How to improve semiconductor process yield?

[NTMBK]

Don't worry, we know all about mining patents in the software world. Reading patent filing legalese is still quicker than reverse engineering binaries!

It's nice on some level to know that even cutting edge physical engineering is so dependent on human know-how and expertise! This is what I think of when I hear "intellectual property", not rounded corners (to choose the easy example).

This GloFo/Samsung deal seems unusually generous. Have you ever seen anything like it before? I saw some mutterings that it partly came about because Apple demanded a second source, though I don't know much to believe that- they are obviously an important client, but important enough to loan the crown jewels?

(As an aside, honest question- have you ever considered writing a boom? You could probably find a modestly profitable readership amongst us nerds, especially if you reduced overheads by publishing digitally. Obviously the revenue would hardly compare to a high end engineer's salary, but it could be an entertaining side project.)

 

[Idontcare]

So yields are the same?

Not the same, in fact never the same for the same reasons no two wafers are ever the same even when processed one after the other in the same production line. (answer: entropy)

The only way to get close to same yields is to resort to what Intel resorted to, copy-exact when fanning out process nodes across fabs.

If GloFo did a copy-exact of Samsung's 14nm then they would be entitled to expect essentially identical yields. But copy-exact is the most expensive method conceivable for fanning out a process node, which is why pretty much no one does it in practice (even Intel allows for some exceptions, albeit in a highly controlled and regulated manner where risks have to be vastly outweighed by the potential reward).

In general no two fabs will have the same yields, even within the same company running the same process node on the same tools. The human element still makes a difference, including time-zero decisions on the tool and plumbing layout within the fab itself.

But the highest yielding fab will always be used as the benchmark for the other less yielding fabs. If one fab has 80% yield and another fab has 75% yield, then the 75% yielding fab will be held with its feet to the fire to close the gap.

This GloFo/Samsung deal seems unusually generous. Have you ever seen anything like it before? I saw some mutterings that it partly came about because Apple demanded a second source, though I don't know much to believe that- they are obviously an important client, but important enough to loan the crown jewels?

There is a lot to the story, most of which I suspect will probably never enter the public domain. It is a tale of desperation combined with opportunism, on all sides.

(As an aside, honest question- have you ever considered writing a boom? You could probably find a modestly profitable readership amongst us nerds, especially if you reduced overheads by publishing digitally. Obviously the revenue would hardly compare to a high end engineer's salary, but it could be an entertaining side project.)

I'll never say never, but the traditional article publication process just isn't something that excites me. I've tried, multiple times, and have been the benefactor of some wonderful publishers doing everything they could to work with me to make something happen. But every time something gets underway I find real life interrupts and kills all the momentum.

Forum posting has the flexibility that you can do it in whatever frequency and duration that fits into the minutia of free time available during the day, week, or month. Article formation and publication doesn't work that way, not effectively at least based on my experience.

But thanks for the kind thoughts and considerations! It is always a rewarding moment to read and realize that others appreciate what I post on occasion :$ 🙂