Not Everyone Needs Leading Edge: TSMC’s 22 nm ULP, 12 nm FFC and 12 nm FFC+

Now let’s discuss something less advanced, but what is required for hundreds of millions of devices sold every year.

Advertised PPA Improvements of TSMC's Low-Power/Compact Nodes
Data announced by TSMC during conference calls, press briefings and in press releases
  CLN28HPC
vs
CLN28HPM
CLN28HPC+
vs
CLN28HPM
CLN22LPU
vs
CLNHPC+
CLN16FFC
vs
CLN16FF
CLN12FFC
vs
CLN16FFC
12FFC-ULP
vs
CLN12FFC
Power 20% 30% 35% lower 25% lower
Performance - 15% 15% unknown 10% unknown
Area Reduction 10% 10% 10% optional 20% unknown
HVM Start started started 2018 Q1 2016 2018 2019
Note Planar
28 nm-based
FinFET
16/20 nm-based

Development of FinFET-based chips is more expensive of ICs featuring planar transistors and their manufacturing is more costly as well. As a result, FinFET is virtually unavailable for many smaller designers of SoCs that usually build various solutions for emerging IoT applications. GlobalFoundries and Samsung offer their FD-SOI manufacturing processes to such companies (and these technologies have a number of other advantages in addition to being more cost-effective), whereas TSMC intends to introduce its new 22 nm ULP technology aimed at such applications. The CLN22ULP is an optimized version of the company’s 28 nm HPC+ (high-performance compact plus) manufacturing process that has been available for a while. The 22ULP offers a 10% area reduction and either a 15% performance improvement over the 28HPC+ process, or a 35% power drop. The 22ULP process joins a family of other ultra-low-power processes offered by TSMC and will compete against GlobalFoundries 22FDX as well as Samsung’s 28 nm FD-SOI offering.

Next up is TSMC’s 12 nm FFC manufacturing technology, which is an optimized version of the company’s CLN16FFC that is set to use 6T libraries (as opposed to 7.5T and 9T libraries) providing a 20% area reduction. Despite noticeably higher transistor density, the CLN12FFC is expected to also offer a 10% frequency improvement at the same power and complexity or a 25% power reduction at the same clock rate and complexity. Further down the road, TSMC also plans to offer a ULP version of the CLN12FFC with reduced voltage, but that is going to happen only in 2018 or 2019.

Sources: Samsung, TSMC, SemiWiki (123).

Related Reading:

Beyond 10 nm at Samsung: 8 nm and 6 nm
POST A COMMENT

89 Comments

View All Comments

  • Demon-Xanth - Friday, May 5, 2017 - link

    A silicon atom's width is about 110pm, so they are literally going into widths that are only double digit numbers of atoms wide. Reply
  • MananDedhia - Friday, May 5, 2017 - link

    Processes that deposit single atomic layers are already used commonly in processes - even in 28nm.....For some layers, atomic layer deposition is the only way to go. The complexity increased here because we are now increasing the number of layers and devices that need to be defined at those scales. Reply
  • bug77 - Friday, May 5, 2017 - link

    Yes, but you can't take 3 atoms, call then source, drain and gate and assemble them into a transistor. Reply
  • ddriver - Friday, May 5, 2017 - link

    Why not? IBM have already demoed atomic assembly. The downside - it is very slow, it is one atom at a time VS etching septillions of atoms with acid at the same time.

    BTW I finally get how they will get to 5 nm - by lying about it. How much of a 10nm chip's features are at 10nm resolution? Not many. Area decrease is already falling behind the process scale number and it is only going to get worse.

    On the upside - no biggie - we already have enough performance to run terminators. So our extinction is well assured.
    Reply
  • philehidiot - Friday, May 5, 2017 - link

    Don't you go worrying about terminators. I've already started work on the first of many. I was just so sick of not being able to get a seat on the bus. Was thinking no one wants to sit next to a cybernetic killing machine so I can send that to the earliest bus stop, get it to reserve a seat and I can ride to work without the smell of unwashed, practically rotting human being next to me.

    Far safer than one of those self driving car things. Bloody death traps.
    Reply
  • Kevin G - Sunday, May 7, 2017 - link

    Are you Sarah Conner? Reply
  • Xajel - Sunday, May 7, 2017 - link

    Yeah they're actually moving these actual atoms... atom by atom like a lego Reply
  • Kevin G - Sunday, May 7, 2017 - link

    The node size is mostly marketing now which is why Intel went out of their way to define some new metrics ~6 weeks ago. While I wasn't a fan of that marketing spiel, there is a point that there needs to be a new metric as traditional node shrinks are few and far between going forward.

    What I think the foundries are waiting on is a new big break through as they realize that they cannot currently continue on the existing path indefinitely. Germanium can come in as an exotic material as a substitute for silicon but wafer prices are extremely expensive. Even then, germanium doesn't even solve the node problem but rather just provides better material properties at existing nodes. Carbon nanotubes and graphene are two related materials seen as potential for replacing silicon as we get even closer to the atomic level. Both have some good properties for circuit design but no one has found a means of economical mass production.

    Both Intel and IBM has invested heavily into silicon photonics. So far their efforts have lead to advancements in IO but not raw processing but optical logic gates do exist. Much like other exotic solutions, these suffer from mass production problems to bring them out of the research lab. (Notice a trend starting here?)

    I think strategies like interposers and EMIB are emerging to side step the absolute need for shrinks in the sense of limiting transistor counts. Granted interposers/EMIB do nothing with regards to power consumption. The one nice thing about these techniques is that they do potentially allow for mixing some of the more exotic solutions with bulk processes. For example, a die with slicon photonics could interface with some high speed optical circuits in the package and also interfaces with more traditional bulk processes for its SRAM cache. Very expensive but worth considering when there are other new node alternatives available. Granted, such choices are not going to happen tomorrow but they're clearly on the horizon.
    Reply
  • eachus - Sunday, October 1, 2017 - link

    My read is that the first application of nanotubes or graphene will be laying down a copper layer, then growing graphene on top of it. The trick will be to get the graphene to align on top of the copper, which will probably take another layer in between, perhaps silver. Could silver be substituted for copper in bulk? Good question. It is a better conductor and solves the alignment problem.

    You may think of silver as a precious metal along with gold and platinum, but over fifty per cent of the silver mined goes into silver solder for brazing or soldering metals together. Most silver solder is used for brazing, go figure. Silver is also used in thermal compounds for getting a good seal between a CPU chip and the heat sink. Obviously replacing a few grams of copper with silver inside the chip won't raise prices significantly.

    Getting copper to bond to the graphene is not a problem--even if the reverse is a significant problem. However high-temperature processes may damage the graphene. Best is probably a "wet" process to put a thin layer of copper on the graphene before building the next litho layer. Putting the graphene in a copper sandwich like this should significantly improve the characteristics of the layer. This will show up as a reduced capacitance with adjacent conducting traces--less cross-talk and faster signal propagation.
    Reply
  • beginner99 - Monday, May 8, 2017 - link

    "BTW I finally get how they will get to 5 nm - by lying about it"
    Process tech numbering hasn't been about feature size for the past 2 decades.
    Reply

Log in

Don't have an account? Sign up now