3 ms·

There is something weird about this story. The glitch happened at 5am on December 8. They got everything up and running at 100% capacity by 15:00 on December 10

There is something weird about this story. The glitch happened at 5am on December 8. They got everything up and running at 100% capacity by 15:00 on December 10. That's 2.5 days, at most. 2.5 days isn't 20% of two months, so why are they losing 20% of shipments over two months? The article says it's because "wafers in process were ruined". But in a fab, at any given time, most of the wafers are sitting in carriers waiting for robots to move them in and out of the tools. It's easy to see how a power glitch might kill wafers that are sitting in an etcher being etched, but not so easy to see how it might kill wafers that are sitting on a conveyor belt. It might stop the belt, maybe even for days, but then don't you just fall behind by a few days? Could it be a post-bottleneck disaster? Cartoon example: You spend six months making wafers, put the six-month supply into a big box, then drop the box. But that would be quite odd. Engineers don't tend to let nearly-finished wafers pile up in (metaphorical) big boxes and then process them in huge batches... precisely because such a practice makes you vulnerable to a tiny glitch like this one. Not that I disbelieve the story or anything. The number of crazy things that can go wrong in a fab is truly unbounded. And, though I once worked in a fab, that was a few years ago and it wasn't a flash RAM fab, so maybe I just don't appreciate the range of possible disasters. But I wish we had more details.

DougBTX 16y ago
But in a fab, at any given time, most of the wafers are sitting in carriers waiting for robots to move them in and out of the tools. If that was the case, I would have expected them to have bought some more tools by now!
sparky 16y ago
They could mean that wafer-starts-per-day were back up to their normal levels within 2.5 days, but the pipeline bubbles that result from destroyed wafers anywhere in the process almost certainly won't be filled for another month. It's easy to believe that the numbers cited in the article are fudged slightly -- fabs tend to be tight-lipped -- but there are also several stages in the process where many wafers could get screwed up at once, e.g., most ion implant machines process wafers in big batches. From a vulnerability point of view, you're right that it's foolish to do so, but you need to balance satisfying Little's Law with the cost and space of equipment -- it'd just be too expensive most of the time to buy enough single-wafer implant machines to match the rate of other, much faster machines.

DougBTX 16y ago

But in a fab, at any given time, most of the wafers are sitting in carriers waiting for robots to move them in and out of the tools. If that was the case, I would have expected them to have bought some more tools by now!

sparky 16y ago

They could mean that wafer-starts-per-day were back up to their normal levels within 2.5 days, but the pipeline bubbles that result from destroyed wafers anywhere in the process almost certainly won't be filled for another month. It's easy to believe that the numbers cited in the article are fudged slightly -- fabs tend to be tight-lipped -- but there are also several stages in the process where many wafers could get screwed up at once, e.g., most ion implant machines process wafers in big batches. From a vulnerability point of view, you're right that it's foolish to do so, but you need to balance satisfying Little's Law with the cost and space of equipment -- it'd just be too expensive most of the time to buy enough single-wafer implant machines to match the rate of other, much faster machines.