Voltage controlled current source

[[email protected]]

Hi,

I'm designing a PCB that needs to provide several bias currents to a
custom IC. The receiving end in the IC are basic current mirrors, i.e.
the input stages are diode-connected MOSFETs.

There are several constant currents that I need to source, ranging
from 1uA to 100uA. I plan to use LM234's with trimpots for these. What
is your opinion on using them for such small currents?

There is also a current that needs to be voltage controlled, so I can
change its value using a bench power supply, instead of tuning a
potentiometer. The current range from 100uA to 500uA. I may also
connect it to a waveform generator someday (a sub-MHz frequency). I'm
planning to use a opamp driving a MOSFET for this. The circuit is
similar to this one http://www.edn.com/contents/images/233726f2.pdf
I would like to use LT1801 for the opamp because it is easy to use
(single-supply, rail-to-rail), and seems to have decent performance.
My question is what MOSFET or JFET should I use? I see thousands of
choices on Digikey. Which one should I pick to have the best accuracy
for my application?

Thanks a lot,

Scott

 

[Joerg]

Hi,

I'm designing a PCB that needs to provide several bias currents to a
custom IC. The receiving end in the IC are basic current mirrors, i.e.
the input stages are diode-connected MOSFETs.

There are several constant currents that I need to source, ranging
from 1uA to 100uA. I plan to use LM234's with trimpots for these. What
is your opinion on using them for such small currents?

Assuming you mean the LM324, keep in mind that its input current can
exceed 100nA so the lower ranges might not be accurate enough.

There is also a current that needs to be voltage controlled, so I can
change its value using a bench power supply, instead of tuning a
potentiometer. The current range from 100uA to 500uA. I may also
connect it to a waveform generator someday (a sub-MHz frequency). I'm
planning to use a opamp driving a MOSFET for this. The circuit is
similar to this one http://www.edn.com/contents/images/233726f2.pdf
I would like to use LT1801 for the opamp because it is easy to use
(single-supply, rail-to-rail), and seems to have decent performance.
My question is what MOSFET or JFET should I use? I see thousands of
choices on Digikey. Which one should I pick to have the best accuracy
for my application?

P-channel? Any old small FET would do, depending on how much compliance
range you need. My usual standby is the BSS84. For the opamp I usually
use the LM324 as well but with a resistive divider because it can't get
close enough to the positive rail.

 

[Winfield]

I'm designing a PCB that needs to provide several bias currents to
a custom IC. The receiving end in the IC are basic current mirrors,
i.e. the input stages are diode-connected MOSFETs. [ snip ]

There is also a current that needs to be voltage controlled, so I
can change its value using a bench power supply, instead of tuning
a potentiometer. The current range from 100uA to 500uA. I may also
connect it to a waveform generator someday (a sub-MHz frequency).
I'm planning to use a opamp driving a MOSFET for this. The circuit
is similar to this - http://www.edn.com/contents/images/233726f2.pdf

P-channel? Any old small FET would do, depending on how much compliance
range you need. My usual standby is the BSS84. For the opamp I usually
use the LM324 as well but with a resistive divider because it can't get
close enough to the positive rail.

"Small" mosfet or jfet is the operant word, because
ordinary mosfets, like the one in the EDN article,
are "large" in the sense that they can conduct amps
of current, etc. They have a high gate capacitance,
Ciss, crippling their performance at low currents.
The OP can use a medium-sized mosfet, like the bss84,
with Ciss = 30pF (not that low), but for 0.1 to 0.5mA
currents he'll likely be better off using some common
small PNP transistor, like the 2n3906, instead.

 

[Joerg]

I'm designing a PCB that needs to provide several bias currents to
a custom IC. The receiving end in the IC are basic current mirrors,
i.e. the input stages are diode-connected MOSFETs. [ snip ]
There is also a current that needs to be voltage controlled, so I
can change its value using a bench power supply, instead of tuning
a potentiometer. The current range from 100uA to 500uA. I may also
connect it to a waveform generator someday (a sub-MHz frequency).
I'm planning to use a opamp driving a MOSFET for this. The circuit
is similar to this - http://www.edn.com/contents/images/233726f2.pdf

P-channel? Any old small FET would do, depending on how much compliance
range you need. My usual standby is the BSS84. For the opamp I usually
use the LM324 as well but with a resistive divider because it can't get
close enough to the positive rail.

"Small" mosfet or jfet is the operant word, because
ordinary mosfets, like the one in the EDN article,
are "large" in the sense that they can conduct amps
of current, etc. They have a high gate capacitance,
Ciss, crippling their performance at low currents.
The OP can use a medium-sized mosfet, like the bss84,
with Ciss = 30pF (not that low), but for 0.1 to 0.5mA
currents he'll likely be better off using some common
small PNP transistor, like the 2n3906, instead.

Yes, I go for BJT's as well. But for the usual reason: They cost a
couple of pennies less ;-)

BJTs do introduce a small error though. Their hfe often ain't that great
at low currents and the base current enters into the equation because it
also flows across the sense resistor.

 

[Scott]

Thank you all for your help. When you say you'd choose a BJT instead,
is it because of the smaller input capacitance? Or higher
transconductance? I thought since the opamp already has a large open-
loop gain, the gain of the transistor shouldn't matter so much. Am I
right?

My application needs <100nA accuracy for all the currents, therefore
the base current of the BJT may be too much.

For the LM324 (sorry for the typo), what would happen for small
currents? The datasheet specifies it for down to 1uA, but I can't find
any information on what would happen at the low end or beyond. Is it
thermal noise? Also can I have such accuracy for larger currents (say
100uA)? I don't really want it to be 99.9uA--100.1uA, just the
variation has to be smaller than 100nA.

Thanks again.

 

[Winfield]

Thank you all for your help. When you say you'd choose a BJT instead,
is it because of the smaller input capacitance? Or higher
transconductance? I thought since the opamp already has a large open-
loop gain, the gain of the transistor shouldn't matter so much. Am I
right?

I'd say yes, inside the feedback loop, the mosfet's
low transconductance isn't an issue. Actually, at
low currents the mosfet's transconductance/current
ratio improves dramatically, but it doesn't get as
good as a BJT, with gm/Ic = q/kT = 1/25mV at 25C.

My application needs <100nA accuracy for all the currents, therefore
the base current of the BJT may be too much.

Then consider useing a Darlington transistor, or wire
two ordinary ones in the Darlington configuration.

For the LM324 (sorry for the typo), what would happen for small
currents? The datasheet specifies it for down to 1uA, but I can't find
any information on what would happen at the low end or beyond. Is it
thermal noise?

The LM324 input bias current, which is the spec you
should be concerned with, can be as high as 250nA at
room temp, according to the datasheets by ON Semi,
Fairchild, TI, etc. NSC's LM324A datasheet says 100nA
but if you're serious about that parameter, I'd select
a better opamp.

Also can I have such accuracy for larger currents (say 100uA)?
I don't really want it to be 99.9uA--100.1uA, just the
variation has to be smaller than 100nA.

There are two current-source issues having to do with
speed, capacitance, and MOSFETs. The first is gate
capacitance, which will determine how fast the opamp
feedback loop can be, and how much capacitive gate
current shows up on the sense resistor, causing an
error at high frequencies. The second is the drain-
source output capacitance, which will make a current
source look like a voltage source at high frequencies.
A BJT has Cobo = 4.5pF max, 3pF typ above 5V. By
comparison, a bss84 mosfet, being a reasonably-small
part, has Coss = 10pF typ, which is about 3x higher.
Other common power MOSFETs will be much higher yet.

Consider, 10pF at 100kHz looks like 160k ohms, which
can be an issue if your goal is 100nA predictability.

So, as you can see, there are tradeoffs on each side
considering BJT vs MOSFET for low-current circuits.

 

[MooseFET]

Scott wrote: [...]

My application needs <100nA accuracy for all the currents, therefore
the base current of the BJT may be too much.

Then consider useing a Darlington transistor, or wire
two ordinary ones in the Darlington configuration.

Or:
-------------+--------+---
! !
!/e !
--------! PNP !
!/ !\ !/ ------> Repeat if needed
---! NPN -------! NPN
!\e !\e
! !
----+----------------+----

The EB voltage is the same as the single transistor and the HFE can be
really huge

 

[Scott]

Thank you (and MooseFET), I'll look into the circuits you suggested.

For my constant current sources, I'm actually using LM234 (3-terminal
adjustable current source), not LM324 after all. Sorry about all the
confusion! Any comments on these parts?

Thanks a log,

Scott

 

[Joerg]

Thank you (and MooseFET), I'll look into the circuits you suggested.

For my constant current sources, I'm actually using LM234 (3-terminal
adjustable current source), not LM324 after all. Sorry about all the
confusion! Any comments on these parts?

Keep in mind that they are temperature-dependent and that the inityial
accuracy is only 3%.

Thanks a log,

I just put three of those into the wood stove

 

[Joerg]

I'd say yes, inside the feedback loop, the mosfet's
low transconductance isn't an issue. Actually, at
low currents the mosfet's transconductance/current
ratio improves dramatically, but it doesn't get as
good as a BJT, with gm/Ic = q/kT = 1/25mV at 25C.


Then consider useing a Darlington transistor, or wire
two ordinary ones in the Darlington configuration.


The LM324 input bias current, which is the spec you
should be concerned with, can be as high as 250nA at
room temp, according to the datasheets by ON Semi,
Fairchild, TI, etc. NSC's LM324A datasheet says 100nA
but if you're serious about that parameter, I'd select
a better opamp.

Also, mind the 7mV or so offset voltage. This adds to the total error
and its impact will be larger for smaller voltages across the sense
resistor. With Scott's specs I'd pick a better one.