Going discrete means flexibility in choice of the R value - lower resistance giving less thermal noise. It would also allow a higher Vref.
Matching lots of resistors to the highest accuracy requires laser trimmed precision thin-film networks usually, so discrete is probably better, although you could argue such networks of resistors are an integrated circuit themselves... Laser-trimmed resistors on a silicon IC are good, but not as good as thin-film networks I believe (which are used for instance in 6.5 digit multimeters etc)
I think the fundamental simplicity of an R2R DAC is appealing to a lot of people: "you only need one resistor value to build it!". As is so often the case, however, the devil is in the details. The more bits your DAC has, the higher the precision needed for the resistors AND the switching element, in order to maintain monotonicity and linearity. So, really good R2R DAC's are going to be expensive. That said, simpler design solutions are often the best!
No. In fact, a discrete R2R DAC will have a hard time matching the performance of an R2R DAC in an IC. @Mark Tillotson explains why: Matching. Discrete resistors won't match well because of their absolute tolerance and because they aren't all at the same temperature. Even if you get good resistors, say ±0.01 % tolerance, few ppm/ºC temperature coefficient you still can't guarantee better than ±0.01 % matching. That may sound great until you realize that's only about 12 bits of precision.
You can get better matching with a resistor network, but even those that offer excellent matching seem to sit at ±0.01 % matching, at least from what I can glean from a quick search on Mouser. It's possible that someone out there makes better resistor networks. But with a, say, 20-bit R2R DAC you need good matching between 60 resistors (3 per bit, right?). That's a very tall order.
By contrast, resistors on an IC can be sized and laid out such that good matching is ensured by design. 14-bit R2R DACs were SOTA 40 years ago. In a modern precision analog process you should be able to get closer to 16-bit performance.
The central point of the R2R topology is that it can be constructed from a string of identical resistors. Just connect two in series for the 2R legs. That's easy to get to match.
Tom
One could re word and ask if it is feasible to have discrete the same bit resolution of a well made R2R dac chip like the ad1862 for illustration.
People claims it is feasible : Holo DACS ? I don't think because what Tom said. (then add drifting of the resistors in time plus dust in the pcb substrate, then good luck !)
It is said a 20 bits, but if having 18 real bits due to my RED BOOK cd libraries, I'm happy enough for that, all he thing is to do it well around the chip then.
Then comes some others design that are not 100% R2R and works good with less bits resolution if not better to the ears : TDA1540, TDA1541....
People claims it is feasible : Holo DACS ? I don't think because what Tom said. (then add drifting of the resistors in time plus dust in the pcb substrate, then good luck !)
It is said a 20 bits, but if having 18 real bits due to my RED BOOK cd libraries, I'm happy enough for that, all he thing is to do it well around the chip then.
Then comes some others design that are not 100% R2R and works good with less bits resolution if not better to the ears : TDA1540, TDA1541....
It's in the details.... The problem with single chip or thin film networks, is matching, it can be done during manufacturing with laser trimming, but that gets expensive, that why TI (Burr-Brown) discontinued all their R-2R DACs.... You can overcome the problem with very tight matching by using the Sign Magnitude R-2R principle, then you don't need that precise matching. And as the precise SMT thin film resistors have become relative cheap, together with automated SMT assembly, you can get very good results, see the dam1021 thread, I might be biased as I'm the designer....
Søren
Søren
Only the upper 6 bits need ultra high precision in a 16-bit 2R-R DAC. Fun fact: the LSB resistors can be as high as 20% tolerance. The MSB and other upper resistors can be trimmed by an external element. Important is a low TC, or a temperature controlled environment.
"Driiiiift are my reality, ...."😀
Off topic, sorry : did you have decoupled AOL/AOR to +5V finally ? (3 to 10 nF)
Off topic, sorry : did you have decoupled AOL/AOR to +5V finally ? (3 to 10 nF)
I have a Ferranti ZN425 8 bit R-2R DAC. It's old.
"Bitstream" DACs are inherently superior.
Time is more accurate than resistance.
"Bitstream" DACs are inherently superior.
Time is more accurate than resistance.
Are you sure? Won't you get crappy INL if parts of the R2R ladder is imprecise? I haven't done the math, so it's possible that you're right, it just doesn't sound right to my ears.
Same trick can be employed on die.
I doubt TI discontinued their R2R DACs because they required trimming, though trimming in production is generally frowned upon for a number of reasons, cost just being one of them. I bet people stopped buying R2R DACs when they could get higher resolution from a delta-sigma DAC.
If TI wanted to bring back R2R DACs they do have a good process for it. One of the processes they bought from National Semiconductor is optimized for precision analog and, among other things, in includes precision thin film resistors with excellent matching. There's a blurb about it here: https://www.electronicdesign.com/ma...conductor-op-amps-based-on-new-bicmos-process
Tom
No it was too high cost for production and on chip calibration both at Philips and (then) Burr Brown. This was published several times then.
The pcm1704 was Sign Magnitude, the EOL of them inspired me to make the dam1021.... You do need a certain precision, but the requirement go down with the LSB bits, that's the whole point of Sign Magnitude, while Regular R-2R need high precision even for the LSB's.
All thin film resistors need calibration to end up precise.... I once visited Vishay's thin film resistor factory in Germany, partss are manufactured to the required temperature coefficient, then laser trimmed to desired precision, easy with "large" discrete parts, more complicated on a chip....
As Jean Paul said, cost killed TI's R-2R DAC chips, it's much cheaper to manufacture in a bulk CMOS process, than a low volume linear process that need laser trimming.... But TI should consider to use their e-trim technology they use for offset trimming in precision opamps....
All thin film resistors need calibration to end up precise.... I once visited Vishay's thin film resistor factory in Germany, partss are manufactured to the required temperature coefficient, then laser trimmed to desired precision, easy with "large" discrete parts, more complicated on a chip....
As Jean Paul said, cost killed TI's R-2R DAC chips, it's much cheaper to manufacture in a bulk CMOS process, than a low volume linear process that need laser trimming.... But TI should consider to use their e-trim technology they use for offset trimming in precision opamps....
Ah. Thank you!
They could do that. But trimming is still an expensive process. They'd have to test the parts at least twice: Once before trim and once after. They'd likely run the wafers through sort too, to ensure that they don't package the obvious duds. So that's three test steps. Plus the E-trim step. That's a lot of cost. They'd need a customer who's willing to bear that. And, sadly, a bunch of DIYers wanting free samples doesn't make for a good business case... 🙂
Tom
Why did Philips need on-chip calibration? They never manufactured pure R-2R audio DACs, but DACs with dynamic element matching for the higher bits.