[HN Gopher] Thinner Films Conduct Better Than Copper
___________________________________________________________________
Thinner Films Conduct Better Than Copper
Author : rbanffy
Score : 56 points
Date : 2025-03-27 13:59 UTC (9 hours ago)
(HTM) web link (spectrum.ieee.org)
(TXT) w3m dump (spectrum.ieee.org)
| sopchi wrote:
| The research described was done at Stanford by Profs Eric Pop and
| Krishna Saraswat. Paper in Science:
| https://www.science.org/doi/10.1126/science.adq7096
| sbierwagen wrote:
| Also posted on arxiv: https://arxiv.org/abs/2409.17337
| bee_rider wrote:
| The title here is sort of confusing-thinner films? Thinner than
| what? And what are these thin films made of?
|
| Maybe something like "thin film semi metals conduct better than
| copper" would be better?
| motorest wrote:
| > The title here is sort of confusing-thinner films? Thinner
| than what? And what are these thin films made of?
|
| You answer all your questions if you read the first paragraph
| of the article. The second paragraph even specified the exact
| material and thickness.
| themaninthedark wrote:
| The only thing I think that the title could make clearer is
| that this is at or below nm scale, but that too is in the
| first paragraph.
| NikkiA wrote:
| "New niobium thin film conducts better than copper" would have
| been my choice, but it would probably get butchered by the
| autofilter to something as useless as the current title anyway,
| because 'new' is probably elided, as would 'than'.
| jonasenordin wrote:
| Conductivity: Film at Eleven
| infogulch wrote:
| Cool! Or lukewarm. Or both since they tested from 5K to 300K. And
| found that the thin films (5-18nm) 'behaved like' metals wrt
| temperature dependence of resistivity, where thicker films of the
| same material (80nm) 'behaved like' insulators.
|
| I wonder how traditional superconductor materials would fare in a
| super thin film regime like this.
| larkost wrote:
| The linked article does not go into many details about what they
| are measuring. So I have to wonder: could this property be used
| to make thinner wires for conducting electricity in bulk (e.g.:
| hosing wires, or transmission lines) by stacking lots of very
| thin layers into a cable. I imagine there would be a good amount
| of development work to make manufacturing such a layered cable,
| so it would not be right around the corner. But is there any
| physical barrier to such a development?
| lawlessone wrote:
| heat might be an issue
| croemer wrote:
| Correct title is "New Films Conduct Better the Thinner They Get"
| deepsun wrote:
| I wonder if that's only about direct current. The higher
| frequency is AC, the more it conducts on sides of the wire (hence
| insulated strands). So I wouldn't be surprised by their discovery
| if we talked about very high frequency AC.
|
| But since it all reversed here it might by opposite for AC in
| their case.
|
| UPDATE: paper says something about RF (radio frequency), but I'm
| not sure I understand what it means, looks more like
| manufacturing process.
| adrian_b wrote:
| Because copper is no longer a good enough conductor when it it is
| used in very thin layers, the latest and densest CMOS fabrication
| processes have begun to replace copper with other metals for the
| first 2 metal layers, which are the thinnest.
|
| It is not known with certainty what metals are currently used by
| Intel or TSMC, though cobalt and ruthenium have been considered
| as the most promising choices.
|
| There has been a lot of speculation about whether the choice of
| cobalt for the thinnest metal layers has been an important cause
| of Intel's woes with their "10 nm" CMOS fabrication process.
|
| While there were chances for this hypothesis to be true, the
| actual reasons for Intel's failure to achieve the predicted clock
| frequencies and fabrication yields with their "10 nm" processes
| remain unknown, because Intel has never published any information
| about this.
|
| Whichever was the reason, eventually Intel has succeeded to fix
| this fabrication process, even if only around 5 years later than
| in their initial plan, after rebranding it in "Intel 7", at least
| from the point of view of the achievable performances, which have
| culminated in Raptor Lake Refresh, though perhaps Intel's
| incapacity of predicting accurately the reliability behavior as a
| function of the supply voltage, as demonstrated in many failures
| of Raptor Lake/Alder Lake CPUs, may indicate that they have never
| succeeded to understand the exact characteristics of this
| fabrication process.
| q3k wrote:
| > It is not known with certainty what metals are currently used
| by Intel or TSMC, though cobalt and ruthenium have been
| considered as the most promising choices.
|
| Wait, shouldn't that be easy to find out? Section/delayer a
| chip and throw it under a SEM-EDX system?
| hnuser123456 wrote:
| Alright, who here has the delidding tools and SEM? Let's
| crack this egg
| q3k wrote:
| I mean I have access to SEM with EDX [1] and I can (poorly)
| crack open chips [2], 'just' need to get some LN to run the
| EDX... I just can't believe this isn't something that
| another lab has already done and published!
|
| [1] - https://wiki.fa-fo.de/equipment:zeiss-dsm-962 ; it's
| still not 100% restored
|
| [2] - Here's what I believe to be metal1 on a RP2350:
| https://object.ceph-
| eu.hswaw.net/q3k-personal/c98c23b8db73df...
| IlikeKitties wrote:
| Der8auer on Youtube is the most likely candidate to figure it
| out if he so choses.
| dieselerator wrote:
| The research looks detailed and interesting. However, I don't
| follow this summary article.
|
| Digital circuits dissipate most of the energy charging and
| discharging capacitance. It must necessarily dissipate that as
| heat (except for a minor amount of electro-magnetic radiation).
| The interconnect resistance hardly matters. Of course RC relay
| can be a factor for some circuits. We can hope this reasearch
| leads to improvement there.
|
| Power supply bus resistance can lead to voltage drops, but this
| research apparently studies layers much too thin for that
| application.
|
| Did I missing something?
| mystified5016 wrote:
| The interconnect resistance _does_ matter as resistance is a
| function of cross-sectional area. It 's related to the physical
| size of the conductor. Lower resistance conductors can be
| physically smaller while carrying the same amount of current.
|
| But the real trick is if you can increase your switching speed,
| you lose less energy in the transistor. All the time in between
| 0 and 1, the transistor is burning energy as heat rather than
| conducting current. Lower R in your interconnect means your RC
| time constant goes down and your switching speed goes up. Your
| transistor spends less time in the linear region and wastes
| less energy.
|
| But yes, these are pretty small effects on the whole. That's
| really just where the industry is at: incremental improvements
| until the Next Big Thing comes along.
|
| Additional nit: up to 50% of the energy put into gate
| capacitance _could_ be recovered. It 's not _necessary_ to
| waste 100%, it 's just dramatically cheaper and easier.
| Honestly I doubt there's any practical benefit as the chip
| would become quite a lot larger.
| rbanffy wrote:
| > Honestly I doubt there's any practical benefit as the chip
| would become quite a lot larger.
|
| There will be at some point when area becomes too small to
| accommodate the heating and the added complexity becomes a
| way to shrink the entire chip to sizes that wouldn't be
| possible otherwise.
___________________________________________________________________
(page generated 2025-03-27 23:01 UTC)