Low yield on Samsung's 4nm process node prompts Qualcomm to go with TSMC for future chips

[nanoguy]

What just happened? Samsung has committed to spending over $100 billion to improve its chip manufacturing arm, but the company is experiencing serious yield issues on its advanced process nodes that may drive some of its customers away. For instance, Qualcomm will be making its Snapdragon 8 Gen2 chipset on TSMC's 3nm process node.

Traditionally, Qualcomm has partnered with either TSMC or Samsung Foundry for its chip manufacturing needs. However, it turns out the latter is experiencing yield issues on its cutting-edge production lines that are preventing it from meeting expected order volumes.

According to a report by Korean publication The Elec, Qualcomm has decided to switch back to TSMC for the Snapdragon 8 Gen2 Mobile Platform. The chipset will be manufactured on a 3nm process node and is expected to power next year’s flagship phones.

Samsung’s 4nm process node is plagued by a yield rate of as low as 35 percent. In other words, for every 100 Snapdragon 8 Gen1 chips etched on a wafer, 65 don’t meet the quality standards imposed by Qualcomm. For reference, TSMC’s 4nm process node — which is used by Qualcomm to manufacture the Snapdragon 8 Gen1 Plus — is said to have a yield rate of over 70 percent.

The Elec said Samsung is also having issues making its own Exynos 2200 chipset, with yields even lower than the ones achieved for the Snapdragon 8 Gen1. This prompted the Korean tech giant to conduct an internal investigation into the matter, according to local press. It’s possible this may be a case of fraudulent reporting on the part of Samsung Foundries executives, so the company is looking into current and former managers of its 5nm, 4nm, and 3nm production lines.

Qualcomm isn’t the only company experiencing issues with Samsung’s advanced process nodes. Nvidia is said to have spent no less than $10 billion to secure a chunk of TSMC’s 5nm manufacturing capacity for its upcoming RTX 4000 series GPUs.

Permalink to story.

https://www.techspot.com/news/93520-low-yield-samsung-4nm-process-node-prompts-qualcomm.html

 

[Nintenboy01]

TSMC must really have the best engineers in the biz or signed a deal with the devil to advance so much.

 

[Dimitriid]

TSMC must really have the best engineers in the biz or signed a deal with the devil to advance so much.

Could be, they certainly get a lot of money to put into R&D from several really big customers.

But I remember many years ago people used to think the same of intel fabs and just how far ahead they were and then 10nm happened to them, or rather didn't happen to them for an extremely long time.

All I'm saying is that as we race closer and closer to the limits of what's possible on a particle level, there's more and more chances that *any* of the forges runs into massive problems like intel did with 10nm but this time could be Samsung or could be TSMC.

So I'd be interesting to see what happens to the world if TSMC starts getting trouble on future nodes: does Apple attempts to move back to intel? What about AMD that owns much of their current success and unprecedented market cap to TSMC and the ability their nodes had to be untouched by intel for many of the critical Ryzen launch years what do they do if intel is back as the top dog and they're stuck on 3nm for 5 years because now TSMC can't get anything smaller to work properly?

 

[hahahanoobs]

Samsung?!
If what I read was right, TSMC 3nm is in trouble too. And TSMC has supply issues already. Here's a sneak peek;apparently Intel shipped 7.3M server CPU's to EPYC's 1.1M. No 5800X3D release date. No word on Zen 3 support on 300 series boards while getting trounced by ADL sales. The list literally goes on. Not looking good guys. Nvidia may go with TSMC 4nm instead of 3nm. It's tough out there.

Things are about to get interesting.

 

[kiwigraeme]

The other day when someone blithely suggested setting up a foundry- I was a bit bemused .
Assuming you can even get the parts from that Dutch etching (UV lasers ) company - and then book their engineers for 6 months to help set up - or their competitors .
This is much harder than brewing beer.
Let's invest in football in the USA - throw $5 Billion at it at win the World Cup - not happening .
There is probably 1001 processes that have to go right .
Part of me would love to be a Project Engineer bringing Billion dollar projects on time and under budget - ( but don't have a family ) - there is a reason those people get pay mega bucks.
This is no Elon - go fast break things - this is paint prep x 1 million .
Even if you look at surgery team success rates - some ops are 12 to 20 people working really well together - it's not just the superstar surgeon that can illegally park his Lambo in front of the hospital doors.

Saying that - maybe in 100 years we can buy and over the counter that can print 4nm circuits at a slow rate - or nearly anything else at that scale given the correct ingredients - Fluorine excepted- that stuff is brutal

 

[Mr Majestyk]

All this means is higher prices from TSMC. Samsung's incompetence is really hurting the consumer with hardly any viable alternatives. Putting all our eggs in the TSMC basket is dangerous, and even more so now. Now China has seen the pathetic West stand by while Putin invades Ukraine, China will be emboldened to have a go at Taiwan. Sure the West will probably not just sit by due to the importance of Taiwan to the global manufacturing chain, but the outcome could be catastrophic.

 

[Theinsanegamer]

Could be, they certainly get a lot of money to put into R&D from several really big customers.

But I remember many years ago people used to think the same of intel fabs and just how far ahead they were and then 10nm happened to them, or rather didn't happen to them for an extremely long time.

All I'm saying is that as we race closer and closer to the limits of what's possible on a particle level, there's more and more chances that *any* of the forges runs into massive problems like intel did with 10nm but this time could be Samsung or could be TSMC.

So I'd be interesting to see what happens to the world if TSMC starts getting trouble on future nodes: does Apple attempts to move back to intel? What about AMD that owns much of their current success and unprecedented market cap to TSMC and the ability their nodes had to be untouched by intel for many of the critical Ryzen launch years what do they do if intel is back as the top dog and they're stuck on 3nm for 5 years because now TSMC can't get anything smaller to work properly?

TSMC almost had an intel moment. Anyone remember their aborted 20nm node? Imagine if they had gotten hung up on spending years trying to fix it.

Intel should have backed off from the more exotic parts of 10nm superfin, they could have gotten it out most likely only a year or two behind. The node only does so much as well, their core line needs a major overhaul, which alderlake is the first full step towards.

Any of these big companies can have major issues down the line. We are talking properly anal levels of engineering to get any more gains at this scale. What happened with intel with 10nm may be the future of the entire sector, not just them. TSMC is having a lot of luck, but remember that nm numbers dont tell the whole story TSMC 10nm is not directly comparable to intel 7nm.

It's been suspected that samsung has had yield issues for awhile now, this is pretty much confirming it. Samsung's upper management and its corruption may be catching up with them.

 

[zamroni111]

Well, nvidia can produce more ampere if qualcomm moves to tsmc

 

[zamroni111]

Could be, they certainly get a lot of money to put into R&D from several really big customers.

But I remember many years ago people used to think the same of intel fabs and just how far ahead they were and then 10nm happened to them, or rather didn't happen to them for an extremely long time.

All I'm saying is that as we race closer and closer to the limits of what's possible on a particle level, there's more and more chances that *any* of the forges runs into massive problems like intel did with 10nm but this time could be Samsung or could be TSMC.

So I'd be interesting to see what happens to the world if TSMC starts getting trouble on future nodes: does Apple attempts to move back to intel? What about AMD that owns much of their current success and unprecedented market cap to TSMC and the ability their nodes had to be untouched by intel for many of the critical Ryzen launch years what do they do if intel is back as the top dog and they're stuck on 3nm for 5 years because now TSMC can't get anything smaller to work properly?

Indeed. silicon atom size is only .2 nm

 

[Watzupken]

All this means is higher prices from TSMC. Samsung's incompetence is really hurting the consumer with hardly any viable alternatives. Putting all our eggs in the TSMC basket is dangerous, and even more so now. Now China has seen the pathetic West stand by while Putin invades Ukraine, China will be emboldened to have a go at Taiwan. Sure the West will probably not just sit by due to the importance of Taiwan to the global manufacturing chain, but the outcome could be catastrophic.

Samsung's failure could allow Intel to step up as the next go to fab. I am no expert in this field, but from my observations over at least the past half a decade, Samsung's fab is consistently behind TSMC in terms of yield and power characteristics. I don't know what is TSMC doing right, or Samsung doing wrong, but if Samsung's node is always like 1 generation behind TSMC despite the same nanometer naming, than Samsung really needs to up the game or do something drastic to catch up.

 

[Watzupken]

Well, nvidia can produce more ampere if qualcomm moves to tsmc

I feel Nvidia cannot afford to cheap out on fab now that they have to contend with AMD and Intel. it is not just about yield in my opinion, but I feel the node is less efficient than a TSMC equivalent. So at the given clockspeed, Samsung's node will require more power to get there. I think we will have a clearer picture when Qualcomm releases their TSMC based 8 gen 1 chip. That way, we can see between the TSMC and Samsung chips, how high can they boost clockspeed and how much throttling.

 

[zamroni111]

I feel Nvidia cannot afford to cheap out on fab now that they have to contend with AMD and Intel. it is not just about yield in my opinion, but I feel the node is less efficient than a TSMC equivalent. So at the given clockspeed, Samsung's node will require more power to get there. I think we will have a clearer picture when Qualcomm releases their TSMC based 8 gen 1 chip. That way, we can see between the TSMC and Samsung chips, how high can they boost clockspeed and how much throttling.

Nvidia can't simply move 3090 and lower to tsmc.

Every chip model needs 40+ lithography glass masks that are designed to specific manufacturing process.
Signal buffers etc. in the chip are also designed to signal latency characteristics of Samsung 8nm.
So, moving 3090 or lower to tsmc will not be economical for nvidia because they have to redesign many things.