Low power , low voltage Current Source

[raducrv]

I need to design a l-p l-v current source , Vin=3->5,5 V ; Iout=1µA +/- 20% ; T=-40 degree(C)...90 degree(C)

 

[keanhong2]

T=-40 degree
(C)...90 degree(C).i not understand ?

 

[erikl]

Celsius, I'd guess ;-)

 

[raducrv]

yes , celsius

 

[erikl]

I need to design a l-p l-v current source , Vin=3->5,5 V ; Iout=1µA +/- 20% ; T=-40 degree(C)...90 degree(C)

Use something like this:

 

[raducrv]

like what ?

---------- Post added at 20:17 ---------- Previous post was at 20:11 ----------

ok thanks , i'll try those . thanks .
any further information is welcome

 

[erikl]

... any further information is welcome

Not easy, without knowing your process tech.

 

[raducrv]

by process you refer , using tranzistors (pmos,nmos), i have to do the spice model (ltspice)

 

[erikl]

Ok, then you can start with the suggestions given above, and analyze with your spice models!

 

[raducrv]

thanks a lot

 

[leo_o2]

It is hard to have +/-20% current accuracy in standard process. Trimming circuit is needed.

 

[raducrv]

that's the ideea , because it was much more easier to obtain a constant current source , so our teacher had the stupid ideea

 

[erikl]

It is hard to have +/-20% current accuracy in standard process.

True. But it can be done, at least for currents in the several nA range. See here:

... from this paper:

[P73] G. De Vita, G. Iannaccone, A 109 nW, 44 ppm/°C CMOS Current Reference with Low Sensitivity to Process Variations, IEEE International Symposium on Circuits and Systems, 2007 (ISCAS 2007), pp. 3804-3807, New Orleans, USA, 27-30 May 2007

... available from here [P73].

 

[Sonnenmann]

the supply voltage is variations are not the problem, the structure of a simplke ptat+ctat source should be able to handle this. The variations of technologie parameters are the problem.
If you take this source and simulate with worst power worst speed parameters, you will se a variation of +/- 20% ,too.
Current sources are still depending on a resistance and this could not realize so accurate.

 

[leo_o2]

Result from most of papers can't be trusted for mass production. +/-20% current accuracy on chip without trimming for mass production is hard to achieve. Usually resistor will take >+/-30% variation.

 

[erikl]

... Usually resistor will take >+/-30% variation.

Hi Leo,
didn't you read this paper from above? A CC source can be designed without resistor, only for an individual, process-dependent (µn/µp ratio dependent) current, however. But this current can be multiplied arbitrarily, of course.

 

[FvM]

You showed examples of a current source insensitive to VT variations. But it will be still affected by P? At best, the variation is lower than with usual resistors, I assume. Is the originally requested 20% initial accuracy feasible without trimming?

 

[erikl]

... Is the originally requested 20% initial accuracy feasible without trimming?

Actually, I don't know, FvM .
The current dispersion from one single wafer - measured on 20 samples only, admittedly not much for good statistics - showed a 1σ std. deviation of about 2% (s. Fig. 6 of the a.m. paper), so a <20% (in-)accuracy seems feasible without trimming, IMHO.

 

[raducrv]

this is the circuit , MOS Diode M6 as load . The teacher told us to do also a start up circuit , something by applyng a current bethween M2 and M5

---------- Post added at 15:37 ---------- Previous post was at 15:34 ----------

I(R)={2*[1-sqrt(1/k)]^2 }/ (beta*R^2) , where k is M2 multipicity and beta=50uA/V^2 ,

 

[leo_o2]

Thanks for reminding, erikl.
However, it is worth raising doubt for that the circuit will achieve 20% current accuracy.
I check the equation. There is some items of Vt, so other items left can be looked as a equivalent resistance. It's still hard to achieve accurate equivalent resistor. It is related to k5 that is process dependent. k5 effect is not described in the paper.
In additional, Vds5-Vds3 will be very small. It will be influenced much by mismatching of M1 and M2. Other MOS vth mismatch might lead to serious variation.

Hi Leo,
didn't you read this paper from above? A CC source can be designed without resistor, only for an individual, process-dependent (µn/µp ratio dependent) current, however. But this current can be multiplied arbitrarily, of course.