Not bad.

Here's what I don't get: If they already know 1.4nm is possible, what exactly is preventing them from just skipping right to that? It's one thing where you're limited to how small you can get because you need an entirely new process, but it seems to me (and maybe I'm wrong) that for a lot of semiconductor producers, they're mostly just fine-tuning their process. I understand this is no simple task but I imagine the R&D is less expensive and time consuming to push the equipment to their physical limits rather than go in these incremental stages. On the other hand, if they don't show progress with their nodes then not only do they lose interest in their manufacturing process, but it makes it harder to improve the chips produced by them. In other words, companies like Intel, AMD, ARM, and Nvidia will have a lot harder time selling products that are barely any better than last-gen. These incremental node improvements probably help make next-gen products just a little more enticing. I know it's not exactly unheard of for industries to deliberately not release the best product they can, so next-gen products can be more enticing.

schmidtbag:

Here's what I don't get: If they already know 1.4nm is possible, what exactly is preventing them from just skipping right to that? It's one thing where you're limited to how small you can get because you need an entirely new process, but it seems to me (and maybe I'm wrong) that for a lot of semiconductor producers, they're mostly just fine-tuning their process. I understand this is no simple task but I imagine the R&D is less expensive and time consuming to push the equipment to their physical limits rather than go in these incremental stages. On the other hand, if they don't show progress with their nodes then not only do they lose interest in their manufacturing process, but it makes it harder to improve the chips produced by them. In other words, companies like Intel, AMD, ARM, and Nvidia will have a lot harder time selling products that are barely any better than last-gen. These incremental node improvements probably help make next-gen products just a little more enticing. I know it's not exactly unheard of for industries to deliberately not release the best product they can, so next-gen products can be more enticing.

My understanding is that there is a large amount of time required for the process to be refined until mass production (at a reasonable yield) is possible.

schmidtbag:

Here's what I don't get: If they already know 1.4nm is possible, what exactly is preventing them from just skipping right to that? It's one thing where you're limited to how small you can get because you need an entirely new process, but it seems to me (and maybe I'm wrong) that for a lot of semiconductor producers, they're mostly just fine-tuning their process. I understand this is no simple task but I imagine the R&D is less expensive and time consuming to push the equipment to their physical limits rather than go in these incremental stages. On the other hand, if they don't show progress with their nodes then not only do they lose interest in their manufacturing process, but it makes it harder to improve the chips produced by them. In other words, companies like Intel, AMD, ARM, and Nvidia will have a lot harder time selling products that are barely any better than last-gen. These incremental node improvements probably help make next-gen products just a little more enticing. I know it's not exactly unheard of for industries to deliberately not release the best product they can, so next-gen products can be more enticing.

New process nodes introduce new technology that take time to develop and costs to implement. Also consider that the bigger the jump, the greater the risk. Intel 10nm is an example of this. They started with some rather ambitious design goals, but had to cut back a bit to make it work. This cost billions for Intel and delayed it's launch several years.

schmidtbag:

Here's what I don't get: If they already know 1.4nm is possible, what exactly is preventing them from just skipping right to that?

There is a big difference going from theory to practice, or should I say, a single wafer test to a full fab working out thousands of wafers. It´s not like they turn a few screws, adjust some software and boom: new improved process. When they make a multi billion dollar fab, they expect it to work for a certain time (10 or more years) non stop on that process to recoup the investment and make money. To start using a new process it might take new hardware, new chemicals, new processes, and that means renewing a still useful factory that now has to be stopped, or build another one. That's why generally you see leading fabs investing in more plants and more area with new processes. New stuff is made at new fabs and old fabs start making chipsets and less important components for washing machines or toasters. PS: making a new plant can take 3 to 5 years, generally. So we're on track for 2027!

Catspaw:

My understanding is that there is a large amount of time required for the process to be refined until mass production (at a reasonable yield) is possible.

Understood, but that's why I said it might make more sense to spend a little more time and money just skipping straight to the smallest size the technology can theoretically handle.

Horus-Anhur:

New process nodes introduce new technology that take time to develop and costs to implement. Also consider that the bigger the jump, the greater the risk. Intel 10nm is an example of this. They started with some rather ambitious design goals, but had to cut back a bit to make it work. This cost billions for Intel and delayed it's launch several years.

Like I said, if a node shrink requires all new tech, I totally get the incremental change. Even if the first generation isn't as finely tuned as it could be, that also makes sense, because as long as it's an improvement, they need to get some ROI. But when they're fine-tuning their existing technology to a level they already knew was possible, that's where it starts to not make so much sense. I'm not saying that's what Samsung is doing here, but it just seems like in the past, that's how things worked.

schmidtbag:

Understood, but that's why I said it might make more sense to spend a little more time and money just skipping straight to the smallest size the technology can theoretically handle. Like I said, if a node shrink requires all new tech, I totally get the incremental change. Even if the first generation isn't as finely tuned as it could be, that also makes sense, because as long as it's an improvement, they need to get some ROI. But when they're fine-tuning their existing technology to a level they already knew was possible, that's where it starts to not make so much sense. I'm not saying that's what Samsung is doing here, but it just seems like in the past, that's how things worked.

Refinement is done in a half node. Such as N6 is a half node of N7. Advancements are done on full nodes. Like N7.

schmidtbag:

I know it's not exactly unheard of for industries to deliberately not release the best product they can, so next-gen products can be more enticing.

I think your comment above is a big part of it. Would you go straight to technology Z if you could make trillions milking technology A, B, C...? The next big leap in computing power is probably already road mapped out, but it won't happen quickly unless hands are forced. The current path is clear, less risky, and highly profitable. It's obviously not as simple as I'm stating here and there are many more variables in play, but I think generational milking is definitely factored into their profitability strategy.

How hard you push your tech boundaries depends on what your competitors can do.

schmidtbag:

Here's what I don't get: If they already know 1.4nm is possible, what exactly is preventing them from just skipping right to that? It's one thing where you're limited to how small you can get because you need an entirely new process, but it seems to me (and maybe I'm wrong) that for a lot of semiconductor producers, they're mostly just fine-tuning their process. I understand this is no simple task but I imagine the R&D is less expensive and time consuming to push the equipment to their physical limits rather than go in these incremental stages. On the other hand, if they don't show progress with their nodes then not only do they lose interest in their manufacturing process, but it makes it harder to improve the chips produced by them. In other words, companies like Intel, AMD, ARM, and Nvidia will have a lot harder time selling products that are barely any better than last-gen. These incremental node improvements probably help make next-gen products just a little more enticing. I know it's not exactly unheard of for industries to deliberately not release the best product they can, so next-gen products can be more enticing.

it would cost you $1500 to buy the cheapest card manufactured on a node that's not even in mass production.

schmidtbag:

Understood, but that's why I said it might make more sense to spend a little more time and money just skipping straight to the smallest size the technology can theoretically handle.

You're describing intels 10nm. In like 2015 they tried to moonshot down (lol) to a significantly smaller process but ended up in the weeds until ASML effectively bailed them out years later. It's easy to demonstrate this stuff in a lab. It's much harder to pilot and in some cases it's impossible to mass produce. If you don't go through the steps you might find your self several billion in a hole with little to show for it.

Again people fall for the trap of the marketing naming .... 1.4 is not really nm not as they where traditionally used . The tsmc 16/14 glofo 14 are the old 20nm but finfet 20nm traditional existed but the yields where so poor it was more expensive and it really did not offer much for the cost this is why we where stuck to 28nm for so long . The 16/14 nodes where renamed for marketing reason the tsmc 7nm are more in line with Intel's 10 nm density wise and the 5nm with Intel 7 .... Although not all waffers are the same between Intel Samsung Intel and tsmc and the density etc is not 1:1 identical. Honestly the most fair naming would be nand gates count per 1mm²

schmidtbag:

Here's what I don't get: If they already know 1.4nm is possible, what exactly is preventing them from just skipping right to that? It's one thing where you're limited to how small you can get because you need an entirely new process, but it seems to me (and maybe I'm wrong) that for a lot of semiconductor producers, they're mostly just fine-tuning their process. I understand this is no simple task but I imagine the R&D is less expensive and time consuming to push the equipment to their physical limits rather than go in these incremental stages.

Yield rates most likely. Even if you can do 1.4nm, it doesn't say it's commercially feasible at this time.