Maintaining a Vbe Multiplier's bias value

[Ban]

On Thu, 11 Feb 2010 07:34:40 -0700, Jim Thompson

snip>
In other words, although OnSemi has a realistic model for
their own parts, which is fine for simulating their parts
more accurately -- is there a reason to shop around and
actually _select_ someone else's parts for some application
reason. And in what cases would you not bother wasting time
shopping around and for what other cases would you decide to
spend the time, because you know enough about how they are
made and what differences that can make to be worth that
effort to test and verify when making a selection?

Jon

Hi Jon,
I like your approach coming from equations to analysing practical circuits.
But then follows the verification with real parts. I wouldn't worry with
small details, as long as the general idea is understood and followed. When
you have performed some measurements on real parts, it is good to understand
the important parameters, so you can tweak them yourself.
ciao Ban

 

[Jon Kirwan]

Once upon a time there was JEDEC, and all 2N2222A's had to be the
same in regards to essential specifications.

Okay. That's gone, then.

But I'd use the OnSemi model, irrespective... the LTspice version lets
too many variables drop back to their default values... might not
matter, but who knows.

...Jim Thompson

I'd still like to _learn_ about FAB processes, geometries,
mask steps, subtrates (and if any BJTs include a bond to such
things), and differences between them. For example, I've
heard you talk about processes that include gold as a step
(or more?) I'd like to know what does what. I can (and have
attempted) a few 2D spatial integrals aka Hauser's analysis
of crowding on r_b many years ago, and I'm vaguely aware of
the fact that he neglected to account for lateral base
diffusion which happens when the crowding and some local base
widening takes place. I actually _did_ take measurements of
real Hamamatsu diodes, years ago, and reversed out from the
measurements what the dopant concentrations had to have been
so that I could better model the behavior over a wide range
of temperature operations (Hamamatsu flatly refused to give
me any such information.) The resulting model I created
_did_ model that photodiode at -40C to 55C better than I'd
expected it to do and much better than the gross models I had
at the time were able. So at some point, I'd like to study
these things to get a better feel... but I'd like to know who
has what FABs and what the processes are capable of and
produce.

I hope it's not as difficult as pulling dopant numbers out of
Hamamatsu was! I'm not wanting to know specific recipes or
anything -- just process capabilities. Hopefully, FAB and
process capabilities and locations are something that is
known about and published. I can hope.

Jon

 

[Jon Kirwan]

I like your approach coming from equations to analysing practical circuits.

It's the thinking processes that I'd like to encourage in
myself. Being able to deduce to specific cases from theory
seems a vital part of not just copying others but being able
to think on my own, at times. Of course, theory doesn't
necessarily tell me what the constants are -- to get those, I
can always scout for models... but as this part of this
thread clearly shows, that only gets you within some very
vague area. But theory remains VERY important all the same,
even if in practice the detailed constants applied may be
just about anything. Even with the same part numbers, as the
recent discussion shows.

But then follows the verification with real parts.

Yes! Of course. Theory can be used to deduce math models
and models to express the math order of error terms ignored
in them or the frequencies involved (for example, theory can
tell you whether or not even order harmonics are possible
from the deduction in hand), sans calibrated constants. To
compute quantities, though, it helps either to have accurate
model constants or else go to the bench and test out the
facets you care about at the time.

I wouldn't worry with
small details, as long as the general idea is understood and followed. When
you have performed some measurements on real parts, it is good to understand
the important parameters, so you can tweak them yourself.

I have some idea how to measure most useful parameters of
BJTs. Problem is, it can be pretty hard to get for some.
(Read: "work+equipment.") One of the things I'd like to do,
when I get further along, is to design and build a BJT tester
with a micro that can extract parameters and generate models
at the EM1, EM2, EM3, and GP levels, at least. Maybe even
VBIC, if I can grasp for it. That will teach me so much and
I can take it in steps, so long as I can think out what I
need in the first place. For example, I will need to be able
to set voltages, monitor pin currents; or set currents, and
monitor pin voltages; or just observe voltage after setting
high impedance; etc. It would be nice to do a lot more than
just curve tracing. And I'd learn a great deal from all
this, spread out over a time by which I can assimilate each
part in mind.

Thanks,
Jon

 

[Jon Kirwan]

I'm not sure how you can sleuth out doping profiles in that manner.

It's been 20 years now, but I used 1D and 2D models I found,
which applied dopant concentrations to predict behavior. I
could go find the book, I suppose, but I figured the
information must be _generally_ understood. Certainly, when
I was looking, it wasn't that hard to find. The issue was in
plugging in the numbers. What I did was to take a series of
measurements over a few weeks' time and then used a short
program I wrote to "tinker" the parameters until the
predictions came very close. I knew physical dimensions for
the die.

When I've had the need to get such numbers, I would use an outside
lab.

This was a one-off and the entire company was 20 people,
including part time. I don't think they could have afforded
the price. In my case, I got paid a few hours' work but the
data collection just ran on its own. Most of what was needed
to set it up was there and we bought a Burr-Brown board for
the rest.

They can actually profile the device, so you don't just see the
net dopant, but how it changes.

I can't say that I knew the dopant levels, for sure. It's a
high quality photodiode, though. It's not complicated and
the integrals weren't anything to write home about, as I
recall. What I needed to model was dV/dI over temp. Not
everything else. So maybe I got lucky. That was my focus,
though.

In the end, when I used those dopant figures and then re-used
the equations from the book to predict behavior at points
where I had _not_ measured it, but interpolated as well as
extrapolated, and predicted very well. We had taken
measurements down to -5C but the model nailed the observed
dV/dI at -40C and at 55C, as well. Which is an impressive
range to someone like me. Seemed to solve a problem, anyway.
Equipment is still in the field and 'doing fine.'

I've managed to discover a few trade
secrets doing this.

I can imagine.

What had surprised me is that a neophyte like me could pick
up some basic books on the subject, cold, and arrive at
something that worked reasonably well and allowed me to
deduce equations for automatic corrections that worked so
well. Although it could be entirely accidental, when you get
such results from the application of semiconductor theory it
is hard to believe one could be _that_ lucky (so I think
something was nailed down right and the physicists working on
these things actually know their stuff pretty well.)

In the days before using epi for CMOS was common,
a company where I worked would ion implant the wafer from the back.
That was like a faux epi. A piece of cake for the outside lab to
spot.

Hehe.

I wish I knew more of this stuff. It's interesting.

Jon

 

[Jon Kirwan]

Big Grins!

Yeah I applaud your effort, I wait for further posts.

For me, I’m building electronics to either detect something or drive
something that’s detecting something. So the fun is in making good
detectors or drivers.

George H.

Well, I am wanting, eventually, to build something I need.
Something I cannot buy in the market because the need is
unique.

This divides into two parts. Design and build. Since the
item is unique, I can't just go out and buy it. And getting
the features I need cannot just be "hacked" into existing
designs without at least knowing _some_ stuff, first. I
might as well turn the "design" part into a fair learning
experience, as a separate project of its own. Get past that
and when it comes time to build what I want I'll be able to
build on what I learned and add what I need and then do a
modest hobbyist level whack at actually making what I want to
make.

If someone else were to do this for me (hire a designer),
they'd get all the fun of learning on the job and taking my
money with it. They get the money, they get to further their
own education, and I get a tool. One tool. Once. Next
time, I get to pay someone else to learn for me.

It almost feels like paying someone to go do your exercising
for you. No satisfaction and no weight loss. They get all
the _real_ benefits.

Part of the fun isn't the destination itself but it is what
you see and enjoy while getting there, too. You take a plane
when all you need is to "get there" quick, but you drive when
you want to enjoy stops along the way. I used to fly to
Burbank every week for a year and a half. Slept in a hotel
for 3 nights a week, worked day and night in between, flew
home. Barely saw anything but hotel room walls, cubical
walls, a few cement roads, pollution so thick you couldn't
see the Burbank hills from the Lockheed center, and not much
else. The destination was important, of course. Paid the
bills and I enjoyed the work, too. But there is a lot more
to see in the 1000 miles from here to there.

Anyway, I'm driving this time, not flying.

Besides, I'd rather _keep_ the money and _keep_ the education
for myself. That way it pays off, again and again.

Jon

 

[Phil Allison]

"George Herold

So how do you do push pull with tubes, or say with only Jon's npn
transistors?


** There are literally *millions* of push-pull tube amps in use - the vast
majority of tube hifi and guitar amps are push pull designs.

NPN output transistor amps are called "quasi-complementary push pull " -
many millions of them made and sold too.

Use Google to find the schems.

Idiot.



..... Phil

 

[Ban]

On Thu, 11 Feb 2010 11:50:23 -0700, Jim Thompson






I'd still like to _learn_ about FAB processes, geometries,
mask steps, subtrates (and if any BJTs include a bond to such
things), and differences between them. For example, I've
heard you talk about processes that include gold as a step
(or more?) I'd like to know what does what. I can (and have
attempted) a few 2D spatial integrals aka Hauser's analysis
of crowding on r_b many years ago, and I'm vaguely aware of
the fact that he neglected to account for lateral base
diffusion which happens when the crowding and some local base
widening takes place. I actually _did_ take measurements of
real Hamamatsu diodes, years ago, and reversed out from the
measurements what the dopant concentrations had to have been
so that I could better model the behavior over a wide range
of temperature operations (Hamamatsu flatly refused to give
me any such information.) The resulting model I created
_did_ model that photodiode at -40C to 55C better than I'd
expected it to do and much better than the gross models I had
at the time were able. So at some point, I'd like to study
these things to get a better feel... but I'd like to know who
has what FABs and what the processes are capable of and
produce.

I hope it's not as difficult as pulling dopant numbers out of
Hamamatsu was! I'm not wanting to know specific recipes or
anything -- just process capabilities. Hopefully, FAB and
process capabilities and locations are something that is
known about and published. I can hope.

Jon- Hide quoted text -

- Show quoted text -

A very comprehensive book, bible lets say is
The Crcuits and Filters Handbook by Chen
don't get fooled by "Handbook", it's the biggest book I got. )
ciao Ban

 

[Ban]

"George Herold

So how do you do push pull with tubes, or say with only Jon's npn
transistors?

** There are literally *millions* of push-pull tube amps in use - the vast
majority of tube hifi and guitar amps are push pull designs.

NPN output transistor amps are called "quasi-complementary push pull " -
many millions of them made and sold too.

Use Google to find the schems.

Idiot.

.... Phil

Thanks Phil I'll try google. Why are there still so many tube amps?
You'd think someone could make a solid state amp that sounded
'right'.

A lot of transistor amps sound right, but some people prefer to add a bit of
2nd harmonics, or they want to enjoy the gloom.
ciao Ban

 

[Ban]

IIRC the LM395 is basically an LM309 with the voltage reference removed.
Emitter-follower regulators are nearly bulletproof unless you discharge a
cap into the output.

or the reference input. But some power diode antiparallel will take care of
that.
ciao Ban

 

[Jon Kirwan]

Usually two "NPN" tubes whose plates drive a center-tapped output
transformer.

John

Just to be clear, with B+ at the center-tap.

Jon

 

[Jon Kirwan]

<snip>

A very comprehensive book, bible lets say is
The Crcuits and Filters Handbook by Chen
don't get fooled by "Handbook", it's the biggest book I got. )
ciao Ban

Cripes. If you walked into the house you'd see walls of
polished bookshelves from floor to ceiling covering every
spare bit of wall space in my library off the mail hall. I'm
already overflowing out of there into another room and my
wife isn't exactly happy about it. But what's another book
or two? I'll add it to my next order from somewhere.

Jon

 

[Jon Kirwan]

The fish/fish rule ?

...Jim Thompson

Almost. It's more like: I'd like to get to Fresno Flats
(aka Oakhurst, near Yosemite) from Portland, but rather than
just fly there (okay, so I have to parachute out) I'd like to
enjoy the trip this time, too, and detour around a bit to
some of the sights along the way.

First off, maybe to the Opal Creek Wilderness area to see
some of the 800 year old cedar groves; then maybe south to a
short visit at White's Electronics in Sweet Home; then to
Crater Lake; maybe then to Oregon Caves; off perhaps to do
some panning for gold in Jackson County while also taking a
hike in the Siskiyous; then on over to hwy 101 at Brookings
to do some whale watching; down to Crescent City to visit the
Jedediah Smith and Del Norte parks and see what's left of the
redwoods there; ....

Well, might as well make a thing of it this time 'round.
Oakhurst won't go away while I'm on the road.

Jon

 

[krw]

ST Arm Cortex-M3 has an internal 1.8V regulator for the core, and can take
any input voltage from 2.0 to 3.8V.

Same with the Altera Max-II CPLDs. They'll take 1.8V, 2.5V, or 3.3V.

 

[krw]

Got snow?

We have about 1" now. They cancelled work for today about 9:30 last
night. Since there was no snow this morning (it started about
10:00AM) I went into work. I was the only one there. My wife's
employer closed about 2:00, with less than 1" on the ground. The
streets are just now starting to get some slush on them. I guess it's
a good thing the locals are scared shitless of snow.

 

[Jon Kirwan]

This is an example of a common emitter voltage amplifier. It might be one of
the easiest stages to design. Normaly, I would start off by knowing what I
need for the stage, for example I need a voltage gain of 20, an input
impedance of 10K, Z out of ?? etc ----
<snip>

I take your point about selecting a BJT for the desired Iq.
After that, my next consideration is in setting the
collector's average Vc, given an Iq. I want to leave 1V for
the dc bias of Ve to stay well above (kT/q)/Iq. And since I
want to keep the BJT unsaturated I also preserve 1V for Vce.
In your 15V rail case, that leaves me 13V for Vc to wander
around in. Half of 13V is 6.5V. Add back in the 2V I'm
saving, and that sets Vc at Iq as 8.5V.

I set the collector resistor to (Vbat-Vc)/Iq. I get 6.5k for
that. What the heck, make it 6.8k. Iq is now 956uA. Oh,
well.

The emitter resistor is easily set to 1/956uA or slightly
more than 1k. Call it 1k. This means Ve will be about
0.956V. Livable.

On the biasing, I start by assuming that about Iq/5 should
flow in the divider. Say 200uA? (You started out thinking
in terms of the thevenin, instead, and a different rule of
thumb for it. I just use the 1/5th rule.)

Assume Vbe about 0.7V, the base should be at 956mV + 700mV or
about 1.66V. Maybe a little less, maybe a little more.
Assume less, hope for a little more. So 1.6V/200uA is 8k and
(15V-1.6V)/200uA is 67k. Let's go for gusto and pick 68k and
8.2k.

Now, I want a gain of 10? I can either bypass the emitter
resistor with another R+C in series across it or I can divide
up the emitter resistor into two pieces in series and bypass
just one of them, leaving the other one active at AC. With
the collector resistor of 6.8k, a gain of 10 would suggest an
AC resistance of 680 ohms in the emitter. To get that from
the 1k DC for the emitter, we put that 680 in series with a
330 and bypass the 330 with a cap sized appropriately for the
lowest frequency of interest to be close to a 'dead short.'

AC impedance is going to be about the thevenin of the base
pair of resistors divided by (1+5/beta) -- the 5 comes from
my 1/5th factor I earlier chose. If I keep that very much
smaller than the beta, it doesn't affect things much... as
you say. The thevenin of 68k and 8.2k is about 7.3k. With a
beta of 100, for example, this drops to about 6950 ohms.

I'll stop at this point and plug it into LTspice with a
2N2222 model they include (who knows if it is 'good'?)

Adding a signal source through a cap to base, I get a gain of
9.6, an average base voltage of 1.58V, average Iq of 921uA,
and a center Vc of 8.74V. AC impedance is computed as 6961
Ohms in LTspice. Pretty close, really.

Now... to the breadboard for a quick DC check. OnSemi
PN2222A just taken from an ammo pack. Collector resistor of
6.800k, emitter resistor 1.008k, base to ground resistor of
8.360k, base to V+ of 69.53k. Measures (rounded):

Vbat = 15.1
Vb = 1.59
Ve = 0.95
Vc = 8.73

I haven't hooked up the signal generator, yet. I'll need to
move upstairs to do that and get the scope fired up. But
that's a quick check of reality.

Now, plugging the Vbat back into LTspice and the real values
of the resistors I used, I get this from LTspice (rounded.)

Vb = 1.59
Ve = 0.94
Vc = 8.8

Which is ... pretty close. (The OnSemi PDF for the part
doesn't include a spice model for it and a search on their
site only comes up with the MMBT2222, so I'm using the
LTspice model for now.)

Jon

 

[Jon Kirwan]

Actually you found the "art" part needed to rough-in a design. Now go
back and fix the flaws ;-)

...Jim Thompson

One improvement is to boostrap from emitter to base divider
and use a resistor from there to the BJT base, with the
signal tying in directly to the base via the cap.

Jon

 

[Jon Kirwan]

Don't try to get too exotic too fast... you'll get peaking, or worse,
oscillation, if you don't know what you're doing.

But that's a good thing... try it and learn from it ;-)

...Jim Thompson

Okay. Slow down.

So how about a simple (non-Wilson, for now) current mirror in
the emitter? It's easy to select a resistor for the other
side of it to set the current. Then an R+C leg can be used
to set the AC gain!

Jon

 

[Jon Kirwan]

And do the math as you go... so you understand what you're seeing.

Don't worry, that I will do. It's what drives me. I enjoy
developing different questions and doing the differentials to
see where that takes me. Gives me an excuse.

Anything in particular to include in the math? (I may choose
to simplify, so it wouldn't hurt to have a clue where not to
do that.) I assume I should be looking at instantaneous
effects due to T_ambient, but what about differential T in
various parts? Anything important here? (The current mirror
does help a little here, I'd guess, without doing the math
yet.)

I was fortunate, grew up in a radio and TV repair shop, and I was
already math-heavy when I trotted off to MIT... had a Russian battle
axe of a teacher named Evelyn Truchovesky for my first round of
Algebra... she pounded me so good I adopted her way of writing "E" in
my signature... to this day ;-)

I entered college already knowing _some_ calc. Enough to
score a perfect 800 on my SAT math part before entering.
Didn't help me pay the bills, though, or deal with
people-created paper work barriers.

Washed dishes first year.

I worked a burger joint, washing dishes and cleaning; and
worked also as a janitor doing commercial steam-cleaning of
carpets at restaurants while taking more than a full load.
Couldn't handle two jobs _and_ school _and_ the rest, at one
time, at 18. Almost, maybe. But unlike in horseshoes and
a-bombs, _almost_ doesn't cut it. Dad was dead, mom wasn't
able to help out, and eventually decided to become self-
employed. Which is how I've been, almost all the years since
then.

Second year onward, tech'd in Woodson,
Jackson, Melcher MHD lab (in Building 20)... invaluable experience!

Learn Laplace short-hand, it'll be invaluable!

...Jim Thompson

I'm gradually getting more comfortable with Laplace, as I
continue to work more problems. It is an especially nifty
way to solve some differential equations, which is what it
was designed to do, I think.

Jon

 

[JosephKK]

That one takes an approach that I'm not familiar with and
didn't take. I'll have to consider the approach more.
However, I did take a look at the end of it. It says:

R = (R1+(R2||re)) / (1+(1/R1+gm)*(R2||re))

If I understand the value gm, and I may not, it's just 1/re
or else re=1/gm. Basically, just the (kT/q)/Ic I'd mentioned
when I wrote. If that is the case, I used these to see how
that page predicts:

ic=.005
vt=k*300/q
gm=ic/vt
re=1/gm
r1=1000
r2=1000
r2p=r2*re/(r2+re)

and then computed:

(r1+r2p)/(1+(1/r1+gm)*r2p)

and got:

502.5719049 Ohms.

This is so far from my own calculations of about 15.4 Ohms
that I just _had_ to put this into LTspice and test it. To
do that, I simply set up the basic circuit with the two
resistors and BJT and then hooked up a variable current
source to the topside. I set it up as an AC source of 5mA
with peaks of 500uA, and then ran a .TRAN on it and plotted
the upper rail of the structure's voltage. I used a 2N2222
BJT, as well. Convenient, and I have them laying about.

Anyway, so I ran the sims and got 17.44mV, peak to peak.
Divided by the peak to peak current variation of 1mA gives an
apparent R of 17.44 Ohms. My calculations arrived at 15.4
Ohms, or so.

All this could be operator error. I may be operating the web
page you suggested incorrectly, so that the 503 Ohms I get is
because I didn't know what I was plugging in and where. I
may be operating LTspice incorrectly, so that it's results
aren't usable and it's just luck that the numbers worked out
in my favor.

But there it is.

Here is the LTspice file:

Version 4
SHEET 1 880 680
WIRE 128 0 16 0
WIRE 224 0 128 0
WIRE 288 0 224 0
WIRE 128 32 128 0
WIRE 16 112 16 0
WIRE 224 112 224 0
WIRE 128 160 128 112
WIRE 160 160 128 160
WIRE 128 208 128 160
WIRE 16 224 16 192
WIRE 128 320 128 288
WIRE 224 320 224 208
WIRE 224 320 128 320
WIRE 128 336 128 320
FLAG 128 336 0
FLAG 288 0 V_rail
FLAG 16 224 0
SYMBOL npn2 160 112 R0
SYMATTR InstName Q1
SYMATTR Value 2N2222
SYMBOL res 112 192 R0
SYMATTR InstName R1
SYMATTR Value 1k
SYMBOL res 112 16 R0
SYMATTR InstName R2
SYMATTR Value 1k
SYMBOL current 16 192 R180
WINDOW 123 0 0 Left 0
WINDOW 39 0 0 Left 0
SYMATTR InstName I1
SYMATTR Value SINE(5m 500u 50)
TEXT -76 296 Left 0 !.tran 1


Well, let's assume that I got lucky and LTspice and I agree
on the figure of about 16 Ohms. With a signal at 20Hz, we
are talking:

C = 1/(2 PI f (R_ac/10)) = 5000uF

Yikes. John L. wasn't kidding when he wrote "big." Luckily,
in steady state it could be a low voltage cap!

Jon

That is substantially larger than what i have seen in commercial
audio amplifiers (about 5X to 25X), but not particularly surprising.

 

[Jon Kirwan]

<snip of discussion about a 5000uF cap across a Vbe multiplier>

That is substantially larger than what i have seen in commercial
audio amplifiers (about 5X to 25X), but not particularly surprising.

I like the idea of first making the function blocks
themselves better behaved.

Jon