Current mirror matching for DAC

[edinburghtech]

Hi

Can some one please help me understand the current matching requirement related to the resolution of a DAC. For example, if I am designing a 10 bit current mode DAC, how well the currents should match (Current through bias network is 100uA) . Also please suggest me some books/papers from where I can study this.

Once I have the matching requirements fixed, how can I verify this?

Thanks
Tech

 

[erikl]

... I am designing a 10 bit current mode DAC, how well the currents should match (Current through bias network is 100uA) . Also please suggest me some books/papers from where I can study this.

Current matching should correspond at least to the required resolution, i.e. 1/2ⁿ or better. In your case 1/2¹⁰ ≈ 1‰ or less. G00GLE for the corresponding literature.

Once I have the matching requirements fixed, how can I verify this?

By simulation of either all input voltages bit by bit, or using 2 or more different input voltages separated by 1LSB each, at several 2ⁿ borders. e.g. digital input values 0, 1, 2, ... 15, 16, 31, 32, 63, 64, 127, 128 ... 1022, 1023 bit, and checking the output voltages' accuracy.

 

[Dominik Przyborowski]

You need to use any mismatch model (e.g. Pelgrom model which is simplest but works only for saturated, strong inversion devices and overestimating current source mismatch). The methodology for estimate a current source dimensions you can find in Jose Bastos paper - "A 12-bit intrinsic accuracy high-speed CMOS DAC", IEEE J. Solid State Circuits, vol. 33, no. 12, pp.1959 -1969 1998.
The some issues about latch desining for high speed You can find at the Borremans, Van den Bosch papers (for the last 20 years ESAT-MICAS group at Leuven produce many valuable works about DACs).

Also Behzad Razavi's book "Principles of Data Conversion System Design" could be useful for You. If you want to increase matching between Yours current sources You can applies a degeneration which improves matching by 1+gmR factor (recomended for DSM technologies with relatively high transconductance factor).

 

[edinburghtech]

Current matching should correspond at least to the required resolution, i.e. 1/2ⁿ or better. In your case 1/2¹⁰ ≈ 1‰ or less. G00GLE for the corresponding literature.


By simulation of either all input voltages bit by bit, or using 2 or more different input voltages separated by 1LSB each, at several 2ⁿ borders. e.g. digital input values 0, 1, 2, ... 15, 16, 31, 32, 63, 64, 127, 128 ... 1022, 1023 bit, and checking the output voltages' accuracy.

Hi Erikl

Thanks for the reply. Since my required resolution is 10bits, as you mentioned the matching should be atleast 1/2^10 = 1/1024 = 0.1 % right? I am unable to find corresponding literature in this area.
I will try to check each and every output point as you mentioned. Also could you please help me how to do montecarlo analysis to verify matching?

Thanks
Tech

- - - Updated - - -

You need to use any mismatch model (e.g. Pelgrom model which is simplest but works only for saturated, strong inversion devices and overestimating current source mismatch). The methodology for estimate a current source dimensions you can find in Jose Bastos paper - "A 12-bit intrinsic accuracy high-speed CMOS DAC", IEEE J. Solid State Circuits, vol. 33, no. 12, pp.1959 -1969 1998.
The some issues about latch desining for high speed You can find at the Borremans, Van den Bosch papers (for the last 20 years ESAT-MICAS group at Leuven produce many valuable works about DACs).

Also Behzad Razavi's book "Principles of Data Conversion System Design" could be useful for You. If you want to increase matching between Yours current sources You can applies a degeneration which improves matching by 1+gmR factor (recomended for DSM technologies with relatively high transconductance factor).

Hi Dominik

Thanks for the reply. I find the Bastos paper very useful. Unable to get the Borremans paper. Please share if you have.
I will read about degeneration for matching and get back.

Regards
Tech

 

[erikl]

Since my required resolution is 10bits, as you mentioned the matching should be atleast 1/2^10 = 1/1024 = 0.1 % right?

Yes, correct!

I am unable to find corresponding literature in this area.

May be because this is self-evident: your tool's accuracy has to be at least as good - preferably better - as the results' accuracy you expect from it, isn't it? Think of measuring length with a foot rule!

Also could you please help me how to do montecarlo analysis to verify matching?

See the MC tutorial linked to in this thread.

Slainte, edinburghtech!