Array
(
    [content] => 
    [params] => Array
        (
            [0] => /forum/index.php?threads/samsung-fab-equipment-overlap-for-28nm-and-14nm.5178/
        )

    [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] => 2021370
            [XFI] => 1050270
        )

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

Samsung Fab Equipment Overlap for 28nm and 14nm?

benb

Well-known member
One more thought, perhaps the insomnia talking, but consider this:

-Fab8 cost: ~$8B
-Fab14P7 cost: ~$5B
-Austin Fab2 cost: 0; reusing fully depreciated 28nm equipment.

If Samsung is really ramping 14nm in Austin, using 28nm equipment, it will have a TREMENDOUS cost advantage over GF and TSMC. A really decisive advantage.
 
One more thought, perhaps the insomnia talking, but consider this:

-Fab8 cost: ~$8B
-Fab14P7 cost: ~$5B
-Austin Fab2 cost: 0; reusing fully depreciated 28nm equipment.

If Samsung is really ramping 14nm in Austin, using 28nm equipment, it will have a TREMENDOUS cost advantage over GF and TSMC. A really decisive advantage.

If we assume the litho is NXT:1950i's then although feasable they would need a major update and would still be about 30% slower than brand new NXT:1970i's which also have 20% better overlay performance. Best to keep the existing equipment for 28nm designs (FD-SOI anybody ?)
 
If we assume the litho is NXT:1950i's then although feasable they would need a major update and would still be about 30% slower than brand new NXT:1970i's which also have 20% better overlay performance. Best to keep the existing equipment for 28nm designs (FD-SOI anybody ?)

Samsung 28nm is gate first HKMG, FD-SOI 28nm is gate-first HKMG so there is major equipment overlap. TSMC 20nm and TSMC 16nm have a 90% overlap (per TSMC). I do not know the Samsung 28nm HKMG / 14nm FinFET equipment overlap number but I highly doubt it is close to the TSMC 20nm/16nm 90%. Can someone make an educated guess please?

<script src="//platform.linkedin.com/in.js" type="text/javascript">
lang: en_US
</script>
<script type="IN/Share" data-counter="right"></script>
 
Last edited:
Samsung 28nm is gate first HKMG, FD-SOI 28nm is gate-first HKMG so there is major equipment overlap. TSMC 20nm and TSMC 16nm have a 90% overlap (per TSMC). I do not know the Samsung 28nm HKMG / 14nm FinFET equipment overlap number but I highly doubt it is close to the TSMC 20nm/16nm 90%. Can someone make an educated guess please?

And of course the Samsung 28nm process was only introduced to Austin for the Apple A7 and thus is quite likely to be the same equipment first used for the 32nm Apple A6. Half of the equipment cost is litho (according to GF) so if that isn't suitable the number has to be below 50%. Plus it won't be quite fully depreciated yet anyway.
 
Last edited by a moderator:
The reports in the Korea media were of 14nm wafer starts in Austin. It sounds like those starts are more likely 28nm FD-SOI than 14nm, if not a complete fabrication.

My understanding is that immersion ArF lithography has been used since 32nm at most leading logic fabs, and this was a major capital investment. The 14nm lithography is the same, immersion ArF, just with double patterning. To get better ROI, maybe, just maybe, Samsung has figured out how to extend the old generation immersion ArF tools to 14nm. Keep in mind Samsung has DRAM and NAND wet litho experience to draw on that other logic manufacturers do not. NAND and DRAM have been using wet lithography much longer than logic.

One more litho-related question: Given the delays of EUV, why is there no interest in 157nm lithography?
 
Last edited:
The reports in the Korea media were of 14nm wafer starts in Austin. It sounds like those starts are more likely 28nm FD-SOI than 14nm, if not a complete fabrication.

My understanding is that immersion ArF lithography has been used since 32nm at most leading logic fabs, and this was a major capital investment. The 14nm lithography is the same, immersion ArF, just with double patterning. To get better ROI, maybe, just maybe, Samsung has figured out how to extend the old generation immersion ArF tools to 14nm. Keep in mind Samsung has DRAM and NAND wet litho experience to draw on that other logic manufacturers do not. NAND and DRAM have been using wet lithography much longer than logic.

One more litho-related question: Given the delays of EUV, why is there no interest in 157nm lithography?

I mentioned in my earlier post that there is an upgrade kit for the NXT:1950i so it is possible to do this but it still has a lower throughput and poorer overlay performance so it's far better to buy shiny new NXT:1970i's and move the older machines to less critical layers, either in FEOL or BEOL. I agree Samsung have lots of experience with immersion but Intel and TSMC weren't that far behind and in an industry where progress is measured in weeks I suspect they can be regarded as equals.

Proposals for 157nm went the way of the dinosaurs once TSMC invented immersion lithography as it was supposed to extend 193nm enough until EUV was ready. Maybe evolution gets things wrong sometimes :D But I can't see anybody wanting to invest in a short term measure now.
 
One more litho-related question: Given the delays of EUV, why is there no interest in 157nm lithography?

157nm was looked at over two separate investigation periods and finally rejected as having poor cost/benefit ratio, and because of major problems getting enough high-quality CaF for the optics, among other issues. A few tools were built, but the potential gain over 193i did not justify the cost.
 
Thanks for the comments on 157nm litho. Googling "193nm optics" I found reference to CaF optics, it mentions "157nm grade" option. That makes it sound like 157nm is an extension of 193 (but I know nothing about litho). Maybe 157 isn't compatible with immersion?

Back to the topic of overlap. TSMC devised a clever way to overlap 16nm and 20nm, sharing 90% of the equipment, at the cost of some density. That provides two advantages: Cost, obviously, but also manufacturing stability and yield learning, plus routing flexibility (a fab with more flexibility can reduce cycle times and operate more efficiently). Samsung also has 20nm in production, raising the question of how much their 20nm and 14nm equipment overlaps.

My questions are: What are implications of 90% overlap for competition (if you don't match it, are you still competitive?), for equipment vendors (this can't mean anything good for TSMC's equipment vendors), and the future (will 10nm also have high-percent-overlap, more than 50%). Also, beyond litho, what equipment fits into that last, irreducibly important 10% that must be upgraded at every node?
 
TSMC devised a clever way to overlap 16nm and 20nm, sharing 90% of the equipment, at the cost of some density. That provides two advantages: Cost, obviously, but also manufacturing stability and yield learning, plus routing flexibility (a fab with more flexibility can reduce cycle times and operate more efficiently). Samsung also has 20nm in production, raising the question of how much their 20nm and 14nm equipment overlaps.

My questions are: What are implications of 90% overlap for competition (if you don't match it, are you still competitive?), for equipment vendors (this can't mean anything good for TSMC's equipment vendors), and the future (will 10nm also have high-percent-overlap, more than 50%). Also, beyond litho, what equipment fits into that last, irreducibly important 10% that must be upgraded at every node?

From 20nm to 16nm, the 10% of non-reused equipment is probably mostly new litho machines for the most critical layers, with older machines shuffling out to less critical layers. Other equipment that would need upgrade includes metrology - a critical in-line part of the things once you adopt double-patterning - and implant due to the move to FinFETs, but neither is on a par price-wise with litho.

But I think the thing to consider is you generally aren't upgrading just a half-node. For instance TSMC still has to produce 20nm for several years as even Apple will need more wafers and other designs generally have a longer lifespan, and probably much longer for 28nm. Thus it is probably older modules that get major upgrades to 14/16nm, or you build new modules. I think TSMC's comment on 90% overlap is more to show they have understanding of the process technology than of actual equipment reuse, although obviously test wafers would be done on the 20nm lines whilst other lines are upgraded. At the end of the day being a fab is expensive and there are few cost-cutting measures or magic bullets. As Jerry Sanders famously said it takes real men to have fabs and Morris Chang certainly meets that score.
 
Back to the topic of overlap. TSMC devised a clever way to overlap 16nm and 20nm, sharing 90% of the equipment...
Benb,

In 2015, very little, if any, 20nm capacity will be converted to 16nm. A8 production may decline, but many others will come online, such as QCOM SD 810, NVDA Integra X1, AMD, MediaTek, etc.

The 90% equipment overlap between 20nm and 16nm is a result of the phrased approach at TSMC. I wrote about this before; see the two links below.

Phrased approach and high initial yield of 16FF+
https://www.semiwiki.com/forum/f293/samsung-strikes-chip-deal-apple-4864-2.html#post16998

https://www.semiwiki.com/forum/f293/samsung-strikes-chip-deal-apple-4864-3.html#post17016

Furthermore, old processes usually don’t go away at TSMC, unlike the Intel model. Linked below:

TSMC’s processes vs. the Intel model
https://www.semiwiki.com/forum/f2/samsung-strikes-chip-deal-apple-4864-4.html#post17097
 
157nm was looked at over two separate investigation periods and finally rejected as having poor cost/benefit ratio, and because of major problems getting enough high-quality CaF for the optics, among other issues. A few tools were built, but the potential gain over 193i did not justify the cost.

Small nitpick. 157nm was investigated before 193 immersion as the scanner manufacturers thought it the engineering was easier. (I still remember the discussion here when Bruce Smith of RIT came to talk here at imec about his immersion small field demo tool and the ASML people wondered if they had now to put their machines in a swimming pool or so.) 193i killed 157 when 157 still had big engineering problems and the immersion ones seemed to be solvable. Actually 193i immersion is an improvement over 157 as wavelength in water of 193nm light is 193/1.44=134nm with 1.44 refractive index of water.
 
134/4-109/4=33.5-27.25=6.25nm 6.25/109=5.7%
157 stops due to multiple patterning. The gain after multiple patterning by introducing 157i is ~5%, is too little for a single tech node step.
 
I am not sure but I thought 157nm light is absorbed by water. Optical path in 157nm already had to be nitrogen purged as air is too absorbing; EUV is done in near vacuum.
 
Back
Top