Skip to main content

TSMC to Initiate 1.4nm Process Technology R&D

TSMC
(Image credit: TSMC)

At processor manufacturers, fundamental and applied research and development work never stops, so now that Taiwan Semiconductor Manufacturing Co. has outlined a timeline for its N2 (2 nm-class) fabrication process that will enter high-volume manufacturing (HVM) in 2025, it is time for the company to start thinking about a succeeding node. If a new rumor is to be believed, TSMC is set to formally announce its 1.4 nm-class technology in June. 

TSMC plans to reassign the team that developed its N3 (3 nm-class) node to development of its 1.4 nm-class fabrication process in June, reports Business Korea. Typically, foundries and chip designers never formally announce R&D milestones, so we are unlikely going to see a TSMC press release saying that development of its 1.4 nm technology had been started. Meanwhile, TSMC is set to host its Technology Symposium in mid-June and there the company may outline some brief details about the node that will succeed its N2 manufacturing process.  

Standard process technology design flow includes pathfinding, research and development phases. Pathfinding involves things like fundamental exploration of materials and physics and in many cases, it is performed simultaneously for numerous nodes. By now, pathfinding for TSMC's N2 has probably been concluded, so appropriate teams specializing in fundamental physics and chemistry are working on a successors for N2, which may well be called 1.4 nm, or 14 angstroms.  

TSMC's N2 relies on gate-all-around field-effect transistors (GAAFETs), but will use existing extreme ultraviolet (EUV) lithography with a 0.33 numerical aperture (0.33 NA). Given the details about TSMC's N2 that we know today, it is possible that its successor will retain GAA transistors, but what really remains to be seen is whether it is going to move to EUV tools with a 0.55 NA (or High NA).  

Keeping in mind that TSMC's N2 enters HVM in late 2025 (so expect the first 2 nm chips from the company to be delivered around 2026) and TSMC's two-and-a-half to three-year node introduction cadence, we can potentially expect TSMC's 1.4 nm (or 14 angstroms) process to be used for commercial products starting in 2028. Given the timeframe, it will be beneficial for the node to use High NA lithography, which Intel plans to start using in 2025. 

Speaking of Intel, it remains to be seen which of Intel's node is set to compete against TSMC's 1.4 nm. Intel is set to introduce its 18A (18 angstroms) technology in 2025, so by 2028 the company will roll out at least one new fabrication process. Whether it will be called 16A (since Intel seems to be cautious with node advancements these days) or 14A will be interesting to see.

Anton Shilov is a Freelance News Writer at Tom’s Hardware US. 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.

  • tennis2
    I feel like fractions of nm is so much harder to garner a relative % difference from compared to integers. 2nm to 1.4nm is the same % as 7nm to 5nm, yet is seems so much less significant.
    Reply
  • peachpuff
    Intel: pfffft our 10nm is just as good...
    Reply
  • gfg
    tennis2 said:
    I feel like fractions of nm is so much harder to garner a relative % difference from compared to integers. 2nm to 1.4nm is the same % as 7nm to 5nm, yet is seems so much less significant.

    are ~logarithmic equivalents, not linear scales

    proceso (nm)note i vs i-1nodo i vs nodo 45 nm4511280,62x1,61x220,79x2,05x140,64x3,21x70,50x6,43x50,71x9,00x30,60x15,00x1,40,47x32,14x
    Reply
  • AtrociKitty
    peachpuff said:
    Intel: pfffft our 10nm is just as good...
    I'm glad Intel rebranded, because too many people misunderstood the actual ranking of nodes when you compared them based on half-pitch scaling of an equivalent (in transistor density) planar transistor node:

    15. 32nm Intel 14. 28nm TSMC / 28nm UMC / 28nm Samsung/GlobalFoundries/IMB 13. 22nm Intel /22nm IBM 12. 20nm TSMC /20nm Samsung/20nm Intel (marketed as “22FFL) 11. 18nm TSMC (marketed as “16nm”) / 18nm Samsung/GF (marketed as “14nm”) 10. 17nm GF (marketed as “12nm” (12LP) by GF) 09. 16nm TSMC (marketed as “12nm” (12FFC) by TSMC) 08. 14nm Intel / 14nm Samsung (marketed as “10nm” by Samsung) 07. 13nm Samsung (marketed as “8nm” by Samsung) 06. 10nm TSMC (marketed as “7nm” (N7/N7P))/ 10nm Samsung (marketed as “7nm”) 05. 9nm Intel (marketed as “10nm”)/ 9nm TSMC (marketed as 7nm (N7+)) 04. 6.7nm TSMC (marketed as “5nm”) 03. 6.4nm Intel (marketed as “7nm”) 02. 5nm TSMC (marketed as “3nm”) 01. 4.5nm Intel (marketed as “5nm”)
    Reply
  • gfg
    is logarithmic...
    65 vs 45 = ~ - 30% area...
    7 vs 5 = ~ -30% area...
    2 vs 1.4 = ~ -30% area

    65-45= 20 nm and 2 vs 1.4 = just 0.6 oooo noooo.... by -30% node by node....
    Reply
  • tennis2
    peachpuff said:
    Intel: pfffft our 10nm is just as good...
    "But is it REALLY 1.4nm???
    Reply
  • derekullo
    TSMC to Initiate Quark Process Technology R&D.

    Physicists puzzled.
    Reply
  • Kamen Rider Blade
    If you're going to use decimal points, shouldn't we just migrate to Angstroms at that point?
    Reply
  • tennis2
    Kamen Rider Blade said:
    If you're going to use decimal points, shouldn't we just migrate to Angstroms at that point?
    IIRC that's what Intel is doing.
    Reply
  • Kamen Rider Blade
    tennis2 said:
    IIRC that's what Intel is doing.
    True, Intel is going to do that. What about the rest of the foundaries, when are they going to start using Angstroms?
    Reply