Saving a resistor on an NPN transistor current source (sink) when driving LEDs

Question

The typical circuit to drive an LED from a low-power microcontroller digital output pin, using a transistor, might look something like this (assume a 20mA/2V LED):

^{simulate this circuit – Schematic created using CircuitLab}

Now, of course individual transistors have current gain values that vary widely from one to the next. However, when building a circuit myself, I know the exact individual transistor I'm going to use. Therefore, I could measure the current gain of that transistor when it's not in saturation, with a circuit like this, for various values of R1:

^{simulate this circuit}

And then (in my example), I could keep trying different values of R1 from my stock of resistors until the output current is exactly 20 mA:

^{simulate this circuit}

And then use that in the LED circuit, now omitting the current-limiting resistor and letting the transistor do that instead:

^{simulate this circuit}

Of course if the transistor were swapped for a different one, the process needs to be repeated.

This saves one (potentially high-power) resistor per LED, by simply choosing an appropriately-measured resistor in series with the transistor base.

Is there any particular problem with this method?

Answer 1

^{simulate this circuit – Schematic created using CircuitLab}

This would be a better circuit. It should work fine also if SW1 is replaced actually by 5V CMOS output. There is no need to rely on calibrating each transistor or the stability of beta.

Answer 2

This is a bad idea. Beta can change drastically with die temperature:

Also there can be significant change with age of the device, especially at high temperature. Your die may be relatively hot if you're dissipating all the power in the transistor.

Any change in your supply voltage will also cause relatively large fluctuations in the LED current.

Bottom line is that you could get it to work for a time at one operating point, but it's bad design. It's not that bad to include the resistor and remove the strong dependence on an exact value of beta.