Author here: I'm sure there are people on HN who have experience with modern chip layout, and I'd be thrilled if you can explain more about the transistor structures and so forth. I mostly look at chips from the 1970s and 80s, and a lot has changed since then :-)
EE here. I agree with most of your guesses as far as layout goes. There are definitely some diodes in there - the devices with a more square aspect ratio and large square contacts in the middle.
The large power transistors are laid out in multiple "fingers" so current is distributed among them. This both reduces the effective diffusion resistance and alleviates concerns related to electromigration.
There are a lot more guard rings than I'm used to seeing, and they're very wide. This is probably to avoid latchup due to the potentially high switching currents, and possibly also for moisture protection ("seal rings" would be on every metal layer in addition to the front end layers).
There's one very odd layout structure that has me puzzled. In your "collection of transistors", the top left image has a non-rectangular structure in the middle. I wonder if there was a metal structure there that required a diffusion opening. Or maybe its a BiCMOS BJT?
I have a post about my setup [1]. The main secret is a metallurgical microscope: unlike most microscopes, this shines light from above, through the lens, so you get good illumination of the die. (A ring light is easier, but it doesn't help at high magnification. Also, light from the side doesn't work as well.) I use a low-end metallurgical microscope that I found on eBay. You can see interesting stuff with a regular microscope, so take a look at chips if you have one, but a metallurgical microscope makes a huge difference.
For the die photos, I stitch together multiple photos using the "Hugin" software package. I might use a dozen to 50 individual photos to produce a high-resolution image. Hugin has a learning curve, though.
The final thing is how to decap the chip. Usually, I stick to ceramic or metal packages that I can open with a chisel or hacksaw. For epoxy packages, I don't want to mess around with boiling acids, but I have friends who can do this.
Would you mind digging into those cheap Chinese isolated DC-DC converters in little black rectangular boxes potted in epoxy of similar wattage? "b0505" or similar
Those are hybrid circuits with real components and magnetic cores inside, not a single-chip solution, but I wonder what types of chips are used inside.
I didn't test the isolation. I wanted to avoid accidentally destroying the chip, since that would mess up the analysis. I was tempted to hook the chip up to the 40 kV spark generator we had sitting around, but that would probably be a bad idea.
Probably the package would need to be over 1cm tall to withstand that.
(I think that the electric power guys get "30kV/cm" etched into their brains - the field strength at which sparks start to fly between conductors separated by air.)
I love this sort of stuff on HN: seems pretty complicated at first, but the author dived deep into the meat of how it works while making each step extremely easy to understand.
These tiny isolated DC to DC converters are amazing.
When you think that they can withstand kV, some for minutes at a time, it's pretty impressive engineering, all in a tiny form-factor that still manages to be highly efficient and avoid over-heating.
I'm amazed by the knowledge that goes into making these, all the various disciplines that need to work together to make these tiny devices...
My guess is it's a mixture of analog and digital stuff. On the analog side, there's probably a bandgap voltage reference, temperature and current sensing, soft-start circuitry. It needs to detect the primary's clock and lock on it (probably with a PLL), and adjust the phase as necessary. Then there will be drivers for the power transistors.
The digital side is probably fairly simple logic to stop the chip if there's a fault and restart it if the fault goes away. From the datasheet, there are various different overload states it can go into, limiting power, changing to passive rectification, and then shutting down entirely. So the logic would need to manage these states.
There's always the possibility that they threw a microcontroller in to handle this, but it seems like overkill. Also, I didn't see anything that looked like ROM. So I think the control is hard-wired.
Nothing as fancy as calculations. If you look at the datasheet, there is an Enable pin as well as a clock input and open drain debug output. There's also an internal oscillator. The standard cells are probably used for those features, and possibly additional internal features used for testing.
It’s a switching regulator, so it’s generating PWM and adjusting duty cycles based on voltage feedback to produce a certain voltage on the output. Part of that means controlling the mosfet gates in the H-bridge. Some of that kind of thing can just be done in analogue circuits but often these kind of power supplies also can do things like hiccup modes for overcurrent, different waveforms for low load current, etc. so it could have a little state machine in it.
