When I sat down to reverse engineer every Nixie enthusiast's favorite Soviet chip, the К155ИД1 BCD-to-decimal decoder/driver (K155ID1 when you're at home), I had no idea that I would have to spend a day staring at it and trying to make sense of it.
Here is an NPN transistor on the chip:
Here it is with the contacts:
And here is the type of transistor I'm more familiar with, on a different chip:
The above image matches pretty much every diagram you see on the web of an NPN transistor. It's basically concentric: the outer region is the collector, the inner region is the base, and the region(s) inside that are emitters. The above transistor has two emitters. Here, let me annotate that:
Red is the collector region, blue is the base region, and green is the emitter region. The contact that touches the base region also touches the collector region, but through a bit of metal. That structure forms a Schottky diode which is connected from base to collector, but that's not the important point.
The key here is that the regions are concentric, which makes the terminals very easy to identify.
But we don't seem to have that in the К155ИД1.
After staring at the image, and doodling various attempts at cross-sections, I finally came up with this explanation. Let me put the original image in here again, marked up with what I think is going on, and follow it up with the cross-section I settled on.
NOTE: All of the following discussion about transistors turns out to be wrong. This was my first analysis.
Unlike in other chips, the n+ buried layer is very prominent, raising a distinct hillock-looking region. Another major difference, looking at my drawing above, is that where normally the p-doped base surrounds the n-doped emitter, here the base appears to be separated from the emitter on the surface, although I think under the surface the base does surround the emitter. So on the surface, it only looks like the base and emitter are separate.
The whole rectangular area is n-doped, and surrounded by p+ isolation. This isolation is grounded.
Another source of my puzzlement was that the emitter and base overlap the hillock. But I guess that as long as they are all within the n-doped collector region, it doesn't really matter.
I'm not enough of an IC fabrication person to say for certain whether this is correct. But it seems to explain all the visual characteristics.
Update: All this turned out to be wrong, as I discovered when I tried to trace out the circuitry. None of it made sense. The next post discusses another analysis.
The next unexpected thing is how resistors are formed. Here's what I'm used to, from a 74LS01 chip:
Resistors are formed of polysilicate or highly n-doped "wires". The longer and thinner, the more resistance. So here's what greeted me for the many resistors that are supposed to connect transistors to VCC:
The black squares are contacts. Where are the resistors? Here is the image with the metal layer:
The wires that terminate in contacts in this VCC-connected region are clearly meant to go to resistors to VCC. So my thought is that the whole thing is just one big planar sheet resistor. The designers perhaps had a good model of what resistance would be formed to VCC at any point in the area.
This VCC-connected area seems to be surrounded by p+ isolation, is probably n-doped, and seems to have a large n+ buried layer underneath. The disturbing thing is that the wires terminate closer to each other than to VCC terminals, implying that they are connected by smaller resistances to each other than to VCC. Presumably that doesn't seem to matter, since the IC clearly works just fine like that.
Anyway, this just confirms my idea that each group of chips is its own special snowflake.