Electron microscopy shows ‘mouse bite’ defects in semiconductors
news.cornell.edu88 points by hhs 5 days ago
88 points by hhs 5 days ago
Metrology is probably the most important part of semiconductor manufacturing aside from actually putting features on the wafer.
Statistical process control is at the heart of profitability, and measurement of what we've actually built is how it gets its data. If the accuracy and frequency of measurement goes up, the control loop tightens accordingly.
Parameterizing features and defects is a really interesting multidisciplinary process. Figuring out how to correlate defects at EDS time with something that occurred 80 process steps ago is where all the money lives in the business. Once you draw the correlation, you can place it under SPC and people will automatically get paged in the middle of the night the moment something starts to drift into an unhappy range.
And people wonder why I laugh when they say "non-mechanical devices are more reliable." Sure mechanical devices need pieces to stay moving in the same way over and over again, but electronic devices need a huge number of very precisely-placed atoms to not move in any way, including chemically.
There's a question about the right units here. Your CPU performs more operations in its first millisecond* than most mechanical devices do in their entire lifetime. So in per operation units, they are staggeringly reliable.
Actual operating life is often determined by the economic feedback loop which causes manufacturers to cut costs until basically all consumer products have roughly the same expected lifetime, regardless of the potential of the underlying technology.
* Or at least, the first millisecond after it starts using its normal operating clock, which might not be the very first millisecond
You can easily make it more reliable just by making larger features, and then no mechanical thing can even come close to compare to how reliable digital circuits are. There is a reason 40 year old NES consoles still works today, they are very reliable because the circuits are so large. To break those old machines you have to essentially melt them.
They work, as long as you replace all of those leaky caps (and maybe the ceramic resonator)!
Reliable in the sense that after initial testing, it'll be stable over time.
In what context are people saying this? I’ve never heard anyone proclaim that electronic devices are more reliable than mechanical devices. See for example how people desire cars with “no computers”.
The problem is not that modern cars are somehow less reliable than old cars. They are much more reliable. But they’re also much less repairable without specialized equipment. You can with somewhat accessible technology repair almost all defects on a purely mechanical car. You cannot do the same for a modern car unless you happen to have a chip fab.
Ok, but how are you going to make a billion mechanical transistors? Even if you could, latency would be limited by the speed of sound in whatever materials you use.
> At Bell Labs, Muller and fellow scientist Glen Wilk ’90, who is now vice president of technology at ASM, tried replacing silicon dioxide - the prevailing gate material, which leaked too much current at small scales – with hafnium oxide.
They are naming professors like "Now That's What I Call Music" albums now?
(I genuinely can't find why there's a '90 there, suspect it's a copy/paste error?)
Presumably because he is a Cornell alumnus from 1990. The article is at cornell.edu .
Listing alumni degree year is generally an "insider" thing (noone who isn't also a Cornell alum really cares which year, especially for a bachelor's degree; likewise Cornell doesn't mention the Harvard '95 PhD in Applied Physics, even if it's probably more relevant to the work...)
"insider" thing, you can be certain that an exempliary mind such as his did not get fired up in a vacume, and that 90, was a year and place that likely produced an iteresting mixed bag of characters, or in other times would be refered to as "schools of thought", and then some went off to bell labs which still functions as an intelectual singuarity that leaks concepts through it's event horison ocasionaly, or in this case displays time dialation effects.
I first thought it was supposed to be a comma, and that he's 90 years old.
University news sources often cite alums of the university with their year of graduation.
Put out some really tiny mousetraps then?
any hope that this could be applied to improving memory fab yields and ease some of the capacity constraints on consumer devices? asking for a friend
For now, this is useful only for the processes used to make the latest CPUs, like Intel Panther Lake and Clearwater Forest in the 18A process and various CPUs for smartphones or computers that have been launched recently or which will be launched around the end of the year and which are made with the TSMC 2 nm process.
Memories use more mature fabrication processes, for which it is likely that electron microscopy already worked well enough.
The article is about a better method for processing the output of an electron microscope, which enables a better image resolution than in the past and the 3D reconstruction of the surface of the device. This is needed for the 2 nm/18A processes and their successors, for which the existing tools were insufficient.
Silicon has 23 known isotopes, and now you why it will unlikely ever be economical to reach 0 defects in a business context.
Modern chip designs do include over-provisioned features, so designers can often selectively downgrade areas that are not viable.
Chenming Hu books about solar cell physics and semiconductors are quite accessible. =3
Solid work. The technical details are appreciated.