Array
(
    [content] => 
    [params] => Array
        (
            [0] => /forum/index.php?threads/have-finfet-foundry-nodes-fallen-in-cost.14027/
        )

    [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] => 2020470
            [XFI] => 1040170
        )

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

Have FinFET foundry nodes fallen in cost?

BlueNode

New member
Hi all – Is it cheaper to go with FinFET process nodes in 2021 than it was a few years ago? I mostly mean the initial foundry nodes, and maybe one or two shrinks beyond that, coming up just shy of the current cutting edge nodes (5nm). So for example:

-- TSMC 16/12nm and Samsung 14nm (presumably their nth-generation versions of it, like 14LPP or the "14LPU" that I heard about somewhere)
-- TSMC 10nm and Samsung 10/8nm (same assumption about mature versions like 10LPU or 8LPU)
-- TSMC 7nm DUV and maybe Samsung's 7nm EUV?

There was always supposed to be a large upfront, non-recurring cost to move down to FinFET nodes, compared to the best planar options like 28nm (and now there are "22nm" planar nodes that improve on 28nm). Has the delta shrunk? Has the absolute cost of designing a new product in FinFET nodes shrunk at all, since they were introduced by the foundries?

I think TSMC didn't want their 10nm to become a long node, so I imagine they steer people away from it these days. Do you think their 7nm nodes are comparable in cost to 10nm, or maybe that in a couple of years they will be?

Some of the nodes out there seem to be designed specifically for migration from an earlier, larger node, not for new products, like Samsung's 11LPP, which is designed for existing customers on one of their 14nm nodes. I'm most interested in scenarios where you can design an ASIC or SoC from scratch, maybe a new family of products. I guess one consideration would be whether you see a product line having to move to FinFET eventually anyway, so then it becomes of question of pay now or pay later.

Interestingly, some of the gains with new planar nodes look so good that I wonder if they actually match 16/14nm processes now. e.g. TSMC's 22nm is supposed to be a 30% performance or power improvement over 28HPC. That sounds similar to the improvement you'd expect from the first-gen FinFET nodes, the 16/14nm portfolios. (Especially since I think 28HPC is already a significant boost over earlier 28nm nodes – I wonder what 28nm reference points the 16/14nm nodes were compared to back in 2015 or whenever.) Some of the SOI processes look good to, the 20-something nm ones, but I wonder if they cost as much as bulk FinFET. Anyway, have you seen or heard of a fall in FinFET costs over the last few years? Will FinFET ever be as cheap as planar, say by 2025 or so?

Thanks.
 

IanD

Active member
Depends what you mean by "Will FinFET ever be as cheap as planar"... ;-)

As cheap for upfront NRE/mask costs -- no, design and manufacture are more complex and more costly.

As cheap per wafer -- no, see above.

As cheap per gate -- yes (they're already cheaper) because of higher gate density, though this is largely driven by metal pitch not the fact that FinFETs are used. But of course all the fine-pitch processes *are* FinFET...

The issue is that the lower cost *per gate* (or per chip with a given function) only applies if you order a *lot* of silicon, meaning big chips or high volumes or both. Saving a few dollars off the die cost when the NRE cost goes up by millions of dollars doesn't always give cheaper chips...

There's also TTM (time to market) to consider; it takes longer to do the design/verification in advanced FinFET nodes and the manufacturing TAT is longer, so the risk is that you might end up with a theoretically cheaper die but your competitor who got there before you gets 80% of the market and you get 20%, which drops their costs and brings in more profit.

There are a few markets where bleeding-edge technology is worth it (mobile phones, CPUs, networking) and these tend to be the ones that make the headlines; there are far more where the added cost and delay means a less advanced process is better, either older FinFET or planar depending on the particular design.

All process costs fall with time, but there's still a "right" process choice for a design which is rarely very old or very new...
 
Top