSCC 2025 Advance Program. Still, Intel's 18A could have other major advantages over N2. 

Intel's 18A fabrication process features a high-density SRAM bit cell size of 0.021 µm^2 (therefore achieving an SRAM density of approximately 31.8 Mb/mm^2), which is a major improvement compared to a high-density SRAM bit cell size of 0.024 µm^2 in Intel 4, but aligns with what TSMC's N3E and N5 offer. By contrast, TSMC's N2 manufacturing technology shrinks HD SRAM bit cell size to around 0.0175 µm^2, enabling SRAM density of 38 Mb/mm^2. 

Both 18A and N2 rely on gate-all-around (GAA) transistors, but unlike Intel, TSMC has managed to shrink its high-density SRAM bit cell size quite aggressively compared to its previous-generation technologies that rely on FinFET transistors. It should be noted that in addition to SRAM bit cell size, a key characteristic of SRAM is its power consumption, and we do not really know how 18A and N2 stack up against each other regarding this metric.

Swipe to scroll horizontally
Row 0 - Cell 0 Intel 4Intel 18AN3N3EN2
SRAM Density27.825 Mb/mm^231.8 Mb/mm^233.55 Mb/mm^231.8 Mb/mm^238 Mb/mm^2
SRAM Cell Size0.0240 µm^20.0210 µm^20.0199 µm^20.021 µm^20.0175 µm^2
HVMRow 3 - Cell 1 Row 3 - Cell 2 Q4 2022Q4 2023H2 2025

Speaking of Intel's 18A, this node has two major advantages over its predecessors: GAA transistors and a backside power delivery network (BSPDN). BSPDN not only promises to improve power delivery to transistors, thereby improving performance efficiency for some designs but also enables designers to make them smaller, thus increasing logic density. 

Although modern chip designs use plenty of SRAM, and its density is crucial for node-to-node scaling, logic density is more important than HDC SRAM density. For now, we cannot compare this metric for Intel's 18A and TSMC's N2. Furthermore, logic density is hard to estimate, as each process technology has high-density, high-performance, and low-power libraries that are usually mixed and matched within a single design. As for logic density for an abstract processor, Intel and TSMC have yet to disclose it. 

One of the hardest things to scale with modern process technologies is SRAM density due to the intricacies of its design, operational requirements for stability and reliability, and increased variability at smaller nodes. That said, it is not surprising that some modern technologies may feature larger SRAM cell sizes compared to other production nodes.

Anton Shilov
Contributing Writer

Anton Shilov is a contributing writer at Tom’s Hardware. Over the past couple of decades, he has covered everything from CPUs and GPUs to supercomputers and from modern process technologies and latest fab tools to high-tech industry trends.

  • EzzyB
    It just demonstrates the uselessness of comparing node names. "NM" means nothing and has meant nothing since the turn of the century. Even at the height of everyone shouting that TSMC''s nodes were "more advanced", Intel's "10nm" node was the equivalent of TSMC's "7nm" node. (There were larger and smaller objects for each like this article describes.)

    TSMC took one look at Intel 14a on their roadmap a couple of years ago and simply changed their own roadmap to indicate their 1.5 node was now 1.3 with no physical changes so it seemed to be more advanced.

    Process node names are no longer indicative of anything but the enthusiasm of the respective companies' marketing departments and should be relied upon for nothing.

    As far as this article goes, don't you think the text in the first comment above would have been a better headline? It certainly seems more neutral as you report in the article that we really don't know yet how this will shake out.
    Reply
  • usertests
    So about +19%, and in the area that Intel could have 30 MiB, TSMC could deliver about 36 MiB. It's not nothing.

    AMD might choose to go from 32 to 36 MiB if they switch to a (standard) 12-core chiplet for Zen 6. That would be a light, easy option instead of 48 MiB. That's expected to use N3 though.
    Reply
  • ET3D
    I think that the major advantage of N2 over 18A is that N2 will actually reach the market on time.
    Reply
  • joeer77
    What about Intel 20A SRAM density? You complete glossed over that. It's also GAA.
    Reply
  • phead128
    After Pat Gelsinger's eviction, I doubt 18A will even pan out. 10% yield is devastatingly low for 18A. They need to outsource 18A Panther Lake entirely to TSMC now.
    Reply
  • Mama Changa
    joeer77 said:
    What about Intel 20A SRAM density? You complete glossed over that. It's also GAA.
    Does it matter, 20A is dead, officially cancelled and was always goiong to be a short-lived stopgap anyway, it's not being used in any consumer products at all.

    As for 18A having BSPD, TSMC has their own solution coming which is actually more advanced than BSPD as used by Intel but will be a few years behind, probably for A16 in late 2026.
    Reply
  • wwenze1
    AMD gamers: *heavy breathing*
    Reply
  • shady28
    If Intel is actually able to get 18A into a shipping product in 2025 or early 2026, it would give them the first 2nm class x86 CPUs given that the first 1-2 years of TSMC N2 will likely be monopolized by Apple A19 and M5.
    Reply
  • phead128
    shady28 said:
    If Intel is actually able to get 18A into a shipping product in 2025 or early 2026, it would give them the first 2nm class x86 CPUs given that the first 1-2 years of TSMC N2 will likely be monopolized by Apple A19 and M5.
    Intel already signaled that 30% of 18A Panther Lake will be outsourced/manufactured by TSMC. I guarantee that will be increased to 100% very soon.
    Reply
  • Artikpi
    Intel has never said that 18A equals 18 angstroms. It's just a marketing name to hide the fact that they're lagging behind in terms of engraving finesse. 18A = 3nm.

    "Intel's 18A manufacturing process (1.8nm-class)", not true
    Reply