What's happening here is a phase transition in the liquid crystal; the stuff we bought is specially formulated to transition from its optically-active nematic phase to its transparent disordered phase at 29 °C, quite close to room temperature. It's very sensitive; temperature differences of less than 0.1 °C can be detected with it--of course, it's limited to telling you whether the temperature is above or below 29 °C, but that's all we really need.

By very carefully adjusting the voltage applied to the chip, we can get to a point where the black spots just barely appear, where the hottest points on the die are *just* above 29 °C. And that tells us, with reasonable accuracy (subject to thermal conduction and heatsinking effects of metal layers on the die, anyway, which can change the apparent location of the hotspot by a few microns) exactly where the fault is located.
-F

and best of all, this liquid crystal is cheap

our work paid $135 for 1 mL of liquid crystal dissolved in 10 mL of carrier solvent (here, ethanol, though i'm kind of wishing we got the one dissolved in dichloromethane instead, or even the 1 mL of pure liquid crystal with no solvent; we've had some difficulties with the ethanol)

That may sound like a lot of money for very little stuff, but almost none of the stuff actually gets used. we could probably do this whole process several thousand times with what we have in that bottle, it's a very thin film of liquid crystal

compare that to spending $10k on a thermal camera, then another $6k on infrared-rated microscope lenses (yes, really), and even *then* having relatively poor temperature resolution making it difficult to pinpoint the hot spot
-F

anyway after two weeks of working on that and several long meetings poring over kLayout and/or Cadence trying to line up photomicrographs of the hot spots with the layout files and figure out why the *fuck* the hot spot would *possibly* be *there* of all places, we've identified a plan for fixing it and, with luck, we'll have working wafers in a couple weeks

after identifying the problem in the first wafers we got, we immediately told the foundry to stop production on the next batch they were working on, and it's looking like we may be able to salvage those by changing one layer that hasn't yet been deposited, which is a major relief--we're not too up to speed on the financials but it seems like those wafers cost several thousand dollars so far, so losing them would be unfortunate for the company wallet (and it's a startup company, so that wallet isn't looking too great to begin with)
-F

@Felthry wow, this thread is *fascinating!* that's cool as hell

@gardevoir it really is!!!
-F

@Felthry that's extremely fuckin' cool

@starkatt it really is! and it's remarkably precise; depending on how close to the surface of the die the fault is located, and how well the liquid crystal is applied (a thicker film means worse resolution), you can pinpoint the location of the fault to within a micron or so
-F

@Felthry frick!!

@starkatt i have pictures and video of it that i'd *love* to show but i don't think i'm allowed to, unfortunately
-F

@starkatt oh yes something i forgot to mention is that you can actually make this even more precise--if, instead of having the wafer under test at room temperature, you use a heated chuck to control the entire wafer's temperature to *just* below the crystal's clearing point (here that would be like, 27 or 28 degrees), you can roughly double the sensitivity

again, though, that's still subject to whatever's in between the actual hot spot and the surface of the wafer where the liquid crystal is; metal layers especially can act like heatsinks and pull heat away, leading to the hot spot appearing to be wherever there happens to be less metal near the actual hot spot
-F

@starkatt so it's not a completely infallible method for finding the problem, but it *is* one of the coolest things we've done this year
-F

@starkatt (you can also get liquid crystal with clearing points at different temperatures, as much as 95 °C, presumably for identifying failures that occur only at raised temperatures)
-F