Thread by @TubeTimeUS

this is an LM168 voltage reference. i decapped it and took these pictures of the die. it's probably the worst analog layout i've ever seen. check out these weird problems! (thread)

there some transistors that are shorted out and don't seem to do anything. there are tons of crossunders in a few spots, and some mystery devices. there are bits of metal and resistors just hanging out, not connected to anything.

the datasheet has a schematic, but the schematic doesn't match the die! why not? turns out, i know the reason because i have a secret weapon (you'll never guess!)

this is the *original LM168 breadboard* which was used by the IC designer to develop the circuit! the metal cans are what's called "process kit parts" which are matched pair transistors from the same IC process. basically a physical equivalent of a PDK SPICE model.

so why all the weird problems? there were two versions of the part! the first version had problems, so a second IC designer was brought in to fix it.

that person was none other than Bob Pease, the "Tsar of Voltage References" at National Semiconductor. he actually mentioned the LM168 in a column: "[it] was the LM168, which I got working. Some other poor turkey designed it, but I got it working and into production. /rap"

you can actually see Bob's rework here. he removed three transistor pairs, tacked on three new ones, and wired them in with the green wires. pretty sloppy, but it got the job done.

so what did he change? here are the two schematics side by side. i had to make a few guesses about the turkey's version, and so there are probably errors in my schematic. mostly he tweaked the current mirrors driving the reference cell Q6A & Q6B.

there are some PNP transistors in the old version that Bob removed, but you can actually still see them on the layout--these spare PNP transistors that Bob just wired up (all three terminals!) to VOUT.

Bob Pease was an old school IC designer who liked to build the circuit with real live transistors instead of messing around with computers. this is what he thought of SPICE simulation.

here's the top metal all colored in and the components identified. these designators match up with the schematic.

i've annotated the schematic to point out the functional bits. the current source i don't fully understand mostly because my die imagery is not great. notice the cute output overload protection circuit made by Q16, R22, and R4.

here's something neat: these are laser-trimmed resistors. during production, there is a machine that measures the output voltage and zaps these resistors to dial it in. check out the test structure on the right. you can see marks from the wafer probe on the pads.

here are two NPN transistors configured as a current mirror. the emitters are tied together and go to ground (orange). the base and collector of the bottom one are tied together and tied to the base of the upper one.

these are lateral PNP transistors. compared with a 2N3906, they have very low beta--i measured 30 on the kit part--but hey, at least they work!

besides all his technical books and columns, Bob Pease wrote this book on car accidents. in a cruel twist of fate, he passed away in a car accident in 2011 as he was leaving the funeral of his friend, famed analog engineer Jim Williams.