LED forward voltage drop with temperature

Depends. It's rare to see a diode data sheet that shows the VI curve at different temperatures, much less at even rated max current.

I've deliberately run some small schottky diodes at the zero TC point, which was about 15 mA.

Big power diodes sometimes show that point on the data sheet. That counters the conventional wisdom about thermal runaway current hogging of diodes in parallel. Some of the big power diode bricks are diodes in parallel.

Here's one, zero TC at 200 amps, well past rated current in this case.

It would be interesting to test an LED. People make cheap flashlights with a battery and an LED, so some must be substantially ohmic.

Thank you Gerhard. I will save your pdf for my future reference.

Thanks, Phil. It looks like LTSpice is giving me what I suspected was incorrect info. Spice is showing about 4mV/C positive voltage with temperature at 2mA. I need to make bench measurements and not depend on spice.

Hi John,

Are you sure the current flowing through the LED is constant ? i.e. temperature independent. AFAIK current should be kept constant to evaluate voltage drift of Vf over temperature.

Some basic math on your circuit would be nice to be explained. Please.

Look at John's circuit, the LED voltage is matched by the V_be and resittor, for it to compensate both V_be and LED must move in the same direction.

Thanks, Jasen. Yes, I studied his circuit and understand how it works.

ode

So I think it's green LED's that 'go the wrong way'.

And I'm assuming the forward voltage changes too.

George H.

The LED current is set by the voltage drop across the 2K resistor. The current TC is what I measured for the whole circuit.

I also measured power supply sensitivity, which isn't bad at all. Variation in power supply voltage directly (actually worse than directly) changes the LED current. A tweak might null out power supply sensitivity, but I had a good supply in my application.

There's no serious math here. There can't be without knowing a lot more about the LEDs than is available. It was easier to build it and test it.

You could Spice it to see the basic functionality, but I wouldn't trust a simulation to predict TC.

It could certainly be better, with a thinfilm emitter resistor and some more tweaking. It was good enough for my product so I moved on. But there are much better circuits if one wants a super stable current source. This was just sort of fun, and it glows in the dark.

Transistor self-heating would be a problem if one wanted serious stability. Base current too.

Right. There are all sorts of LED chemistry and fabs around. I wouldn't trust any of the LT Spice LED models very far.

My green LED really seemed to be an oddball. I don't think my transistor was oscillating, but that's possible. I really like the color of the orange Osram, so I used that one.

Any "basic math" you do on that circuit will result in a transcendental equation that can't be solved in closed-form, anyway so you'll have to resort to numerical methods in any event.

The only way to solve just the diode-resistor circuit exactly for the R-D junction voltage in "closed form" is with the Lambert W function

if the base current assumed to be 0 a simple model would be:

= Is log (Vbe/Vt) + I*R_e.

The diode and transistor Is's are non-linear functions of temperature so unless they cancel neatly like in an a differential pair with matched transistors leaving only the secondary temperature dependence on Vt, without further data about Is this equation doesn't tell you much about real-world temperature stability.

Ooops, I meant:

It's about the best discrete current-source you can build for the money. To build one using discretes that improves on it substantially would require a lot more parts and probably matched transistors.

That's wavelength stability, not voltage drop.

RL

LTSpice model may just assume a constant value for the LED Is parameter, and this works well enough for many purposes. but it's a strong function of temperature in reality like all junction semiconductor devices.

Doing math computerized or otherwise with the diode or transistor equations assuming Is is constant expecting to get meaningful tempco results is hopeless.

You could do better if you allow more voltage drop in the emitter resistor.

ts.

diode

ts

Right. (Sorry just my first hit in google) There are two 'pieces' to the forward voltage drop. One from the current ... the diode equation (and any real resistances) And then from the bandgap of the LED. (Phil H. posted a nice article that covered both.) (And of course the current heats the junction so the two interact strongly.)

George H. reading the article Phil linked to... I'm confused by some of it. (fig 2 shows led wavelength getting shorter at higher led currents?)

Nah, you just apply a suitable sleazy analytic approximation and press on. Binomial expansion, perturbation, asymptotic analysis, Taylor series, no worries.

Tractable analytic approximations are much more useful than some opaque exact result, and way more useful than a stack of simulations--you can optimize analytically, for instance, or set good reliable upper and lower bounds on performance.

Cheers

Phil Hobbs

Middlebrook developed a whole framework for obtaining tractable results for on-paper feedback amplifier design; once you get beyond a couple transistors the exact results become hopeless nobody could make sense of them if written out.

Besides the EET I can't say I've seen it used much in practice. Most designers except those who work at semiconductor cos don't design their own multi-stage discrete feedback amplifiers no mo'.

Wonder how a dual-gate MOSFET would work and hang an LED with opposite tempco off each gate.

I don't think P-channel duals are easy to come by no mo' and dual-gate FETs aren't particularly cheap in general though.

There's an example of a discrete band-gap made with the transistors in e.g. a CA3046 array, for high supply voltages perhaps, there was probably a time when those were cheap and available but most multi-transistor chip arrays seem out of production, now. My ~70 MHz Kikuisi analog scope circa 1982 uses two CA3046 in the input circuit.