Analog Devices AD1139

or maybe some incredibly edge software-defined radio application...

Even average soundcards sport a 24bit DAC. However they're designed for a dynamic output so tend to have quite crap DC precision. What matters to a soundcard is producing a precise waveform, not a precise DC voltage.

This part is designed for DC level setting, ie a programmable reference voltage. Typically it won't change that often (compared to a soundcard), but it needs to be precise.

For example, compare the AD5791[1], a 20-bit DAC with 1LSB DC accuracy costing $40 in 1000 qty, to a 24bit audio DAC like the AD1955[2] costing about $7. The datasheet for the audio DAC doesn't even specify the DC precision, only a 6% gain error from the output buffer.

[1]: https://www.analog.com/en/products/ad5791.html

[2]: https://www.analog.com/en/products/ad1955.html

1) Loudness war means pro sound engineers will mix everything to be full scale all the time so music has 0 dB of dynamic range ideally

2) Almost no one owns equipment capable of rendering 144 dB of dynamic range even if they have a bit stream specifying it. So if a quiet near silent room is 30 dBA where you hear your own heartbeat and OSHA permissible exposure limit is 90 dBA its only 60 dB of dynamic range between "too quiet to hear" and "so loud you legally need earplugs" so you'll be throwing out about 84 dB of theoretical dynamic range anyway even if you owned gear capable of rendering it. Note that its very hard to buy audio amplifiers capable of more than 90 dB or so SNR even if just hooked up to test equipment, nothing on the market capable of 144 dB, so either the bottom end will be lost in the hiss or the high end will be distorted into unrecognizability.

3) Then you get into topics like power supply rejection ratio. A "bad" amp would pass noise from the power supply thru. A "better" amp rejects more noise from the power supply. A good way to look at it is the hummmmm and crackle and stuff that you hear on a bad old tube amp could be modeled as a very large value attenuator directly connecting the power supply to the output. With 1970s tech and capacitors, 80 dB was considered pretty good PSRR. The very top end newest class D amps can just barely exceed 100 dB PSRR under idealized testing conditions. So at full blast (which would just blow your eardrums out anyway) everything more than 100 dB down from peak will merely be power supply noise passed thru the system.

Its kind of like I can spec a piece of wood to 1.5875mm thick. And then convert it to imperial inch measurements. If the best measurement tool I have to measure wood, is a carpenters tape measure ruled to eighths of inches, the best measurement I can do to render those five digits of precision is "about half a division" on that tape measure. The five digits of precision in the mm figure imply I should be using a calibrated micrometer to measure that thickness, but if you assume technology hasn't invented them yet, then there's little point in spec a digital calculated measurement to five digits of mm precision. Likewise I can make a string of binary 1 and 0 that theoretically can be understood or mathematically proven to imply 144 dB of sound precision, but even in 2021 electronics hardware doesn't exist to reliably repeatedly and provably render those 1 and 0 into sound, at least not at audio rates.

Note that 2 to the power of 24 is a eight digit number, implying you're operating at tens of ppb. NIST has really nice new Josephson junction standards capable of reliably operating at tenths of ppb. So you're very optimistically trying to field gear operating at only a hundred times less accurate than the best lab in the entire world can currently measure. Good luck with that. That's why nobody sells voltmeters with more than 8 or so digits of precision, if the world standard calibration system only has ten digits on its best days, no point building voltmeter hardware displaying 12 or 15 digits LOL. We don't really have the technology to do 24 bit accurate voltages out in the field right now. So if we can't build or calibrate testing gear significantly better than 24 bits, like to 32 or so bits perhaps, there's no way to actually measure and tell if 24 bit gear is accurately precisely repeatedly working to 24 actual bits of precision. Maybe your audio gear is distorting in a subtle fashion and only operating to 22 noise free bit and the last two bits are essentially a RNG or stuck on/off or otherwise no relationship with reality; very few electronics labs in the entire world have the gear to prove that claim true or false.

In this case I’m wondering if it’s used to maintain better control of the voltage gradient on a pad.

He uses[1] Helicon Focus[2] for focus stacking.

A lot of these die shots are pieces of art, and also quite interesting. I especially like the LTZ1000 voltage reference[3] or the Thyracont Inclinos[4] which is an accelerometer without moving parts. A lot of the MEMS dies are also very interesting, because of their moving parts, like this[5] ADXL213.

[1]: https://www.eevblog.com/forum/projects/transistors-die-pictu...

[2]: https://www.heliconsoft.com/heliconsoft-products/helicon-foc...

[3]: https://www.richis-lab.de/REF03.htm

[4]: https://www.richis-lab.de/MEMS_04.htm

[5]: https://www.richis-lab.de/ADXL213.htm

Analog still lists them in production!