11 ms·

The demand for leading-edge nodes comes from CPUs and GPUs. Once they move to the newest node, won’t previous capacity be freed to manufacture chips that don’t

The demand for leading-edge nodes comes from CPUs and GPUs. Once they move to the newest node, won’t previous capacity be freed to manufacture chips that don’t strictly need the better process but simply need manufacturing capacity? The machines are surely depreciated by then. Or is the cost of porting the design and making the masks prohibitive?

dragonelite 5y ago
It's usually not technical or capacity usually it's the whole ecosystem surrounding 14, 7, 5nm and lower ecosystems that makes it not really profitable. The initial costs grows exponentially. SOC, CPU and GPU might be one of the few consumer goods the economics work.
- MrBuddyCasino 5y ago
  > The initial costs grows exponentially. I understand that, but is this unavoidable? After all, the old designs don't need the precision of the advanced node, couldn't they (automatically?) translate their existing coarse designs and maybe produce masks more cheaply because a 40nm design doesn't need to be as perfect as a 7nm one? I'm sure this is nonsense otherwise they would have done that already, I'm just interested in why.
  - dragonelite 5y ago
    Im not a expert or even knowledgable but i was familiar with this image: https://www.extremetech.com/wp-content/uploads/2018/06/nano3.png https://www.extremetech.com/wp-content/uploads/2018/06/nano3... It's probably that the software and extras investment is not worth it right now.
blihp 5y ago
The more recent previous nodes are too new for companies like the autos as they want 20+ year old process nodes like 90nm+. So having excess 14nm capacity, for example, doesn't give them what they want since they can't just retrofit the fab for 130nm. However, not having made arrangements with fabs to keep the older nodes around contractually, they (customers like the autos) may not have an option other than paying to redesign their chips for newer process nodes in at least some cases at this point.

dragonelite 5y ago

It's usually not technical or capacity usually it's the whole ecosystem surrounding 14, 7, 5nm and lower ecosystems that makes it not really profitable. The initial costs grows exponentially. SOC, CPU and GPU might be one of the few consumer goods the economics work.

MrBuddyCasino 5y ago
> The initial costs grows exponentially. I understand that, but is this unavoidable? After all, the old designs don't need the precision of the advanced node, couldn't they (automatically?) translate their existing coarse designs and maybe produce masks more cheaply because a 40nm design doesn't need to be as perfect as a 7nm one? I'm sure this is nonsense otherwise they would have done that already, I'm just interested in why.
- dragonelite 5y ago
  Im not a expert or even knowledgable but i was familiar with this image: https://www.extremetech.com/wp-content/uploads/2018/06/nano3.png https://www.extremetech.com/wp-content/uploads/2018/06/nano3... It's probably that the software and extras investment is not worth it right now.

blihp 5y ago

The more recent previous nodes are too new for companies like the autos as they want 20+ year old process nodes like 90nm+. So having excess 14nm capacity, for example, doesn't give them what they want since they can't just retrofit the fab for 130nm. However, not having made arrangements with fabs to keep the older nodes around contractually, they (customers like the autos) may not have an option other than paying to redesign their chips for newer process nodes in at least some cases at this point.