Durations of process steps?

[Tanj]

Yeah, but if you really *really* need to get samples ASAP it's still "only"*** adds <20% to the 5nm mask costs that you're paying for anyway, so <5% to the total chip NRE cost

Ah, then some of the payment might have been accelerating the mask-making? Jumping the queue on a process taking half a day of machine time per mask, for multiple masks, seems like it would have been much more expensive than a FOUP or two jumping in front at some 30 minute machines.

 

[IanD]

Ah, then some of the payment might have been accelerating the mask-making? Jumping the queue on a process taking half a day of machine time per mask, for multiple masks, seems like it would have been much more expensive than a FOUP or two jumping in front at some 30 minute machines.

I don't know exactly where the time is saved; there are at least two lower (and cheaper) levels of acceleration (e.g. "hot-lot", "super-hot-lot") which are more widely available, I suspect these are speeding up the masks and leapfrogging in the line. The reason given for the high ultra-super-hot cost is specifically that it's "hand-carried" through the line and everything else has to get out of its way, this disruption leads to a net slowdown in the line and reduced throughput for the entire two months or so it takes to get through.

I did think $250k/day was ridiculously expensive, but apparently it was needed to meet the promise of getting devices into customer's hands before Xmas...

 

[nghanayem]

I don't know exactly where the time is saved; there are at least two lower (and cheaper) levels of acceleration (e.g. "hot-lot", "super-hot-lot") which are more widely available, I suspect these are speeding up the masks and leapfrogging in the line. The reason given for the high ultra-super-hot cost is specifically that it's "hand-carried" through the line and everything else has to get out of its way, this disruption leads to a net slowdown in the line and reduced throughput for the entire two months or so it takes to get through.

Hand carrying hot lots is bizarre to me and sounds inefficient. For us it is just a normal foup that gets to cut in queues for an OHV and tool load ports.

 

[IanD]

Hand carrying hot lots is bizarre to me and sounds inefficient. For us it is just a normal foup that gets to cut in queues for an OHV and tool load ports.

Come on, use your common sense -- no wafer boats are ever "hand carried" in a modern 300mm fab, all handling is fully automated -- which is why I put quotes around the term... ;-)

(so we should also say "tape out"...)

What it means is that somebody has to intervene in the automated process flow and scheduling to hold other lots out of the way (which is why line throughput drops) and either ensure the hot-lot never has to sit waiting for another one to clear the next machine or even let them be "overtaken" -- and this tracking has to be done through the entire process flow. If I understand correctly, a lot like this also has process operators specifically tasked with making sure they go through the line as fast as possible.

 

[jms_embedded]

What it means is that somebody has to intervene in the automated process flow and scheduling to hold other lots out of the way (which is why line throughput drops) and either ensure the hot-lot never has to sit waiting for another one to clear the next machine or even let them be "overtaken" -- and this tracking has to be done through the entire process flow. If I understand correctly, a lot like this also has process operators specifically tasked with making sure they go through the line as fast as possible.

OK, so it's like a manual override of factory scheduling.

 

[Jert]

Not sure, it was a standard lot so presumably 25 or 50 wafers. I don't think the fee depends on this, it's because it "queue-jumps" -- all the other products in the line literally stop to get out of the way of a lot like this -- so the productivity of the entire line drops. For the same reason there are only a very small number of such slots per line available each month...

Hot lots (and Hot Hot lots) are very common, practically every fab uses this approach to sometimes speed up certain wafers through the line.

Hot lots are disruptive to fab operations, of course. But the impact should not be so severe, unless the tools need to have a new setup to run the Hot lot (for example, a new PM and many hours of seasoning after the PM, or requiring to open the chamber to change out the parts, etc.)

A lot times Hot lots are engineering wafers (qualifying a new product, needing wafer-level data for resolving a critical process engineering issue, new product samples to be sent to new customers, etc.), so production team people bitterly dislike Hot lots because they steal their tools' production time but does not count as revenue, and risk them not meeting their daily wafer output quota (production goal), and so the production teams try to come up with ways to "charge" (more like punish) people and to discourage Hot lots. Thus the exorbitant "cost" often times are not real but punitive.

 

[Tanj]

the impact should not be so severe, unless the tools need to have a new setup to run the Hot lot

Changing a mask is obviously disruptive if it is not already in the swap-set within a machine.

Are there other tools notoriously likely to be disrupted by a hot lot? If the wafer is running a standard process, mostly there should be machines running the correct formulas for development, etching, deposition, CMP, etc. so the scheduling software should be able to route it to head of queue on a machine already compatible? Or are there some processes doomed to need special setup?

 

[jms_embedded]

What happens to a wafer (or multiple wafers) that are in a tool when it breaks or otherwise requires maintenance? Do they just get rerouted to another tool of the same type? Or have to wait until maintenance is complete? Or are they scrapped?

 

[nghanayem]

What happens to a wafer (or multiple wafers) that are in a tool when it breaks or otherwise requires maintenance? Do they just get rerouted to another tool of the same type? Or have to wait until maintenance is complete? Or are they scrapped?

Depends. If it happens during litho, you just strip and recoat. If it happens in say etch, it is a scrap. Wafers that were completed get split off into a separate lot from the unprocessed ones, and they may be merged back together at a later point. Wafers destined for a downed tool automatically get rerouted to another queue as the down tool gets taken off of the network.