Cameras similar to this have been around for decades, but they weren't as automated, and they were more expensive since they had longer Depth of Field for research purposes. But fundamentally, putting an IR back (or front) illuminator on a microscope with IR lenses and a CCD was necessary and available for wafer stacking decades ago (I used them in the early 90s). This looks like it has a motion stage and automated scanning, which is important for doing the verification!
I don't think you're going to get crazy high resolution. The minimum wavelength of light that goes through silicon at even 0.5% (it's >50% down to 1.2um) is ~0.9um. There are specialized techniques (top illumination) for super-resolution like confocal microscopy that can get quarter-eighth wavelength ~120nm as described below, but that's about it. A bigger issue will be that, if there is too high a density of metal, you just won't be able to see through. However, you will almost certainly be able to see the large features, the structural areas (RAM/Logic/Serial) for identification, and the Through Silicon Vias (TSV) used for aligning die together for MCM, but not the underlying logic.
What a super important project! Very well done to the researchers and
I really, really hope you get funding and/or people joining and taking
this all the way. Please keep it up!
Been involved with a physical supply-chain assurance technology
project and this is the first time I heard of this kind of IR chip
scanning.
For those that don't understand; securing the provenance of physical
hardware is the foundation of all other cybersecurity.
VCs and money people... this is where you should be placing your bets!
It doesn't actually matter if/that it's not terribly effective at the
moment. Keep building and roll it out. Get people using cheap scanners
and publish open hash databases. Even with false positives it becomes
like "effective security theatre" and is a massive deterrent to
anyone attempting implants or stealth logic modification at silicon or
board level.
Given that silicon has quite a high refractive index, I do wonder if it would be possible to pull the same trick as the "immersive lithography" people here. But I don't know enough about optics - I wonder if you can get away with a tiny air gap if most of the optical system is from higher refractive index material.
That's an interesting idea. I think you might be able to do reflective optics around a single crystal silicon core in near contact to the wafer backside. You'd get another factor of 2.5x, which along with confocal imaging would be 50-65nm and that's almost good enough to image modern cells.
I guess this method would put some constraints on PCB design, since the need to scan the die physically would mean that you'd need a margin around it where nothing protruded higher vertically.
I don't think you're going to get crazy high resolution. The minimum wavelength of light that goes through silicon at even 0.5% (it's >50% down to 1.2um) is ~0.9um. There are specialized techniques (top illumination) for super-resolution like confocal microscopy that can get quarter-eighth wavelength ~120nm as described below, but that's about it. A bigger issue will be that, if there is too high a density of metal, you just won't be able to see through. However, you will almost certainly be able to see the large features, the structural areas (RAM/Logic/Serial) for identification, and the Through Silicon Vias (TSV) used for aligning die together for MCM, but not the underlying logic.
https://www.tydexoptics.com/materials1/for_transmission_opti...
https://www.photometrics.com/learn/spinning-disk-confocal-mi...
The resolutions needed for reverse engineering will still rely on layer etching/removal and SEM.