Samsung Electronics boosts foundry utilization with increased production orders
Posted by self-fix@reddit | hardware | View on Reddit | 31 comments
Posted by self-fix@reddit | hardware | View on Reddit | 31 comments
BigBananaBerries@reddit
Intel enters the chat...
SherbertExisting3509@reddit
Intel has their Intel 16 node (upgraded 22FFL) for external use. It has one external customer using it in Europe, but so far, there haven't been any other customers.
AFAIK Inrel-UMC 12 is still in development.
Intel-7 is internal only and is unlikely to be used for external customers due to how expensive it is to make, and it's incompatible with industry standard PDK'S . (Likely due to its cobalt-copper alloy vias)
Intel 4 and 3 so far haven't seen any interest from external customers. Intel is using it for Meteor Lake and Xeon 6
External Interest in 18A has been drying up due to rumored delays and performance not being up to expectations compared to TSMC's N2. We don't know if that's true, but what we do know is that Intel hasn't secured any major customers for 18A yet, which is very concerning.
Did Intel set a bunch of money on fire over the past few years for no reason, or will Lip Bu Tan drag Intel's foundry business out of the dirt?
It will be interesting to see if he can save this thing or if he will end up having to kill it to save the company over the next few years.
EnglishBrekkie_1604@reddit
I’d wager the reasons 18A hasn’t gotten anyone yet is the fact that it’s inherently a bit risky, and that the PDK was totally fucked up and apparently really sucks, it was all outsourced on the cheap. The guy responsible for that mess (as well as the one responsible for the whole 10nm disaster) just got fired, so that’s definitely a sign Intel is trying to clean the gutters. Once 18A starts churning out chips, and the PDK gets better, that’ll be the real test.
blueredscreen@reddit
Intel has no choice but to get it working. But simply getting it working just means the company continues to exist, which is not that much of a bar to clear. Their culture was never designed to have their fabs produce for external customers, so the very concept of an industry-standard PDK is foreign to them.
Anxious-Shame1542@reddit
I agree. 18A wasn’t built with any external customer PDKs in mind despite former CEO Pat betting the farm on it. Also much of the fab culture is in the process of changing to industry standards to assign with external customer expectations. But these changes are coming late on the heels of 18A HVM ramp. So of course there’s little external customer interest on 18A.
SherbertExisting3509@reddit
Real galaxy brain move by Pat
"Let's bet the whole company on 18A and then outsource the PDK so that no one buys our node"
Reeks of the famous Intel(tm) arrogance that was at least in part responsible for:
10nm disaster
Losing the PS6 contract
Canceling Royal Core
Underfunding the product division (wonder why AMD is kicking Intel's ass right now? Guess where all the R and D money went?)
Ponte Vecchio (Why would you build a HPC dGPU with your first dGPU uarch???? facepalm 🤦♂️
NoRecommendation2761@reddit
>Intel 4 and 3
Intel 4 and 3 don't have a PDK, just like Intel 7.
>External Interest in 18A has been drying up due to rumored delays and performance not being up to expectations compared to TSMC's N2
From my understanding, it is not the case of performance not being up to expectations compared to TSMC's N2, but chips from tapeout don't match specifications that were initially promised to exteneral customers, making a PDK for 18A practically useless.
And that quoted prices were significantly higher than what potential customers initially thought.
fuji_T@reddit
it's wild that 4nm is considered legacy!
SilentHuntah@reddit
That line made me pause for a second too.
Also gives me some hope that Samsung will be open to porting the Switch 2 SoC to a lower node within the next 3-5 years.
SuperDuperSkateCrew@reddit
Realistically if they do a node shrink it will be to Samsungs 6nm. It’ll offer about a 15% boast to efficiency and it’ll be the most economical node to port to from 8nm, from what I remember you can make that shrink with very minimal changes to the SoC design.
burd-@reddit
isn't that on Nvidia and Nintendo? Nvidia has to port it and Nintendo just have to pay more.
AuthoringInProgress@reddit
What needs to happen is that the silicon costs needs to reach the right balance with a reduced chip size.
5nm silicon is going to cost more per mm2 than the 8N Nintendo is using now, but hopefully within a few years, as the silicon production matures a little more, they'll be able to fab a chip with identical performance to the current switch 2 soc but in a smaller size, in such a way that production costs are lower or the same as the current chip.
Its not going to happen as quickly as we hope, because the time it takes to shrink nodes and make it affordable is increasing exponentially. It's not there yet, at least not combined with the initial RnD it would take to port Ampere to a smaller process node, but hopefully this is a positive move towards it being there in a couple of years.
Strazdas1@reddit
So its never happening.
NoRecommendation2761@reddit
Highly unlikely. A profit margin on console SoC is slim and Nvidia won't re-design Tegra unless Nintendo pays them a big money. It is either Tegra staying on Samsung 8N or Nvidia designs a new chip on TSMC nodes.
Exist50@reddit
Not necessarily true for Nintendo consoles. And surely the volume is high enough to justify a redesign if desired.
PatchNoteReader@reddit
Hope it happens sooner. It took around 2 years for the original switch to be updated right?
JuanElMinero@reddit
That was when it was still relatively cheap porting to a newer node. Tegra X1 Mariko revision came in 2019, but fabbed on TSMC 16nm from 2015.
We all know how Nintendo feels about using any amount of money to implement up-to-date hardware.
Verite_Rendition@reddit
The situations are so different that I'm not even sure we can use the Tegra X1's history as guidance.
Contrast that with the fact the Samsung 8nm isn't going anywhere, porting the chip to a new node wouldn't deliver as much of an efficiency boost, and the cost of such a program would be much higher.
I could see this going either way. While there are benefits to porting, those benefits are fewer and the costs are higher this time around.
JuanElMinero@reddit
Aside from node porting, they could also go at it from the battery angle, once SiC anode designs are mature and widespread enough.
I'd hope they tackle it from both sides though, as the current battery life is simply not acceptable for me.
It took a big node jump from TSMC 28nm-16nm to make the Switch 1 okayish, at least in that respect. I don't expect something like Samsung 8nm-4nm to bridge that efficiency gap alone.
SherbertExisting3509@reddit
4nm would be so much more expensive that it might not be worth the decrease in margins for Nintendo.
If I were nintendo, I would put a higher mAh battery ( at least a 6000 mAh battery) and an OLED display for a mid-gen "upgrade".
Sell it at the original MSRP and then give a $100-$200 discount for the launch model to grab more customers looking for a cheaper next-gen handheld.
The Ryzen Z2A which is a rebranded steam-deck 6nm APU, could lead to a new generation of cheap handhelds made by OEM's and it's much more powerful than the original switch.
Nintendo would need to release something to compete with in that market segment if that situation happens.
JuanElMinero@reddit
6000mAh would be a 20% bump, so roughly like going from 2:00h to about 2:30h for the most demanding titles, going by current user reports.
Mariko gave Switch 1 a 50% increase for less demanding and up to 80% for the most demanding scenarios. Which is roughly the same as the Switch 2 needs to be properly mobile IMO.
Unfortunately, I also expect Nintendo to go with the ~6000mAh option and call it a day.
theholylancer@reddit
that and well...
everyone and their mother can buy a cheap 20,000 mAh portable battery, like the meme haribo ones (that are fairly good because they are trying to protect their brand), then it just make sense for it to be not as big of an issue if you can spend 25 bucks on that kind of power bank and just live off of it if you wanted to.
that kind of development is fairly recent, and i would wager to continue
so devices gets a 6k or w/e battery and the people spends a bit more for a 20k bank that is still portable.
Cheerful_Champion@reddit
This article clearly means legacy = mature. A bit weird terminology if you ask me.
haloimplant@reddit
Seems like it went from yield finally improving to legacy very quickly, but what does that designation even mean. The price must be good because there is a lot of interest in making new Sf4 chips from what I hear.
Blueberryburntpie@reddit
TSMC already fully amortized the R&D and construction costs of their 7nm production lines from a while back. By accounting standards, it's now a money printer.
996forever@reddit
5nm is from 2020!
NoRecommendation2761@reddit
Samsung can call its 4nm a legacy node since they actually ship chips manufactured on 3nm in millions (Fold 7) and SF4's PDK has been available for a while now and SF4 is actually now yielding. As long as SF4 is cheaper than TSMC N5\~N7, it will attract external customers.
This is what Intel should have done rather than being super ambitious of their unrealistic goals of developing so many new nodes within a short span of time.
Astigi@reddit
Samsung boosts foundry utilization producing at barely no margin
WarEagleGo@reddit
Ghostsonplanets@reddit
Switch 2 is 8nm
996forever@reddit
Switch 2 maybe