about reference currents and buffering

Hello,

I want to design a 12bit current dac. therefore I need a reference current source.
I already have a quite precise reference voltage of a bandgap with less than 1% variation on chip.
How can I generate a reference current with the same accuracy?

The poly resistors in my process are quite bad.

Thus I wondered if it is possible to mirror the current directly out off the bandgab circuit.


When would I need to buffer this dac reference current and how is this normally done?

scoobydoo

Generate current sources using this band gap. Various methods like Voltage followers , B-multipliers are available. Generally Bandgaps are preferred not to be loaded.

Hello scoobydoo,
If you mirror the current from the BGR, this current will be I = delta_VGS/R, am I right? So it will continue to be dependent on the resistor tolerance (some times, 20%).
If you can trimming the generated current, your work can be easier.
It may be better if you use a current source that does not use resistors. (I have never designed 12bit current dac).
Regards,

hey my bandgap is very precise. it varys under process and missmatch simulation by less than 1%.

an external resistor is only the last choice.
is there an other way to generate a reference current with less than 5% variation over process?

Hello Scoobydoo,

I'm sorry....I think I was not clear. I didnt say that you reference is not precise. But there is significant difference betweeen the output voltage and the current.
The output voltage can be very precise, but I'm not sure about the current. Do you agree?
this current depends on directly the value of the resistor....do you agree? So... it seems that it does not varies less than 1%. If yes, your resistor varies less than 1%. Is it true?
There are some examples of current sources that do not use resistors. I would use one of them.

scoobydoo said:

is there an other way to generate a reference current with less than 5% variation over process?

Click to expand...

Try E. Vittoz' resistor-less current reference (Fig. 8.8 (b)) : View attachment Resistor-less_constant-current_reference.pdf

Hello Scoobydoo,

erikl gave a very good example. Another one is :

A Resistor-Free Temperature-Compensated CMOS Current Reference. Wei Liu, Waleed Khalil and Mohammed Ismail.2010.

I hope it helps you.

thanks erikl and palmeiras,

i think that will help.

i just have one question left. assume that there is a current reference as shown by erikl in fig. 8.8 b). is there a problem to connect let´s say 30 currentmirror transistors in the same way as M5 to distribute the current all over the chip?
i read that bandgaps and reference circuits should be buffered because they cannot drive any load.
do i need any kind of current buffering there?

scoobydoo said:

... is there a problem to connect let´s say 30 currentmirror transistors in the same way as M5 to distribute the current all over the chip?

Click to expand...

Not at all -- from a dc point of view, because this is a voltage (Vgs) control method, and there's practically no dc gate input current loss.
If the 30 (or so) gates' connection is rather long, however, there's the danger of EM stray pick-up, so you should perhaps add one or more capacitor(s) between this gates' connection and ground (or VDD) to filter such noise.

scoobydoo said:

i read that bandgaps and reference circuits should be buffered because they cannot drive any load.

Click to expand...

That's true.

scoobydoo said:

do i need any kind of current buffering there?

Click to expand...

No. As mentioned above, the constant current(s) are controlled by a constant Vgs voltage, which, on its part, is already buffered by the M4 "diode" connection, whereat the gate inputs don't constitute any dc load.