That's a pretty nice part, but I have all sorts of specialized requirements. It's driven by a chip I don't control, that expects two series P-FETs that it operates in a particular sequence.
The overall product chains two SMPS, and manages n>=94% input-to-output. (I'm about to improve on that slightly with the beefed-up high-side FETs I'm protecting with this overload protection circuit). Lots of Linear Tech chips.
The switching FETs chosen are highly optimized, after extensive search, to be the very best currently available for this exact spec set. Cool beans, LTC.
No. The part he's using is two adjacent dice from the same wafer. Offset voltage is the same 1-2 mV as a monolithic dual. But, a monolithic dual would have bias requirements for the substrate, lest the collectors swap current. Keeping a monolithic pair isolated is a major cost, not required here.
and on page 3, maximum Vos is listed as 2 mV (same Ic, 5V collector bias)
So, it IS guaranteed, isn't it? There's some scaling to be done, because the emitter currents don't match, but that's an easy design-time calculation. It won't be a unit-to-unit variation, just an operating-point offset with some predictable size and temperature coefficient.
The problem with non-monolithic dual transistors is that they are usually not well coupled thermally. So if they have different power dissipations, chip temps will be different and then you get Vbe offsets.
formatting link
--
John Larkin Highland Technology, Inc
lunatic fringe electronics
Vos of two random 2n3906's from possibly different makers (or even centuries) are not guaranteed. That's why, rather than risk that, I chose a matched pair made from adjacent dice and was willing to pay $0.13 extra.
I vetted Digikey's portfolio before choosing the DMMT3906W (mostly because there were other parts in the same pkg that could replace it if needed). Multiple-sources, and all that good stuff.
It would be nice if there were monolithic PNP pairs; I didn't see any.
Yeah, or if the thermal gradients are high. But, that doesn't have to happen. Just keep the bias current low, so they dont self-heat.
The problem with monolithic dual transistors, is that they have to be isolated by making a fourth layer in the diffusion, which has to be reverse-biased relative to the collector, which requires a seventh bias pin. Or, if you pay enough, you can get dielectric-isolated (oxide and epitaxial collectors, or silicon-on-sapphire).
When they were common (I'm looking at data on the old CA3084), the monolithic multiple PNP transistors had Vce(max) of -40V, Ic(max) 10 mA, beta (min@100 uA) of 15, and gain dropped at about 4 MHz. Pin 4 had to be connected 'more negative than any base' of the transistors.
The 2N3906, though, does the same Vce(max) of -40V, Ic(max) 200 mA, beta(min@100 uA) of 60, gain drops at about 250 MHz. It's just a much better transistor, as well as being better stocked nowadays. That extra diffusion layer makes the monolithics a low performance option.
That's not quite right. His sense resistor gives him 1A per 10mV, kinda hef ty. The cut-in voltage on Q1A is therefore a big unknown. The Vbe cancelati on on Q3-Q4 is a crude attempt, there's no bandwidth to his feedback with t hat set-up and Q1A being driven common emitter.
efty. The cut-in voltage on Q1A is therefore a big unknown. The Vbe cancela tion on Q3-Q4 is a crude attempt, there's no bandwidth to his feedback with that set-up and Q1A being driven common emitter.
I thought I'd already amply explained that it's not a precision circuit, so none of that matters. +/- 20% is fine. A crude, first-order Vbe cancellation is worlds better than no Vbe compensation at all, and more than good enough.
The requirement for an external, logic-level input has been dropped, which might make a R-R op amp or comparator cost or complexity competitive again, that might be worth revisiting.
But either one of those also needs hysteresis, a reference for the threshold, and an output level-translation to produce the required open (normal) / 10V levels.
The LM7301 is a nice candidate. It's $1.65 in 100's, pulls 600uA(typ) of battery, and, of the functions I need, replaces essentially only a matched pair that costs $0.20 in 100's and draws 200uA.
Those are the considerations at hand.
My job is to protect the FETs in the most practical, expedient, economical way, not throw expensive parts at something that doesn't need it.
Speaking of protecting fets, I sometimes use an ADC+uP to compute fet power dissipation, add in the heatsink temp, apply a time constant, and then have a realtime estimate of junction temperature. That pushes fets to their maximum safe output.
Here's an analog version of the power computer:
formatting link
(missing the final heatsink temp and time constant)
That PNP power dissipation is about as low as I could get it.
--
John Larkin Highland Technology, Inc
lunatic fringe electronics
On Friday, January 1, 2016 at 12:15:16 PM UTC-5, snipped-for-privacy@yahoo.com wrote :
e (that's 1000-10,000 x) than a monolithic part.
Offset voltage is
have bias
eeping a monolithic
hefty. The cut-in voltage on Q1A is therefore a big unknown. The Vbe cance lation on Q3-Q4 is a crude attempt, there's no bandwidth to his feedback wi th that set-up and Q1A being driven common emitter.
What volume? As our TI FAE says, you can name the price if I can name the volume. For R-R opamps, I use mostly Microchip MCP68xx (the exact one depends on the GBW/noise/Iq needed). With enough volume, quads are easily in the "few cents" and opamp range.
On Friday, January 1, 2016 at 12:15:16 PM UTC-5, snipped-for-privacy@yahoo.com wrote :
e (that's 1000-10,000 x) than a monolithic part.
Offset voltage is
have bias
eeping a monolithic
hefty. The cut-in voltage on Q1A is therefore a big unknown. The Vbe cance lation on Q3-Q4 is a crude attempt, there's no bandwidth to his feedback wi th that set-up and Q1A being driven common emitter.
d
l
Well you don't have to sacrifice precision for cost. The circuit below is p retty cheap and available from Mouser. The 10V shutdown is a little sloppy but should fit the bill. This circuit can be adapted for way higher voltage s than you're dealing with. The output drive biasing is based on scratchpad calcs using nominal Vbe's, diode drops and current gains. It should be run through SPICE environmentals for best compromise or even correct operation . Then you might want to add some terminal protection or your controller in the event something blows through.
Please view in a fixed-width font such as Courier.
On Friday, January 1, 2016 at 7:35:59 PM UTC-5, snipped-for-privacy@gmail.com wr ote:
te:
rse (that's 1000-10,000 x) than a monolithic part.
Offset voltage is
d have bias
Keeping a monolithic
da hefty. The cut-in voltage on Q1A is therefore a big unknown. The Vbe can celation on Q3-Q4 is a crude attempt, there's no bandwidth to his feedback with that set-up and Q1A being driven common emitter.
,
of
hed
cal
pretty cheap and available from Mouser. The 10V shutdown is a little slopp y but should fit the bill. This circuit can be adapted for way higher volta ges than you're dealing with. The output drive biasing is based on scratchp ad calcs using nominal Vbe's, diode drops and current gains. It should be r un through SPICE environmentals for best compromise or even correct operati on. Then you might want to add some terminal protection or your controller in the event something blows through.
You've found two very reasonably-priced ICs. Ignoring Rs, I count 11 parts and roughly $1.15 off the showroom floor. Nicely done Fred.
FIGURE 1. (I've added a resistor to make the 350mV reference from +5v)
ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here.
All logos and trade names are the property of their respective owners.