Temperature differentials across the board ARE a temperature mismatches. It can be characterized two ways. The currents in the mirrors are not the current that is being programmed, or the currents in the mirrors are not the same as each other. Which is more important? One way to get at this is to ask yourself which is worse - that the current, on average, in the mirror is within 10% of the programmed current but that there is only a 1% range between min and max, or that the current, on average, is within 1% of the programmed current but there is a 10% range between min and max.

With transistors off of the same reel, the difference that I have measured between parts in this circuit is extremely low - non-consequential for my purposes. So given no thermal differences in transistors, I am happy with both absolute and relative accuracy. My only concern is the different current between parts that are at different temperatures.

 

Why don't you tell us what your circuit is supposed to do. Sometimes, when we get the big picture, someone will come up with another way of doing the same thing, but without the problems.

The transistors set the output level of a string of current driver ICs. The drivers are spread out over an 18 inch high board and drive LEDs. The potential is that the boards may be unequally warmed by the sun or by convection within the cabinet. For now, the circuit layout is fixed, and I need to do the best I can to minimize the thermal issue. I am also open to thoughts as to how the circuit may be improved in the future, but don't want to get sidetracked with that now.

 

Another way to control temperature if you want everything the same, is to use an oven. This is more normally done for crystal circuits, but I see no reason in principle why you could not create something here.

If you are saying that different variable currents in different parts of the circuit will create localised heating effects that must be compensated for, then it is a different matter.

 

What current driver ICs do you use? Can you post a schematic diagram of what you have (and by schematic I mean the whole thing, not just a bunch of transistors hidden in the corner of a blank page with no reference to anything else), this is really very theoretic and I doubt we will get anywhere like that.

Couldn't you simply control voltage, and then through a resistor connect each of the current inputs?

 

Using .3 volts as your reference causes a condition where a 1.5C difference in temperature causes a 1% difference in current. Changing the reference voltage to 3 volts would result in 1% difference in current per 15C difference in temperature.

Changing the reference voltage will change the variation of current per degree of temperature.

 

With a set base voltage, I looked at the emitter voltage change with temperature. With the 3 different emitter load resistors listed, that change with temperature was very much the same, one to the other. Or in other words, the emitter voltage drift due to temperature did not change across that range of emitter load resistors.

WHY WON'T YOU PRESENT YOUR DATA?!?!?!

WHY WON"T YOU TELL US WHAT YOU NEED YOUR CIRCUIT TO ACHIEVE?!?!?!

Of course the base-emitter voltage changes if the temperature changes and you have a fixed base voltage. So what? You WANT it to change! You NEED it to change!

You don't CARE about whether or not the emitter voltage changes, you care about whether or not the collector current changes; if the base-emitter voltage DOESN'T change with temperature, then the collector current will change drastically.

At the same collector current, you will have a larger drop across the emitter resistor. With the same change in emitter voltage, you have a smaller fractional change in the voltage across the emitter resistor and, hence, a smaller fractional change in the emitter current (which is the same as the fractional change in the collector current).

Let's run some numbers. Let's use a target current of 100μA. Let's assume that the base-emitter voltage of the programming resistor is 500mV at this collector current and that the transistors are perfectly matched. Now let's say that the temperature of one of the mirror transistors is 9°C higher and another is 9°C lower.

If there are no ballast resistors, then the base-emitter voltage of all three resistors would be held at 500mV, but this would result in the collector current of the first mirror being about 200μA and the collector current of the second being only about 50μA.

To keep the collector current 100μA in the first mirror, we need to reduce the base-emitter voltage to 482mV and to keep it 100μA in the second we need to increase the Vbe to 518mV.

Now let's put in a 100Ω ballast resistor. The voltage drop across this would be 10mV so the base voltage of the programming resistor will be 510mV. Not much of a change, right? But now the current in the first mirror will only increase to 160μA and the mirror in the second will only decrease to about 60μA. Not a huge improvement, but what can you expect for a measly 10mV across the ballast resistor?

Now let's put in a 1kΩ ballast resistor. The ballast resistor now drops 100mV and the programming voltage must be increased to 600mV. Now the collector current in the first mirror only increases to 115μA and only drops to 86μA.

Now let's put in a 10kΩ ballast resistor. The ballast resistor now drops 1V and the programming voltage must be increased to 1500mV. Now the collector current in the first mirror only increases to 102μA and only drops to 98μA.

Note that, up to now, we have neglected the change in resistance due to temperature. Let's assume that it is 100ppm/K. For a roughly 10°C change in temperature, that would be a 1000ppm change in resistance, or a 1mV change in voltage. That will actually reduce the change in current, though not by a huge amout.

But what if we used a resistor with a 200ppm/K coefficient and sized it so that there was a 10V drop across it at the desired current. In our case, that would mean a 100kΩ resistor. Now the voltage across the resistor increases by 2mV/K if the current remains constant at 100μA while the voltage across the base emitter junction decreases by that same amount. Combined, this means that the voltage at the base remains constant as the temperature changes.

Now, this only works at at one particular current and it doesn't hold perfectly as the temperature changes because neither component changes perfectly linearly. But the residual changes are greatly reduced.

 

Bill, he already has ballast resistors. Each one is 470Ω. HE can reduce the thermal-gradient induced variation in current from one transistor to another, and also loosen up the Vbe matching requirements, by making this resistor value larger (but still all equal). But that means that the control voltage range (now 30mV to 600mV) must increase by the same amount. For example, if we make all the resistors 4.7k, the control voltage range must go from 300mV to 6V. This might be OK, if the current sinks still have adequate head room when their bases are all at ≈6.7V.

I want to see the big picture. What exactly are these current sinks controlling? Maybe there is another way of skinning that cat.

 

I know he has 470kΩ resistors in place, and I suspect that those are more than adequate for his needs. But I can't tell because he won't tell us what his needs ARE.

Until he can say how much variation he can tolerate, he doesn't stand much of a chance of finding a way to keep the variation tolerable.

He is also focusing on the wrong thing. He is fixated on the fact that the emitter voltage changes with temperature. But everything he says indicated that what he should be worrying about is whether the collector current changes with temperature.