Fab manufacturing questions

[jms_embedded]

- Unbalancing strategically is both deliberate and necessary at my fab; to achieve targets which are customer imposed

Could you elaborate at all? (hopefully without going into proprietary aspects)

Is it a matter of meeting risk requirements? (e.g. something like "keep > 95% uptime of all machinery aside from lithography")

 

[Scotten Jones]

What do you consider unique to semiconductors? I don't have a copy (yet), but the table of contents looks fairly detailed and includes some aspects of reentrancy.

This is a longer conversation than I have time for but simply put....

The technology needs outweigh the manufacturing needs and the technology is harder than any other manufacturing processes on the planet. You have incredibly expensive tools that are immature and therefore are down a lot. Reentrancy is pretty much required due to the process flows and equipment cost. There is also a mix of batch and single wafer tools and huge differences in throughput between tools. I have seen comments from him about how all those issues should be eliminated but practically speaking that isn't possible.

 

[jms_embedded]

The technology needs outweigh the manufacturing needs and the technology is harder than any other manufacturing processes on the planet. You have incredibly expensive tools that are immature and therefore are down a lot. Reentrancy is pretty much required due to the process flows and equipment cost. There is also a mix of batch and single wafer tools and huge differences in throughput between tools. I have seen comments from him about how all those issues should be eliminated but practically speaking that isn't possible.

(my emphasis on tools that are down a lot)

Huh, that's something I hadn't thought about before... If I catch the idea correctly: perhaps what you mean to imply is that there's a bit of a bathtub-curve going on, where:

1. brand new tools have higher failure rates not only because they're a new design, and they may have infant-mortality failures, but also because the fab staff needs to learn how to get them working correctly on-site
2. after a learning curve, the failure rate is lower
3. after a LONG time, the failure rate increases again, because the equipment has exceeded its design life

Perhaps in other manufacturing industries it makes sense to spend the money to reduce failure rates, but in semiconductor manufacturing the capital equipment is just so expensive, and first-mover advantage is so high, that it makes more economic sense just to spend the labor costs to keep brand-new leading edge tools up and running properly.

And in the case of mature fabs, it also makes economic sense to spend the labor costs to keep old tools up and running properly rather than to purchase new equipment and incur depreciation costs.

I'm probably missing something, but as someone who doesn't work in manufacturing, that's what I read into it. Am I close?

 

[Scotten Jones]

No, that isn’t what I was saying.

 

[Scotten Jones]

Take EUV, even with uptime <90% it was adopted because it was essential for continued scaling. Getting above 90% and ultimately approaching 95% required new/upgraded versions.

 

[benb]

Could you elaborate at all? (hopefully without going into proprietary aspects)

Is it a matter of meeting risk requirements? (e.g. something like "keep > 95% uptime of all machinery aside from lithography")

Speaking very generically, a customer will say, we’ll pay a premium to get x chips by y date, so we‘ll prioritize backend product near completion and slow elsewhere in an attempt to meet it. The cost is having to refill that part of the line later.

Unbalancing is inefficient and revenue reducing, as increased output variation is a drag on revenue. The price premium has to justify it.

 

[benb]

JMS, one thing to emphasize, fabs operate on 25 wafer lots. The ideal would be single wafer lots. This would reduce the inventory from 1 million to 40,000. Fabs maximize revenue by processing 25 wafers at a time, building a large inventory, which buffers variations and ensures, within reason, smooth output and thus the most revenue. But inventory is a big cost too.

It’s all trade offs my friend.

 

[jms_embedded]

Speaking very generically, a customer will say, we’ll pay a premium to get x chips by y date, so we‘ll prioritize backend product near completion and slow elsewhere in an attempt to meet it. The cost is having to refill that part of the line later.

Unbalancing is inefficient and revenue reducing, as increased output variation is a drag on revenue. The price premium has to justify it.

That sounds more like hot lots, rather than a fab that has deliberately unbalanced tool capacity. (for example where there is an intentional bottleneck of 10,000 wafers per month at lithography and 11,000 - 15,000 WPM at other tool groups)

But you do raise the point that dispatching decisions have a major impact.

 

[Fred Chen]

(my emphasis on tools that are down a lot)

Huh, that's something I hadn't thought about before... If I catch the idea correctly: perhaps what you mean to imply is that there's a bit of a bathtub-curve going on, where:

1. brand new tools have higher failure rates not only because they're a new design, and they may have infant-mortality failures, but also because the fab staff needs to learn how to get them working correctly on-site
2. after a learning curve, the failure rate is lower
3. after a LONG time, the failure rate increases again, because the equipment has exceeded its design life

Perhaps in other manufacturing industries it makes sense to spend the money to reduce failure rates, but in semiconductor manufacturing the capital equipment is just so expensive, and first-mover advantage is so high, that it makes more economic sense just to spend the labor costs to keep brand-new leading edge tools up and running properly.

And in the case of mature fabs, it also makes economic sense to spend the labor costs to keep old tools up and running properly rather than to purchase new equipment and incur depreciation costs.

I'm probably missing something, but as someone who doesn't work in manufacturing, that's what I read into it. Am I close?

Long downtime means there is not enough understanding of the tool and its workings. Only when there is sufficient understanding can uptime be improved. Even a well-understood tool, though, can have EOL in some cases, e.g., parts no longer supported. So newer models to cover the old tools' applications would be needed.