4 ms·

The reason why this comes up is Intel historically as a company built it's market dominance on it's manufacturing processes. This in turn lead to profitability.

The reason why this comes up is Intel historically as a company built it's market dominance on it's manufacturing processes. This in turn lead to profitability. I'll use Intel's advertised sizes for processes below, but keep in mind that a smaller number just represents a new manufacturing process. Not that actual size. Intel switched to a FinFET process with its "22nm node" in 2011, then a few years later in 2014 switched to a "14nm node" that is also FinFET. That was effectively the last profitable manufacturing innovation for Intel. A "10nm node" certainly exists, but is practically a black hole of money when you consider the time invested on it and just how far behind they are at this point. The only concrete thing Intel did to address this was to just rename all their process nodes to be smaller. That's basically it. A few years from now is a full decade that Intel will have spent without actually introducing a new profitable manufacturing process. You can make all the press releases you want about how it'll be profitable "one day", it doesn't change the fact that Intel can't really seem to design new manufacturing processes. Intel effectively watched the manufacturing technology move past them and shrugged. The proposed solution to this is to spend several more years doing basically the same thing, but this time they'll hope they get it right. This is magical thinking.

guelo 4y ago
I wouldn't say they shrugged, they tried but this stuff is hard. It was an engineering failure. Though engineering failures are often caused by organizational failures.
- sidewndr46 4y ago
  Not just engineering. I listened to an earning's call once. The CFO or someone in that area was claiming that "14nm was the most profitable process ever!". My thought was "well yes, it's also the most long lived process. I'd hope you haven't been operating at a loss for all these years"
phkahler 4y ago
And they renamed 10nm to Intel 7, but it is not EUV. Both TSMC and Samsung have moved to EUV, and until Intel does as well they are going to remain behind. All that money spent is nice, but until they start taking delivery of tons of equipment from ASML it's not going to help them at the leading edge.
- selectodude 4y ago
  EUV is a tool, not a process. TSMC N5 isn't 100 percent EUV either. Intel can hit similar feature sizes with Intel 7 as TSMC can but it requires more steps, hitting their margins and lowering their yields. The chips that come out of the other end are equivalent though.
  - avs733 4y ago
    > EUV is a tool, not a process. It's kinda both. To be clear, EUV directly refers to a specific type of lithography equipment, ones using extreme UV light sources. However, those changes come with a lot of changes in larger process flow that are all inter-related. The change in EUV means smaller lithography lines, and also means needing photo resist, etching, cleaning, etc. - many other process steps, that support that size change. Add in that now, with EUV lithography, you are operating lithography under vacuum and its a huge challenge. Suddenly you have to have vacuum ready materials. One of Intel's mid 2010 wins was getting immersion litho to work better than peers - similar thing just wet wafers rather than vacuum wafers. EUV is a tool/technology, one used only on critical layers where that precision matters (e.g., gate patterning). But the larger (entire) manufacturing process has to be designed to enable the use of that tool/technology.
- bcrl 4y ago
  Leading edge processes are not where the shortage exists -- it's trailing process nodes that are used heavily for support chips and automotive. Support chips like voltage regulators sure as hell don't need a 5nm or 7nm process, but they are critical components to any piece of electronics being able to ship. Older processes are ideal for these needs because they are very well characterized / understood, making them excellent for chips with an analog component. And older processes were cheap until demand exceeded supply. Edit: and Intel doesn't and won't play into the market for support chips. There's no way in hell that they can support Intel's required profit margins.
  - picture 4y ago
    To add on: Older processes also have lower static current consumption due to less leakage from bigger features. They are quantitatively better in many applications for low power or high precision.
Melatonic 4y ago
I don't think the average user or custom cares all that much about what node they are actually on. Yes, of course battery life and efficiency matters. But if Intel can make some decent incremental improvements I just don't see them completely going away. AMD and Intel have been trading the throne for years and every single time everyone says one of them is about to die. Yet here we are.
- zamadatix 4y ago
  Via still exists and they never had the throne so I don't think either AMD or Intel are likely to die any time soon. The more interesting bet is if either will still be valued at ~$200 billion in 5 years or if these recent performance challengers end up making a large portion of the existing revenue generating x86 products look like Via CPUs did before they manage to get stronger grips into other product markets like GPUs and networking.

guelo 4y ago

I wouldn't say they shrugged, they tried but this stuff is hard. It was an engineering failure. Though engineering failures are often caused by organizational failures.

sidewndr46 4y ago
Not just engineering. I listened to an earning's call once. The CFO or someone in that area was claiming that "14nm was the most profitable process ever!". My thought was "well yes, it's also the most long lived process. I'd hope you haven't been operating at a loss for all these years"

phkahler 4y ago

And they renamed 10nm to Intel 7, but it is not EUV. Both TSMC and Samsung have moved to EUV, and until Intel does as well they are going to remain behind. All that money spent is nice, but until they start taking delivery of tons of equipment from ASML it's not going to help them at the leading edge.

selectodude 4y ago
EUV is a tool, not a process. TSMC N5 isn't 100 percent EUV either. Intel can hit similar feature sizes with Intel 7 as TSMC can but it requires more steps, hitting their margins and lowering their yields. The chips that come out of the other end are equivalent though.
- avs733 4y ago
  > EUV is a tool, not a process. It's kinda both. To be clear, EUV directly refers to a specific type of lithography equipment, ones using extreme UV light sources. However, those changes come with a lot of changes in larger process flow that are all inter-related. The change in EUV means smaller lithography lines, and also means needing photo resist, etching, cleaning, etc. - many other process steps, that support that size change. Add in that now, with EUV lithography, you are operating lithography under vacuum and its a huge challenge. Suddenly you have to have vacuum ready materials. One of Intel's mid 2010 wins was getting immersion litho to work better than peers - similar thing just wet wafers rather than vacuum wafers. EUV is a tool/technology, one used only on critical layers where that precision matters (e.g., gate patterning). But the larger (entire) manufacturing process has to be designed to enable the use of that tool/technology.
bcrl 4y ago
Leading edge processes are not where the shortage exists -- it's trailing process nodes that are used heavily for support chips and automotive. Support chips like voltage regulators sure as hell don't need a 5nm or 7nm process, but they are critical components to any piece of electronics being able to ship. Older processes are ideal for these needs because they are very well characterized / understood, making them excellent for chips with an analog component. And older processes were cheap until demand exceeded supply. Edit: and Intel doesn't and won't play into the market for support chips. There's no way in hell that they can support Intel's required profit margins.
- picture 4y ago
  To add on: Older processes also have lower static current consumption due to less leakage from bigger features. They are quantitatively better in many applications for low power or high precision.

Melatonic 4y ago

I don't think the average user or custom cares all that much about what node they are actually on. Yes, of course battery life and efficiency matters. But if Intel can make some decent incremental improvements I just don't see them completely going away. AMD and Intel have been trading the throne for years and every single time everyone says one of them is about to die. Yet here we are.

zamadatix 4y ago
Via still exists and they never had the throne so I don't think either AMD or Intel are likely to die any time soon. The more interesting bet is if either will still be valued at ~$200 billion in 5 years or if these recent performance challengers end up making a large portion of the existing revenue generating x86 products look like Via CPUs did before they manage to get stronger grips into other product markets like GPUs and networking.