Kai said:

"By the way, to fabricate a 16bit DAC with discretes like R2R ladder and such stuff will never work."

Steve said:

Actually, if I were to roll my own, Charles' method can be remarkably successful - these hybrid film devices track very very well and their temperature drifts are also stable to your required precision. How do you think "16 bit DACs" on monolithic silicon work ?

Steve,

ok, my statement could be misinterpreted. In the phrase 'to fabricate a 16bit DAC with discretes like R2R ladder' I meant R2R ladder build with discretes, means single resistors with a wire at each end.
Charles' method is based on an 'integrated passive network'. Of course, an 'integrated passive network' has a good chance (and is the only that has a chance!) to provide necessary precision.

I wanted to focuse, that the common try to build a 16bit DAC with common single resistors, common 'reference' voltage and common operational amplifiers will totally fail, because common discretes are much too unprecise, even if they are called to be very precise. Resistors must match to less than 15ppm and total offset voltage must be lower than 150µV. And this with all temperature changes and after several months or even years.
An example: Standard precision resistors have a tempco of +-50ppm/K. Ultra high precision resistors show a tempco of +-5ppm/K. To guarantee 15ppm precision, these resistors may only undergo a temperature change of 3K. And any long therm drift is forbidden. Such a discrete resistor is hardly to be found...

With an 'integrated passive network' situation is totally different, because then all resistors are very similar in every physical parameter, as you stated. They undergoed the same production process, same materials were used and because of this very same tempco and long therm drift is to be expected. Also, special selected hybrids are available, which showed in the factory during a 2 years period an ultra low drift performance.

Kai