That’s a Big Microscope | Hacker News Reader

That’s a Big Microscope

bunniestudios.com

95 pointsetiam32 comments

Synchrotron lightsources are a dime a dozen. For real power, you want a free electron laser. The Linac Coherent Light Source at SLAC (right in Silicon Valley's back yard) pumps out an X-ray beam with about 750,000 times as many photons per second, and in pulses that are about 200 times faster.

https://portal.slac.stanford.edu/sites/lcls_public/aboutlcls...

If you want even larger systems, you should check out the European XFEL, which was just opened, and is now the most powerful free electron laser in existence. http://www.xfel.eu/

It’s quite a sight.

As a die hard LCLS supporter, I have to say: XFEL has the more powerful electron beam, but LCLS is still in the lead in terms of X-rays right now (if pulse intensity/peak power is the measure).

XFEL is really an amazing machine, though, and once it is fully commissioned, it will take the X-ray crown as well.

Couple questions:

1) How do you steer an FEL?

2) These seem like they'd be pretty good for destroying hardware (drones, cell phones). Is that true?

On (2), I do know that FEL analysis of small samples such as the ones for protein structure analysis entails completely disintegrating them (off goes all the electrons, then the nuclei quickly realize they have nothing in common anymore and decide to go their separate ways), and analyzing the structure is done on the basis of reading the radiation scatter patterns from the instant between when the beam makes contact and when the molecule has utterly lost the structure in question.

I think it would be safe to say FEL isn't exactly non-destructive for hardware.

That is pretty intense. I wonder sometimes if it is a deterrent to nation states to put 'secret' back doors into silicon because a device like this would be able to prove that the device had a back door.

That said, converting a 3D image of a microprocessor level into a Minecraft world would be pretty awesome!

I think proving that there is a backdoor isn't a concern. A useful backdoor could probably be passed off as an engineering error. The bigger issue is that the backdoor, if exploitable by anyone, could become known.

TFA says it's only good to 14 nm but fabs are already down to 10 nm so I mean this isn't a deterrent. I don't know how STEM are for looking at chips or if we've got even more powerful microscopes but this particular one isn't good enough to stop a dedicated state level actor.

https://en.wikipedia.org/wiki/10_nanometer

I'm pretty sure you'd still be able to recognize structures/logic, even if you weren't able to exactly identify all the gory parts of a transistor.

the "14" in "14nm" does not refer to any particular dimension. Most structures in an 14nm chip are quite a bit larger than 14nm. It has been many years since the X in "Xnm" actually measured anything.

eg: https://www.extremetech.com/wp-content/uploads/2017/03/Slide...

More details exist online, for example this writeup: https://www.extremetech.com/computing/246902-intel-claims-th...

No, quite the contrary

Smaller processes are not very well refined at the moment, and anything that matters is 14nm or bigger

You're not building critical stuff on the smaller processes until they're very mature

TEM really isn't the right tool for reversing large integrated circuits. The sample prep is too finicky and involves slicing out a razor-thin sliver of material to image. You'd use it to get atomic resolution images of say a single transistor, maybe.

SEM with serial sectioning is the way to go. I'm sure people are using conventional SEMs for this already, but you could conceivably map entire circuits at high resolution (4nm) pretty quickly with one of the new multi-beam systems. Check out the SRAM cells on slide 17 o this SANDIA report (pdf warning) [0].

[0]: https://www.osti.gov/scitech/servlets/purl/1296536

A standard Scanning Electron Microscope can do down to 0.1 nm, so this isn't really letting us see anything we couldn't before. SEMs are regularly used to reverse engineer chips. The main difference is that the one in the article is non-destructive, whereas with a SEM you have to strip the sample (and do imaging) layer by layer.

...and a nation state has plenty of budget to buy even multiple chips to RE destructively.

Looking for backdoors could possibly turn up something interesting, but I would suspect that most backdoors if they exist are going to be in the microcode/firmware rather than the hardware. Assuming it's an encrypted blob you can't inspect outside of a running system, you'd probably have to scan both the physical layout of the chip and the electrical state, and then load both into some kind of simulator for debugging.

(Late to the party, but:)

You may be interested in the active line of academic research into making "undetectable" hardware Trojans by subtly messing with circuits. It looks like http://asiaccs2017.com/slides/Paar.pdf (currently down, Google cache: http://webcache.googleusercontent.com/search?q=cache:kHlrZta...) is a good introduction.

If you're interested in synchrotrons, you might want to watch Dave Jones's EEVblog 1 hour long tour video of the Australian Synchrotron, which is a similar size to the Swiss Light Source:

https://www.youtube.com/watch?v=27KRtJ-2sVY

That's the other approach to extreme ultraviolet lithography - a synchrotron. If you don't like the two-story machine that vaporizes tin droplets with lasers to make soft X-rays, one of these is an option. It's been tried experimentally.[1]

[1] http://www.lightsource.ca/

"chips down to a resolution of 14.6nm" so does that mean that that whole giant stadium of a microscope can't image at the resolution that ships in chips today?

(e.g. https://www.extremetech.com/computing/237781-samsungs-10nm-n... ).

please correct me if I've gotten it wrong!

That’s actually correct. If you want to go further, you’ll need to use Stanford’s Linac (0.15nm) or the just opened European XFEL (0.05nm)

Makes me sad working in an open plan office when I could be working in an inspiring building like that baby.

Having visited the old light source building at SLAC, the inside is not an inspiring workspace. It was never intended to be a light source, so the beamlines (i.e. the business end) were crammed under the struts and widgets that kept the synchrotron working. Not an economic win. Also, no windows.

The newer buildings are probably nicer. I bet they have foosball tables, too.

a little OT, but does anyone have some insight on the liability an author has to linking to an article on sci-hub like OP did?

Huh yeah whatever did happen to Judge Kaplan's ruling restricting certain speech that contained hyperlinks? (See http://www.nytimes.com/2000/09/07/technology/assessing-linki... and https://www.wired.com/2000/08/only-news-thats-fit-to-link .)

I find it sad that the reverse-engineering of microchips is considered a worthwhile activity, to the point of being the first (if not the only) thing that the poster thinks of as a potential application of this formidable device.

That is kind of Bunnie's thing and why someone would be most likely tuning in to the blog. Yes it could do a lot of other stuff, but the readers of that blog don't care so much about the other applications.

bunnie is a famous hacker, and this is his personal blog.

Why would it not be worthwhile? In a world or proprietary products and DRM systems, reverse engineering is a way to preserve our technological history and future.

I don't really get this attitude, what's wrong with reverse engineering?

I would really like to get opportunity to work with reverse engineering microchips using some reasonably good tools, and I really have zero reason to do it except that I think it would be super-fun! You can also bet a pretty penny that the first thought that pops into my mind when I see 14nm and 3D X-ray in the same sentence is: That would be awesome to reverse engineer some chips with! Not that I have the skills, not the time.

Figuring out how things work is just fascinating to bothe me, and a lot of other people. Doing it (nostly) without the plans is the only way it can be that real, and fun challenge for me. It's similar in character and as subversive as solving crosswords, or riddles at it's core. Since we don't go about complaining about people solving crosswords, why do reverse engineering always get so much flak?

It's almost literally the engineer version of crosswords or sudokus.

It's the difference between learning what someone else did and learning new, previously undiscovered truths of the natural universe.

Intel 14nm chips are not products of the natural universe. Understanding them does not lead to undiscovered truths. You could get this understanding by going to work at Intel. Or asking the Intel engineers nicely. Or whatever.

It's like saying "I will be the first human to chart my neighborhoods roadways." These charts already exist. Why not chart something that doesn't exist, instead?

If you do not learn what others have learnt, you are likely going to end up reinventing the wheel or worse continue making mistakes that lead nowhere. There is a reason academic papers have pages full of references.

We were talking about reverse-engineering someone else's design, which is in 99% cases amounts to stealing someone's secrets.

Reply on news.ycombinator.com