OT: SOT-523 PNP Question

"John Larkin" wrote in message news: snipped-for-privacy@4ax.com...

What's good, compliance or accuracy? If you're looking for the latter, you probably are better off with a dumb resistor, or something active to mirror the current more precisely than transistors can anyway. If you need compliance but don't care if it's off by a factor of two, go right ahead; best part about discrete is you don't even have to worry about lateral PNPs.

Nice thing about current mirrors (with little or no degeneration), they saturate very near the rail. A one-Vbe-emitter-drop type can't pull that.

Tim

At sufficiently low current, I agree. At higher current, especially with small devices, the output device will run away because the thermal coupling isn't good enough to make the V_BEs track. For instance, Figure 13 of the BCV62 datasheet is worth thinking about--it shows the output device running away at only 15 mW of dissipation. Monolithics are much better.

Cheers

Phil Hobbs

Figs 14 and 15 make no sense. There's no base current for TR1.

But it looks so much nicer without that ugly line right across the bottom. ;)

Cheers

Phil Hobbs

People who call a transistor TR deserve to have them not work.

Here's an NEC dual transistor with, as I recall, 50 mW dissipated on one chip.

Not very isothermal.

You _can_ cascode discretes just like you do monolithically.

Here's a past comparison of current mirror types...

...Jim Thompson

Some "artist" in the datasheet department left off the common ground from both drawings. When I was at Motorola I had many problems making sure that datasheets that went out the door were accurate. ...Jim Thompson

You may as well include an opamp and get some serious accuracy and temperature stability, and an easily programmable mirror ratio. Four parts, or even three if you get a little tricky.

Does anybody make an integrated opamp-based current mirror?

I have many... on custom chips... but it doesn't take an OpAmp to do it, just a diff-pair in CMOS is quite sufficient.

But, in most cases, CMOS doesn't need such frivolity, current mirrors can be made quite accurately with cascoded devices. ...Jim Thompson

Other than type "BARE", none of these have significant temperature effects when implemented monolithically...

And even "BARE" isn't too bad. ...Jim Thompson

So why don't mine at over 20mA and 10V? :)

Rth(j-a) = 500 K/W is a bit unfair relative to a TO-92 in still air (~100C/W), but it's the difference that matters. With Rth(j-j) large (they're adjacent, but not bonded... probably in the 1000s C/W), it shouldn't take much. No, not tested beyond 40C or so.

Besides the funny drawings, it also doesn't specify if this is maximum over temperature, or just at room temperature. The former would seem more reasonable. There's no corresponding datapoint in the tables to compare to, where conditions might be listed.

Also, what happened to their hFE plots?

As for changing my opinion of discrete current mirrors, this datasheet has only convinced me to never buy BCV62s. :o)

Tim

Sure, and that helps the tracking a lot, but it also loses the advantage in output swing that Tim was talking about.

Cheers

Phil Hobbs

it depends on the axis you use to flip it.

rotate 180 degrees about the line from pin 1 centre through the centre between pins 2 and 3,

Because they're bigger devices, silly. The BCV62 has a specified theta_JA of 500 K/W. Given a theta number, it's really easy to calculate the thermal stability limit analytically.

Nah, I use them a fair amount, but only down below 500 uA. They're also useful for making temperature compensated current sources, for instance--think of the diode-connected device as a way of temperature compensating the V_BE of the current source device.

Cheers

Phil Hobbs

It would have to be upside down and rotated. It actually does fit well enough that you don't need bits of wire -- but on the parts we had you couldn't reform the leads without them breaking off, so it's a highly questionable thing to do with a production board. We just did it to verify the rest of the board before giving the layout a hard time and doing it over.

Your adapter board idea is probably better, if they fit, or your rotate- and-wire idea. It's a hell of a lot of work, though.

Try it some time... in CMOS it can actually _improve_ the swing range before the current decreases (patent pending :) ...Jim Thompson

I don't read that set of curves (strange as they are) as implying run-away. When I have a moment I'll run some numbers. ...Jim Thompson

I determined that the transistor wasn't needed, it was overkill, so suck off transistor, replace with a resistor... lots of labor but, fortunately, it's on main-land China. ...Jim Thompson

'Tain't hard to calculate. At constant V_BE, a junction temperature rise of 1 degree produces a current increase of 9%. Assuming a low-Z collector load, device dissipation increases by 9% as well. Instability will occur if that 9% causes a temperature change of another degree, i.e.

The general derivation isn't much harder, even with a more complicated load.

If you bias the output transistor near its maximum-power point, you can even use this effect as a temperature controller:

Cheers

Phil Hobbs