AMD to Tap Samsung's 4 nm Process for Chromebook Processors, Notes the Report from J.P. Morgan

[AleksandarK]

Historically, AMD was working with two semiconductor manufacturing companies: TSMC and GlobalFoundries. According to the latest report coming from Gokul Hariharan, an analyst at J.P. Morgan, AMD could soon tap another semiconductor manufacturer to produce the company's growing list of processors. As the report indicates, AMD could start working with the South Korean giant Samsung and utilize the firm's 4LPP process that represents a second generation of the low-power 4 nm silicon node. This specific node is allegedly the choice for AMD APUs designed to fit inside Google's Chromebook devices, which require low-power designs to achieve excellent battery life.

AMD could realize this move in late 2022, as Samsung's 4LPP node goes into mass production at that point. It means that we could see the first Samsung-made AMD APUs in late 2022 or the beginning of 2023. And apparently, the two company's collaboration could be much more significant as AMD is evaluating Samsung's 3 nm nodes for other products spanning more segments in 2023/2024. There are no official, definitive agreements between the two, so we have to wait for more information and official responses from these parties. Anyways, if AMD decides to produce a part of its lineup at Samsung, the remaining TSMC capacity would ensure that the supply of every incoming chip remains sufficient.


View at TechPowerUp Main Site

 

[Vya Domus]

As long as they keep Samsung's nodes away from high performance products, I don't care.

 

[awesomesauce]

As long as they keep Samsung's nodes away from high performance products, I don't care.

why?

 

[z1n0x]

TSMC is overcrowded, AMD is growing. Moving low to middle-end products to Samsung and keeping TSMC for the high-end products, seems like a solution.

 

[Vya Domus]

why?

Their 8nm node was inferior to TSMC's 7nm, even their nodes intended for low power mobile chips have always been inferior.

 

[RedBear]

Their 8nm node was inferior to TSMC's 7nm, even their nodes intended for low power mobile chips have always been inferior.

Playing devil's advocate, isn't it unfair to assume that they will always stay inferior? After all Samsung is adopting GAAFET for 3nm, while TSMC is sticking with FinFET, it's possible that they will manage to close the gap and maybe even overtake TSMC. At any rate AMD might just need the additional capacity at this point in time and be willing to accept the possibly reduced performace, like Nvidia did for Ampere.

 

[prtskg]

Their 8nm node was inferior to TSMC's 7nm, even their nodes intended for low power mobile chips have always been inferior.

TSMC is at top. So everyone is kind of inferior to it. It's definitely better than GF they're currently using as second foundry though.

 

[ShurikN]

TSMC is at top. So everyone is kind of inferior to it. It's definitely better than GF they're currently using as second foundry though.

Are they actually making anything demanding/complex at GF? IO dies aren't really that dependent on manufacturing process, and benefit next to nothing from a die shrink (aside from a smaller footprint)

 

[Patriot]

Playing devil's advocate, isn't it unfair to assume that they will always stay inferior? After all Samsung is adopting GAAFET for 3nm, while TSMC is sticking with FinFET, it's possible that they will manage to close the gap and maybe even overtake TSMC. At any rate AMD might just need the additional capacity at this point in time and be willing to accept the possibly reduced performace, like Nvidia did for Ampere.

That seems perfectly reasonable, its also possible Samsung yields will be poorer given the new tech, as with anything, time will tell.
Gate all around definitely has the potential to be more efficient.

 

[Fluffmeister]

I just miss the good old days when real men had fabs.

 

[lexluthermiester]

As the report indicates, AMD could start working with the South Korean giant Samsung and utilize the firm's 4LPP process that represents a second generation of the low-power 4 nm silicon node.

4nm for Chromebooks? Seems like a waste of effort and resources to me. 4nm nodes would be better served to ease other shortages.

 

[RandallFlagg]

So did a bit of research since 'nm' means nothing.

Samsung's 5nm Ultra High Density node was 127MT/mm2, but the different 5nm variants range from about 95-127MT/mm2.

For comparison TSMC 5nm is ~170MT/mm2 and their 7nm was around 100MT/mm2.

Intel 10nm is around 101 MT/mm2.

Intel 14nm is ~37.5MT/mm2. Comparable nodes were GloFlo 12nm / 14nm at 36.7MT/mm2 and TSMC 12/14 at 28.9MT/mm2 and Samsung 14nm at 33 MT/mm2.

According to this article, calculated density of Samsung 4LPP is 137MT/mm2.

This puts it slightly more dense than TSMC 7nm and Intel 10nm, and limited to low power applications (high power application of a node is usually less dense).

 

[Wirko]

I just miss the good old days when real men had fabs.

Lisa doesn't care if she's a real man or not.

 

[95Viper]

Stay on Topic.
Discuss the topic... NOT, each other.
Stop the insulting remarks.

Thank You and Have a Nice Day.

 

[TheoneandonlyMrK]

I'm only surprised that it's for Chromebooks, I expected more.

And also , even if Samsung didn't quite have node parity in the high end, I realize apple upped the game but since when is high performance even in the same playing field as the phrase Chromebooks!.

 

[mechtech]

Said it before and I will say it again. Smaller nm is great, but at the rate we're going, I'd rather have a radeon 6600 on GF 12nm for MSRP on the shelf and pay for extra power use, than a non-existent TSMC 7nm at 2-3x the price and no where to be found.

 

[Minus Infinity]

It was only reported a few days ago Samsung's 4nm node was showing very poor yields. I really hope they can sort it out for not just AMD's sake. We need as many players as possible, not just TSMC.

 

[lexluthermiester]

It was only reported a few days ago Samsung's 4nm node was showing very poor yields.

As always, the refinement of the process will improve.

 

[IceShroom]

Their 8nm node was inferior to TSMC's 7nm, even their nodes intended for low power mobile chips have always been inferior.

Samsung's 8nm is quite good. It that Nvidia's Ampere is not efficient as AMD's RDNA2. MI50 to MI100, same 7nm and same TBP but 2x the size and performance. Nvidia's Volta V100 to Ampere A100, TSMC 12nm to TSMC 7nm, slightly bigger size and 2x performance but nearly 40-50% more TBP. From this we can conclude that Ampere is not efficient but Samsung's 8nm used for Ampere is close to TSMC's 7nm, but not better.

Said it before and I will say it again. Smaller nm is great, but at the rate we're going, I'd rather have a radeon 6600 on GF 12nm for MSRP on the shelf and pay for extra power use, than a non-existent TSMC 7nm at 2-3x the price and no where to be found.

Sadly no will buy a 12nm 6600 beside you or some of you. But the market does not work like that. If there were market for it then AMD would have already made one. And people care about power consumption, especially for AMD's CPU/GPU. A 180W-200W 6600 XT would be a bad choice against same TDP Nvidia card. So no dedicated 12nm RDNA2 gpu.

 

[mechtech]

Samsung's 8nm is quite good. It that Nvidia's Ampere is not efficient as AMD's RDNA2. MI50 to MI100, same 7nm and same TBP but 2x the size and performance. Nvidia's Volta V100 to Ampere A100, TSMC 12nm to TSMC 7nm, slightly bigger size and 2x performance but nearly 40-50% more TBP. From this we can conclude that Ampere is not efficient but Samsung's 8nm used for Ampere is close to TSMC's 7nm, but not better.


Sadly no will buy a 12nm 6600 beside you or some of you. But the market does not work like that. If there were market for it then AMD would have already made one. And people care about power consumption, especially for AMD's CPU/GPU. A 180W-200W 6600 XT would be a bad choice against same TDP Nvidia card. So no dedicated 12nm RDNA2 gpu.

If it was available at quantity on the shelves for the right price it would sell. Old saying, is there are no bad products only bad prices. A lot of people sitting on RX 460/470/570/480/580s, gtx 1060s, etc etc. would probably be happy to get 1.6x-2x performance gain over those cards for the same price as those cards. As for choice, that's the challenge, there is none, especially at the price points the above cards were released at, that is unless you want a GT1030 2GB card lol. But unfortunately like you said it won't happen, probably for any card. I am kind of surprised they don't start bringing back older cards like Nvidia did with the 2060, I mean with the demands and shortages they would probably have no trouble selling anything.

Here is basically a selection of video cards I could get, and for reference my RX 480 was $300 CAD when I bought it.

Newegg,In Stock,Shipped by Newegg GPUs / Video Graphics Cards | Newegg.ca

Shop Newegg,In Stock,Shipped by Newegg GPUs / Video Graphics Cards on Newegg.ca. Watch for amazing deals and get great pricing.

www.newegg.ca

I'm not a big gamer so not a huge deal for me, I will wait until prices return to pre-covid, and if they don't, I will use my card until it dies before replacing it.

 

[IceShroom]

If it was available at quantity on the shelves for the right price it would sell. Old saying, is there are no bad products only bad prices. A lot of people sitting on RX 460/470/570/480/580s, gtx 1060s, etc etc. would probably be happy to get 1.6x-2x performance gain over those cards for the same price as those cards. As for choice, that's the challenge, there is none, especially at the price points the above cards were released at, that is unless you want a GT1030 2GB card lol. But unfortunately like you said it won't happen, probably for any card. I am kind of surprised they don't start bringing back older cards like Nvidia did with the 2060, I mean with the demands and shortages they would probably have no trouble selling anything.

Here is basically a selection of video cards I could get, and for reference my RX 480 was $300 CAD when I bought it.

Newegg,In Stock,Shipped by Newegg GPUs / Video Graphics Cards | Newegg.ca

Shop Newegg,In Stock,Shipped by Newegg GPUs / Video Graphics Cards on Newegg.ca. Watch for amazing deals and get great pricing.

www.newegg.ca

I'm not a big gamer so not a huge deal for me, I will wait until prices return to pre-covid, and if they don't, I will use my card until it dies before replacing it.

Sadly AMD cant produce their old Polaris card, as they are only support D3D12 not D3D12 Ultimate and GDDR5 is out of production now days. And there is not cheap and small Vega based card. And RDNA and RDNA2 both uses same process node memory type, so producing RDNA2 is better choice. AMD can mitigate this in future by splitting production of gpu in different foundary and memory types.

Nvidia is better position with Turing in this regard. Not only it uses different foundary and node but it also supports newer D3D12 features.