Monday, February 1st 2021

AMD Reportedly in Plans to Outsource Partial Chip Production to Samsung

It's been doing the rounds in the rumor mill that AMD is looking to expand its semiconductor manufacturing partners beyond TSMC (for the 7 nm process and eventually 5 nm) and Global Foundries (12 nm process used in its I/O dies). The intention undoubtedly comes from the strain that's being placed on TSMC's production lines, as most foundry-less businesses outsource their wafer production to the Taiwanese companies' factories and manufacturing processes, which are currently the industry's best. However, as we've seen, TSMC is having a hard time scaling its production facilities to the unprecedented demand it's seeing from its consumers. The company also has recently announced it may prioritize new manufacturing capabilities for the automotive industry, which is also facing shortages in chips - and that certainly doesn't instill confidence in capacity increases for its non-automotive clients.

That's what originated form the rumor mill. Speculating, this could mean that AMD would be looking to outsource products with generally lower ASP to Samsung's foundries, instead of trying to cram even more silicon manufacturing onto TSMC's 7 nm process (where it already fabricates its Zen 3, RDNA 2, EPYC, and custom silicon solutions for latest-gen consoles). AMD might thus be planning on leveraging Samsung's 8 nm or even smaller fabrication processes as alternatives for, for example, lower-than-high-end graphics solutions and other product lines (such as APUs and FPGA production, should its acquisition of Xilinx come through).
Take this information with a grain of salt; but all things considered, it does make sense that AMD outsources less-critical components to lesser-performing (and thus lesser-requested) manufacturing processes in a bid to keep up with demand on its best sources of revenue, without diluting their production even further. Remember that AMD does have an agreement in place with Samsung for licensing of its RDNA 2 graphics architecture for integration in the South Korean company's mobile Exynos chips.
Sources: via Hardware Info, Twitter
Add your own comment

25 Comments on AMD Reportedly in Plans to Outsource Partial Chip Production to Samsung

#1
Mysteoa
I have expected this move since they made colab with Samsung for the exynos gpu.
Posted on Reply
#2
ZoneDymo
Mysteoa
I have expected this move since they made colab with Samsung for the exynos gpu.
Same, with chiplet design you dont have to have every part at the most highend 7 - 5 - 3nm, so you can send that which does to TSMC and the other parts that can be done on simpler tech to Samsung or Global Foundry, reducing manufactuering cost, increasing yields and therefor reducing product cost
Posted on Reply
#3
jeremyshaw
Mysteoa
I have expected this move since they made colab with Samsung for the exynos gpu.
AMD already made Polaris GPUs on Samsung nodes (14nm family), so it would be more of a continuation of existing business, IMO.
Posted on Reply
#4
Mysteoa
jeremyshaw
AMD already made Polaris GPUs on Samsung nodes (14nm family), so it would be more of a continuation of existing business, IMO.
Wasn't it in GloFlo using Samsung tech?
Posted on Reply
#5
Vya Domus
Bad news, I guess it was unavoidable, AMD is becoming more and more popular but TSMC's volume of wafers limits profit. I just hope only low end products are built on these nodes.
Posted on Reply
#7
dicktracy
The desperation is real. Using TSMC was the only advantage they had versus their competitors.
Posted on Reply
#8
TheinsanegamerN
Anyhting to help alieve the bottleneck of new chips is good. I suspect we'll have a glut of new chips, and resulting low prices, in 12 ish months.
Vya Domus
Bad news, I guess it was unavoidable, AMD is becoming more and more popular but TSMC's volume of wafers limits profit. I just hope only low end products are built on these nodes.
It'd likely be mobile products with lower clock speeds, where samsung's power usage is pretty competitive with TSMC.
Posted on Reply
#9
john_
Considering that Abu Dhabi is the main shareholder of Global Foundries, I wonder if they are getting ready for some public executions. I mean those who decided to pull the plug from the 7nm process just a few years before the world was begging for 7nm capacity.
Posted on Reply
#10
ZoneDymo
dicktracy
The desperation is real. Using TSMC was the only advantage they had versus their competitors.
Well that and just being better then everyone else.
Posted on Reply
#11
TheinsanegamerN
john_
Considering that Abu Dhabi is the main shareholder of Global Foundries, I wonder if they are getting ready for some public executions. I mean those who decided to pull the plug from the 7nm process just a few years before the world was begging for 7nm capacity.
They were well aware that 7nm would be in demand back then, but GloFo was having financial issues, and it was likely they would never see a return on the 7nm R+D investment. I could see them lisencing out samsung's 8nm in the future though. That being said, GloFo has plenty of business with things like networking equipment and industrial controllers where you dont need the latest in nodes.
Posted on Reply
#12
chris.london
dicktracy
The desperation is real. Using TSMC was the only advantage they had versus their competitors.
It is not desperation. Samsung‘s 5nm node is at the very least better than TSMC’s 7nm otherwise Qualcomm wouldn’t have moved the SD888 to it.
Posted on Reply
#13
TheinsanegamerN
chris.london
It is not desperation. Samsung‘s 5nm node is at the very least better than TSMC’s 7nm otherwise Qualcomm wouldn’t have moved the SD888 to it.
By that logic Samsung's 8nm couldnt be any worse then tsmc 7nm or elsenvidia wouldnt have moved to it. Qualcomm may very well be chasing lower prices or less contested production.

Samsung 5nm isnt being used in any big chips yet, we're stillnot sure how it would scale to such a large chip.
Posted on Reply
#14
TechLurker
I'm not surprised; if fact, I was wondering why they didn't tap them a bit sooner, since Samsung's 8nm is only a bit inferior but plenty sufficient for mid-range and lower.

That being said, it's still a shame that GloFo never did risk 7nm; their was showing more potential than TSMC's in early testing, and they could have also joined AMD in a financial revival. I do wonder if they will reconsider in the coming years; as 7nm is going to remain in demand for the foreseeable future.
Posted on Reply
#15
xkm1948
Maybe one day it will go back to GloFo
Posted on Reply
#16
RandallFlagg
I've been saying a while that AMD needed to diversify their fab contracts. Their growth is hard capped at what they can get from TSMC right now, and they are competing not only with Apple and Qualcomm but now Intel as well for that limited TSMC capacity.

Now, Samsung 7nm EUV has been in production for ~2 years now. Most folks here don't know that because their head is buried in PC x86 desktop wars.

The IBM Power10 is made on that node and if a picture were used to describe 'fat, hot, powerful cpu' that would be it. 18B transistors on 602mm2 die, 30 cores with SMT8 *yes 8 threads per core*, 120MB L3 cache, and I believe I read a 400W TDP. That thing also supports DDR3, DDR4, DDR5, GDDR6, and HBM. It can have memory bandwidth up to 800GB/s vs 46GB/s for an Intel desktop CPU (using JEDEC RAM). For comparison, the 10 core SMT2 10900K is just over 200mm2 in size.

So Samsung is quite capable of making big high core count CPUs.
Posted on Reply
#17
Kaotik
You forgot CDNA manufacturing. I'm also pretty sure Vega20 is still produced, as well as RDNA1's
Posted on Reply
#18
Camm
chris.london
Samsung‘s 5nm node is at the very least better than TSMC’s 7nm otherwise Qualcomm wouldn’t have moved the SD888 to it.
It should be noted that Qualcomm moving to Samsung fabs isn't exactly about process superiority, it has a lot to do with a tonne of cross license and minimum use shens.
Posted on Reply
#19
Minus Infinity
Is there any indication TSMC is going to build more fabs? It's crazy for a company to be rate limiting themselves when they are so in demand. I know it takes years to build a fab, but surely they can't be so stupid to have not had any forward planning. This didn't just happen overnight.

What happened to global foundries are they not in the picture at all anymore?
Posted on Reply
#20
RandallFlagg
Minus Infinity
Is there any indication TSMC is going to build more fabs? It's crazy for a company to be rate limiting themselves when they are so in demand. I know it takes years to build a fab, but surely they can't be so stupid to have not had any forward planning. This didn't just happen overnight.

What happened to global foundries are they not in the picture at all anymore?
Most of the time, chip makers don't make new fabs. They "retool" existing fabs for the new nodes.

TSMC is building a new fab, or maybe more correctly a "back end process" facility. It is supposed to come online late 2021.

www.businesskorea.co.kr/news/articleView.html?idxno=52647

TSMC is also saying they expect 5nm capacity to increase from around 60,000 wafers per month to 100,000 by the end of 2021 :

www.gizchina.com/2021/01/05/tsmc-to-increase-its-5nm-production-capacity-from-60000-to-100000-per-month/

However, they are also saying that any additional capacity they get will be prioritized for automotive. The shortage is actually starting to halt auto production :

www.cnbc.com/2021/01/25/taiwan-ministry-says-tsmc-will-prioritize-auto-chips-if-possible.html

So there's a pretty clear picture of the official media line.

What I don't get, is why suddenly we don't seem to be able to use all these other fabs on 28nm / 16nm / 14nm etc. We haven't had these kinds of issues in the past, and demand in many sectors is down. I know that auto sales regressed, 2020 was quite a bit below 2019 sales. So did phone sales.

Maybe it's just the boom in PC Desktop/Laptop sales that's driving it, that affects not just CPUs but also support chipsets of all kinds like sound and network. Might also be multiple 'supercomputer' projects that suck up so much capacity.
Posted on Reply
#21
TheoneandonlyMrK
I'm not arsed what anybody uses for what if it gets more hardware out, crack on.
Sure would ease comparisons too.
Posted on Reply
#22
watzupken
I think it is a matter of time they seek to diversify. TSMC is obviously the first choice, but given the bad bottleneck over at TSMC, it doesn't make sense to stick purely with them. Unless you don't mind not selling anything for the year or you have ultra deep pockets like Apple, then sticking to TSMC works.
Minus Infinity
Is there any indication TSMC is going to build more fabs? It's crazy for a company to be rate limiting themselves when they are so in demand. I know it takes years to build a fab, but surely they can't be so stupid to have not had any forward planning. This didn't just happen overnight.

What happened to global foundries are they not in the picture at all anymore?
They may, but again, not all users of TSMC are considered strategic partners. The likes of Intel may start using TSMC in a larger scale, but that is purely because their 7nm is falling behind. At some point, Intel will leave, and if you have invested heavily in new fabs, you will have a lot of idle resources which is no good from a profit standpoint when you sunk a lot of dollars into the equipments and plants.

GF went silent after they introduced their 14nm. I don't think they have gone with the cutting edge fab.
Posted on Reply
#24
RandallFlagg
So ah, worldwide chips shortages to multiple centibilliondollar companies (made that up on my own) are limited by capacity of a small shop in Tokyo with about 300 employees (Ajinomoto Fine-Techno Co).

I wonder if they have publicly traded stock.
Posted on Reply
#25
TheoneandonlyMrK
I personally would expect AMD to absolutely use Samsung at some point to reduce the choke but think they'll design something for Samsung's nodes, despite similarity in nodes I doubt that their present design's would translate well to Samsung however the new console's will be updated quite regularly even while staying identical in specs etc, all to reduce BOM costs, this is the way.
Or the extra chips could be supportive chips like the recent ML chiplet or Io die.
That aligns quite well with this news but it is. Just an opinion.
Posted on Reply
Add your own comment