Diode-Connected Transistors

Error 404, file not found.

=A0 =A0 ...Jim Thompson

=A0 =A0| =A0 =A0mens =A0 =A0 |

=A0 | =A0 =A0 et =A0 =A0 =A0|

=A0|

=A0 =A0 =A0 |

It's there now :)

Grant.

Thanks!

John L. was going to check BJT diode performance back on the thread:

Hope you find time for this one John?

I though Jim T. was going to do something with these too, now he has.

Might've been a different thread that I can't find?

Time flies differently when one is having all their teeth ripped out :)

Grant.

No, but because of popular demand, I'll try to do it tomorrow. Spice sims aren't especially surprising at high currents. And not likely to be very accurate at sub-pA currents.

I did measure capacitance, BFT25 both ways. The numbers were something like 0.5 and 1 pF.

John

Worked for me...

Only because most discrete-device manufacturers don't characterize their products at those currents... if they characterize at all. Mostly they just leave out those terms.

Win Hill complains regularly about poor MOSFET models. Those manufacturers respond by supplying add-on elements in a subcircuit. They _should_have characterized them as HSpice_48/PSpice_7/BSIM3v3. I have demonstrated here, in the past, a size-scaled I/C MOSFET so that it behaves much like a 2N7000, _and_ has appropriate gate-charge behavior.

John, Please tell us how you are measuring capacitance in Spice.

I have a quite precise way that separates real and imaginary parts. I use this primarily to characterize amplifier inputs when tuned circuits are involved. Spice aficionados contact me via E-mail and I'll share.

Spice is like a sports car... only as good as the person behind the wheel. ...Jim Thompson

Thanks Jim... I'm surprised just *how much* better the BE_C_FLOAT case is than the others.

...although BC_EB_SHORT is best for low currents... interesting...

Caution: I didn't characterize transient behavior. Floating elements tend to cause "funny" storage times.

And "best" is in the eye of the beholder :) ...Jim Thompson

This is interesting as far as it goes, but I have some questions...

What type of model is this - Gummel-Poon, G-P with extensions (e.g. Kull-Nagel modeling of quasisaturation, this is common), Mextram, or ?? I think it's not likely to be HiCUM or VBIC95, which never found widespread use, most likely it's GP It's not as accurate as Mextram, especially at high collector currents, but is pretty much the lingua franca of publicly available simulators.

Where did the model come from, and how accurate is it, especially at high current? Generating accurate models, especially for high collector currents, is rather tricky. The model guys I work with have spent a long time learning how to do it well.

How well do you trust the reverse parameters in the model? These are often sloppily done, or simply left as model defaults.

Does the model include self-heating effects? These have a very strong influence on the high-current behavior. Standard TO-92 can be in the hundreds of degrees per Watt, this effect can't become important! Some of the smaller transistors I've worked with in dielectrically-isolated monolithic processes have thermal resistances of several *thousand* degrees per Watt!

I recall a paper from many years ago that did this sort of comparison, including (possibly) reverse-recovery time. The conclusion was that the shorted base-collector gave the best diode overall. I spent half an hour rifling through all my files this morning but couldn't find it, and I had no luck on the IEEE web site. It's frustrating as I can clearly visualize the pages; it might have been IEEE or IRE, I don't believe it was IEE. If I can find it I'll post the reference. It was probably done in the 1960s or early 1970s.

steve

• Model Generated by MODPEX *

• All Rights Reserved *
• UNPUBLISHED LICENSED SOFTWARE *
• Contains Proprietary Information *
• Which is The Property of *
• SYMMETRY OR ITS LICENSORS *

• by Symmetry License Agreement *

• Model generated on Aug 7, 01
• MODEL FORMAT: PSpice

This is an ON-Semiconductor-supplied library

No quasi-saturation.... only I/C houses seem to supply that information.

Who knows? I didn't do the modeling myself.

Overnight I had a request. I simply spit it thru the mechanism. I only believe (in my I/C world) in BE_CB_Short

Not much! I spent much of my early years writing software to extract NE, ISE, etc., kind of data

Of course not. Though one could create a subcircuit, IF you were so inclined.

Personally I rarely EVER use on-chip diodes for other than signal and level-shifting activities.

Having just been thru a painful episode with IBIS models I would gleefully donate the powder necessary to blow all IEEE committees to hell ;-) ...Jim Thompson

Indeed. We fought the same (IBIS) battle a few years ago, with inconclusive results after a lot of effort expended. It was kind of funny/sad really, the customer requesting this was worried about all the wrong things, and somehow believe that having a model they could simulate would make up for a rather large case of cluelessness. They were all wrapped up in the high-speed digital interconnect between their chip and ours, and not nearly concerned enough about the very high-quality signal path required between our chip's output and the real world...

It has the _look_ of being extracted, but with that IS and VAF I'd be mighty suspicious. RE looks pretty unrealistic at 100 micro-ohms.

Yes, it's difficult to extract, and I don't think any publicly available simulators support those kinds of models.

Ditto, I wouldn't trust anything else.

That's an interesting area, actually. The standard subcircuit (used in the Mextram model) is a parallel RC and current source. The current is based on Ic*Vce, R gives the static self-heating, and C makes the time-constant, which can be low microseconds for small dielectrically-isolated transistors.

The transistor has an additional node that represents the device temperature in Kelvin as a voltage relative to ground. The "logical" thing to do would be to connect together the thermal nodes of two closely-spaced critical transistors like an input pair to model thermal crosstalk, but this gives a VERY wrong, non-physical result. It turns out that modeling anything more than individual device self-heating with any degree of accuracy is a *very* hard problem.

What I don't understand... why not just a behavioral model that runs under Spice? Why all this dumbness of tabulating V-I and T-V data? More Cadence self-aggrandizement ?:-) ...Jim Thompson

PSpice does. As far back as when I put TTL outputs on PLL chips :-)

There's actually a chapter in one of my PSpice tomes addressing thermal modeling.

...Jim Thompson

He ran Spice. I measured actual transistors.

OK, I gave it a try. My measurement resolution is 1 fA, but things get dicey down there and things take minutes to settle. My best guess is that, at 5 volts reverse, the c-b junction of a BFT25A leaks about 4 fA at room temp. If I repeat with b-e shorted, it increases to about 7 or maybe 8 fA, close to twice the leakage.

I did parallel the transistor with a 1T resistor (which leaks 5 whopping picoamperes at 5 volts) and got close to the expected current, so my measurement rig isn't being too obviously silly.

Even if I'm off some, the BFT25 c-b junction is the lowest leakage diode I've ever encountered.

John

Thanks, John.

I was unsure whether you had connected the emitter back when you mentioned how low the c-b junction leakage was. Someone wrote that connecting the emitter (can't find the thread, though) changed things -- can't remember if it was good or bad, I thought it was good, possibly in a totally different context.

So now we now about the very sensitive end of low leakage diodes :) Complete with sanity check!

I settled for a 1N3595 recently as it leaks a lot less (1nA) than the 'HC4053 analog mux chip (100nA, both datasheet 25'C numbers) it is protecting from signal over-voltage.

For the curious, I have two signals, out of OP07s running from +12V and -5V, via 1k6 resistor, a 1N3595 diode to 4V stops the signal exceeding 5V as it goes into a 'hc4053 mux switch. There's a cap to analog ground as well. The 4V is a TL431 with scaling resistors (could use 3.9V zener) sinking several mA so it has low impedance.

I'm using +12/-5 because the OP07 output doesn't go anywhere near supply rails, and the negative signal on one channel may go about a volt or so -ve. Both signals can go to ~3V +ve.

Grant.

Good. Don't count on the HC40xx mux ESD diodes to clamp signals, even if the source resistance is high. Once you get a little beyond the rails, the internal series switch fets can be turned on, and all hell breaks loose, many nanoamps of it.

Central Semi makes some nice dual super-low-leakage surface-mount diodes, 10s of fA.

John

Semi manufacturers treat their models like the family jewels they are. You signed NDAs for all the models you have, didn't you? IBIS models aren't (at least none that I know of) released under NDA because they really are behavioral.