Autopsy of a Pentium III


I took a job for a Japanese DRAM and EPROM manufacturer in the 90s. The management culture was ridiculously rigid and questions or even interest outside of work responsibility were met with suspicion.
My favorite part was the microscope inspections of several sample wafers per carrier, I got the intro to the SEM too which was pretty cool.
I was fired for incompatibility issues after three weeks I had in the fab but still got my signing bonus a few weeks later.

Marvelous! The pictures of the chipped-off edge show the layers very nicely. I don’t think I saw that kind of image yet, it’s always the above or side view, of pristine samples, in the textbooks.

For bonus points, use focused ion beam to mill away parts of the chip to selectively gain access to deeper areas.

Also, a thought. What about a single-pass SEM? Something where the beam would not keep scanning the sample but do a single pass and store the image in computer? Would deliver much less charge to the sample, so less blurring of nonconductive areas.

Similarly, what about ionizing radiation assistance? With a blast of xray or UV, even nonconducting things can be made conductive. Could help to dissipate the charge.


The best way to get a good cross section image at the lower layers is to mill a thin lamella with an ion beam and put it in a TEM.

For perspective, that was 250-180nm node. Now production of cutting-edge logic is mostly on the 22nm node, with 14 rapidly ramping up. The technology that goes into ASML’s current generation of lithography tools is utterly mind-boggling, and there are an enormous array of materials science tricks used to scale down further from the smallest features we can print.

I always liked to crank the voltage on the SEM to 11 and just blast my way through a sample. Nuke it from orbit, it’s the only way to be sure :slight_smile:

Another option would be to gold coat the sample if you want to light up the non-conductive areas.

1 Like

I’m curious about this, if you don’t mind sharing. I’m oddly interested in workplace culture stories.

Cranking the voltage results in getting information from deeper in the sample, but less surface resolution.

We had a PIII (dual processor!) server we in a customer’s server closet, but alas, after I got a bit worried about it yesterday and ran a backup, today it finally went to the big server room in the sky :frowning:
To be honest, I think the strain of tarring/gzipping everything was probably the last straw for the hardrives, but I managed to copy the backup off this morning (and the md5 matched).

1 Like

Symptoms of the failure? Did the CPU/mobo fail? The disks (how, bad sectors, or spindle failure, or…)?

Not sure yet, my boss is up there today for a meeting, so I hope he’ll have a look. Last time I was up there the disks sounded like like all their lubrication had be replaced with fine sand, so that’s my guess.
We might just replace it with a Raspberry Pi.

1 Like

This topic was automatically closed after 5 days. New replies are no longer allowed.