Saturday, November 17th 2018

Samsung AMD's Second Foundry Partner for "Polaris 30"

AMD's "Polaris 30" silicon at the heart of Radeon RX 590 graphics card is the company's first 12 nm GPU. Unlike NVIDIA, which is exclusively sourcing its "Turing" family of GPUs from TSMC, the "Polaris 30" is coming from not one, but two sources. This, according to AMD in response to a question by TechPowerUp. The two foundries manufacturing "Polaris 30" are GlobalFoundries and Samsung. AMD did not provide us with visual cues on how to tell chips made from either foundries apart (such as serial numbering schemes). Packaging of dies sourced from both foundries is done in China, and the national-origin marking for the chip is on the package, rather than printed on the die.

GlobalFoundries' 12 nanometer FinFET node, called GloFo 12LP, shares a lot of similarities with Samsung's 11LPP, because both are "nodelets" that are derived from an original 14 nm FinFET process blueprint Samsung licensed to GloFo, deployed in its facility in upstate New York, where AMD's "Zen" processors are made. GloFo's 12 nanometer process is a refinement of its 14 nm node, in which 12 nm transistors are etched onto silicon using the same lithography meant for 14 nm. It doesn't improve transistor densities, but provides dividends in power, which explains why "Polaris 30" and "Pinnacle Ridge" have the same die sizes as "Polaris 20" and "Summit Ridge," respectively. This WikiChip article provides a good explanation of how GloFo 12LP is a nodelet.
AMD appears highly motivated to be first-to-market with both 7 nm CPU and GPU, which come in the form of its "Vega 20" GPU and "Rome" CPU, which could be formally launched before the end of 2018, with production being ramped up through 2019. AMD confirmed publicly earlier, that its first 7 nm chips could be sourced from TSMC, but even it realizes that TSMC has an exhaustive list of clientele for 7 nm, each with its own foundry allocation. AMD could try to find additional sources for 7 nm chips, and we predict Samsung to be one of them. GloFo canned its 7 nm plans, and AMD is keeping it busy with wafer contracts for "Zen," "Zen+," "Polaris," "Vega 10" GPU die, and the 14 nm I/O controller die at the heart of its EPYC "Rome" MCM.
Add your own comment

41 Comments on Samsung AMD's Second Foundry Partner for "Polaris 30"

#1
theoneandonlymrk
It seams Samsung's ambitious project to claim 25% of foundrie work is going well, it makes you wonder what else Samsung is working on for Amd, they're 7nm is compatible with TSMC's AFAIK and chiplets do provide a large degree of foundrie flexibility.
Posted on Reply
#2
Vya Domus
Surprised Samsung's nodes are competitive enough for AMD to use them for their GPUs.
Posted on Reply
#3
stimpy88
OK, well we now have a silicon lottery regarding which chip we get, and which chip runs hotter, faster etc... Great AMD...

Samsung has the inferior process node and manufacturing quality. We found this with the iPhone...
Posted on Reply
#4
Totally
Doesn't anyone find it odd that Samsung doesn't have fab/foundry time for their own products but has no problem making time for other chip makers?
Posted on Reply
#5
HD64G
stimpy88, post: 3944078, member: 178509"
OK, well we now have a silicon lottery regarding which chip we get, and which chip runs hotter, faster etc... Great AMD...

Samsung has the inferior process node and manufacturing quality. We found this with the iPhone...
Samsung owns the know-how of the processes that GF use since 16nm to produce AMD's chips, so the product is identical (as much as possible since silicon lottery exists in the same factories' outcome after all) and only yields can be different based on who began earlier.
Posted on Reply
#6
Imsochobo
Totally, post: 3944094, member: 90126"
Doesn't anyone find it odd that Samsung doesn't have fab/foundry time for their own products but has no problem making time for other chip makers?
Samsung have capacity on older nodes, 10\7nm is where they have capacity issues and also dram\nand is different manufacturing again.
it's not like magically everything can do the latest node
Posted on Reply
#7
jeremyshaw
Vya Domus, post: 3944062, member: 169281"
Surprised Samsung's nodes are competitive enough for AMD to use them for their GPUs.
Why not? Nvidia already uses Samsung to exclusively manufacture the MX150, 1030, 1050, and 1050Ti. All chips that are quite good on the power/performance scale.
Posted on Reply
#8
Frick
Fishfaced Nincompoop
stimpy88, post: 3944078, member: 178509"
Samsung has the inferior process node and manufacturing quality. We found this with the iPhone...
I've no inkling of the accuracy of this, but they have the resources to get into the game big time. Good time for it too, seing as GloFo pulled out of the shrinking race, and Intel's floundring. That window is probably closing though, so good deal for them.
Posted on Reply
#9
kastriot
90% percent of you have samsung memory(including me!) in PC and you think samsung is not capable to make chips for AMD, ha!
Posted on Reply
#10
cucker tarlson
kastriot, post: 3944120, member: 165334"
90% percent of you have samsung memory(including me!) in PC and you think samsung is not capable to make chips for AMD, ha!
I have micron's ram + micron's vram :D
...and four samsung v-nand ssds
Posted on Reply
#11
prtskg
This also means GF doesn't have enough 12nm capacity for AMD even now, isn't it?
Posted on Reply
#12
Frick
Fishfaced Nincompoop
kastriot, post: 3944120, member: 165334"
90% percent of you have samsung memory(including me!) in PC and you think samsung is not capable to make chips for AMD, ha!
Is DRAM and NAND chips as complicated to make as modern cutting edge microprocessors? Their design is simpler so I assume they should be more simple to manufacture?
Posted on Reply
#13
FordGT90Concept
"I go fast!1!11!1!"
prtskg, post: 3944133, member: 163162"
This also means GF doesn't have enough 12nm capacity for AMD even now, isn't it?
Fab 1 = 40, 28, 22
Fab 2 = 600-350
Fab 3/5 = 350-180
Fab 3E = 180
Fab 6 = 180-110
Fab 7 = 130-40
Fab 8 = 28, 14
Fab 9 = 90
Fab 10 = 14

Presumably the 14nm facilities are doing 12nm so...two of them? Yeah, their production capability is very limited and Polaris 30 is a volume product.
Posted on Reply
#14
Totally
FordGT90Concept, post: 3944160, member: 60463"
Fab 1 = 40, 28, 22
Fab 2 = 600-350
Fab 3/5 = 350-180
Fab 3E = 180
Fab 6 = 180-110
Fab 7 = 130-40
Fab 8 = 28, 14
Fab 9 = 90
Fab 10 = 14

Presumably the 14nm facilities are doing 12nm so...two of them? Yeah, their production capability is very limited and Polaris 30 is a volume product.
Now it makes sense why they're looking into divorcing the IMC from the cpu die.
Posted on Reply
#15
rvalencia
FordGT90Concept, post: 3944160, member: 60463"
Fab 1 = 40, 28, 22
Fab 2 = 600-350
Fab 3/5 = 350-180
Fab 3E = 180
Fab 6 = 180-110
Fab 7 = 130-40
Fab 8 = 28, 14
Fab 9 = 90
Fab 10 = 14

Presumably the 14nm facilities are doing 12nm so...two of them? Yeah, their production capability is very limited and Polaris 30 is a volume product.
Fab 1 also has 12 nm FDSOI.
Fab 9 has 7 nm development node.
Posted on Reply
#16
Kaotik
The same die size isn't because it's based on 14nm too, it's because they're using the same old 9T library (or whatever T it was) as they used with 14nm, at least GloFo 12nm has smaller 7.5T libraries available too but AMD hasn't (so far) used it
Posted on Reply
#17
hat
Enthusiast
FordGT90Concept, post: 3944160, member: 60463"
Fab 1 = 40, 28, 22
Fab 2 = 600-350
Fab 3/5 = 350-180
Fab 3E = 180
Fab 6 = 180-110
Fab 7 = 130-40
Fab 8 = 28, 14
Fab 9 = 90
Fab 10 = 14

Presumably the 14nm facilities are doing 12nm so...two of them? Yeah, their production capability is very limited and Polaris 30 is a volume product.
600-350nm manufacturing? Who is still using this?
Posted on Reply
#18
jeremyshaw
hat, post: 3944223, member: 32804"
600-350nm manufacturing? Who is still using this?
My out of thin air (or somewhere else) guess is high voltage (>1000V), solid state equipment.
Posted on Reply
#19
raghu78
I am surprised that AMD is using GF 12LP and Samsung 11LPP. Since both support CPP=78nm 9T libraries and are derived from 14LPP this is possible. I am interested in seeing if AMD can do the same for Zen+ Pinnacle Ridge to better meet demand given the shortage of Intel CPUs in DIY market.
https://fuse.wikichip.org/news/1425/vlsi-2018-samsungs-11nm-11lpp/
https://fuse.wikichip.org/news/1497/vlsi-2018-globalfoundries-12nm-leading-performance-12lp/
Since Polaris 30 is using the same die as Polaris 10/20 with only the newer transistors there is no redesign, just like Zen+ had no redesign as they used only the 12nm transistors but not the cell libraries. Their foundry strategy is very flexible with AMD using GF, Samsung and TSMC. Good job AMD.
Posted on Reply
#20
Totally
hat, post: 3944223, member: 32804"
600-350nm manufacturing? Who is still using this?
jeremyshaw, post: 3944229, member: 92377"
My out of thin air (or somewhere else) guess is high voltage (>1000V), solid state equipment.
"Fab 2, located in Woodlands, Singapore. This fab is capable of manufacturing wafers at 600 to 350 nm for use in selected automotive IC products, High Voltage power management IC and Mixed-signal products."
Posted on Reply
#21
btarunr
Editor & Senior Moderator
raghu78, post: 3944235, member: 106613"
....just like Zen+ had no redesign as they used only the 12nm transistors but not the cell libraries.
I wonder how AMD achieved faster on-die caches on PiR without etching a different SRAM.
Posted on Reply
#22
ET3D
What I'm curious about is whether all RX 580/570 production will be switched to Polaris 30. Is there a reason to keep producing the Polaris 10?
Posted on Reply
#23
FordGT90Concept
"I go fast!1!11!1!"
Considering they're being made by Samsung and Global Foundries, there's probably not enough silicon available to upgrade the entire product stack.
Posted on Reply
#24
ET3D
FordGT90Concept, post: 3944289, member: 60463"
Considering they're being made by Samsung and Global Foundries, there's probably not enough silicon available to upgrade the entire product stack.
Both Polaris 30 and Polaris 10 use the same production plants, right? Won't they come at the expense of one another anyway? I assume that AMD has quite a big stock of Polaris 10 to get rid of, but once that's gone, why would more Polaris 10 chips be made? (Of course, it's possible that AMD has so many Polaris 10 chips left that's it's an academic question.)
Posted on Reply
#25
FordGT90Concept
"I go fast!1!11!1!"
They might just sell the inferior chips as RX 580/RX 570. Perhaps as RX 585 and RX 575. It really depends on what kind of yields they're getting.
Posted on Reply
Add your own comment