favorite temp sensors

[Bill Sloman]

You're behaving like the brat.

Scarcely. Un-intelligible schematics are very nearly useless - too much risk that a crucial detail will be misinterpreted.

 

[John Devereux]

It's an absolute-temperature measurement, so +/- 0.2 C implies better
than a part-per-thousand error band. Voltage references and resistor
values for the current pulse would have to be better than 0.1%, so probably
the answer is NO. The only way to get components with that accuracy
is with trimming, and that means ALL qualified sensors are dependent
on calibration or trimming. You can choose which, but you
must pay, either way.

Yes this is true. It is not clear that Jims way is simpler. As others
have said, manufacturers have gone to a lot of trouble to get IC sensors
that still do not achieve this. A 0.1% ratio reference resistor is
cheap. If you don't want Bills bridge+adc, you can do something like this:

<http://ee.devereux.me.uk/ratio-coded-adc.png>

(opamp choice just for simulation)

 

[Bill Sloman]

Power dissipation is 0.7mW and thermal resistance to ambient air is 37C/W,which is 0.026C of self-heating, which isn't too shabby.

http://www.analog.com/static/imported-files/data_sheets/ADT7320.pdf?ref=ASC-PR-479

I like it.

Good news, thanks. Now all we need is the same thing in a package that
isn't such a good thermal insulator.

There's an "exposed pad" under the device, which you could solder to a copper ground pad, or couple - without electrical connection - with thermal conductive adhesive. What kind of package did you have in mind - one of the options that National Semiconductor offered for the LM12?

http://www.ti.com/lit/ds/snosby8d/snosby8d.pdf

They did have an order of magnitude less thermal resistance to ambient ...

 

[Spehro Pefhany]

Power dissipation is 0.7mW and thermal resistance to ambient air is 37C/W, which is 0.026C of self-heating, which isn't too shabby.

Ummm.. Bill, I _really_ question the general validity of that 37°C/W
number in the datasheet .. I bet that's measured on a good-size chunk
of FR4*.. think about putting a watt(!) into something that tiny. They
don't state the coupon size in the datasheet. Fine, if you're
measuring PCB temperature, maybe not fine otherwise.

A Pt100 at 0.2mA is 4uW, even a Pt1000 at the same current is still
only 40uW- with similar surface area and a much nicer form factor.

*Okay, EIA/JESD 51-X (if that's what they're using) calls for
something like a 3 x 4" chunk of 4-layer FR-4 with 1-oz ground and
power planes!! A TO-220 (with maybe an order of magnitude more
surface area) has typical Theta-JA of 60°C/W (and that would include
heat flow down the leads). WAG for the Theta-JA on a thin bit of flex?
Maybe 500-1000°C/W?

Often the thermal impedance and response times of temperature sensors
are 'optimistically' rated in air blasting at 3m/s or stirred oil.

BTW, note too that there is a temperature dependency in the power
dissipation, especially in continous mode. At 1sps & 3V, power goes
down to a fairly constant 150uW, which is more reasonable.

Can't complain too much though, it's the sort of thing that keeps food
on the table and the wine fridge fairly well-stocked.


Best regards,
Spehro Pefhany

 

[Bill Sloman]

Ummm.. Bill, I _really_ question the general validity of that 37°C/W
number in the datasheet .. I bet that's measured on a good-size chunk
of FR4*.. think about putting a watt(!) into something that tiny. They
don't state the coupon size in the datasheet. Fine, if you're
measuring PCB temperature, maybe not fine otherwise.

The exposed pad on the under-side of the package is for thermally coupling it to the board you mount it on. If you thermally couple that board to the object whose temperature you are measuring - in David Lesher's case, a copper pipe - I can't see that there's going to be much of a problem.

The tiny packages haven't got much surface area, but they are thin,and there's not a lot of plastic packaging to slow down the heat flow. You may wellbe right, but it's not guaranteed.

A Pt100 at 0.2mA is 4uW, even a Pt1000 at the same current is still
only 40uW- with similar surface area and a much nicer form factor.

It depends on the Pt100. If it's a thin film sensor on an alumina substrate, thermal resistance can be quite low. I got enthusiastic about that in print back in 1978

Sloman, A.W. "On microdegree thermostats", Journal of Physics E: ScientificInstruments, 11, 967-968 (1978).

*Okay, EIA/JESD 51-X (if that's what they're using) calls for
something like a 3 x 4" chunk of 4-layer FR-4 with 1-oz ground and
power planes!! A TO-220 (with maybe an order of magnitude more
surface area) has typical Theta-JA of 60°C/W (and that would include
heat flow down the leads). WAG for the Theta-JA on a thin bit of flex?

Maybe 500-1000°C/W?

Probably not. Your TO-220 thermal resistance of 60°C/W is from junction to air. The thermal resistance from junction to case is about 1.5°C/W, andit's a much fatter package than that of the ADT7320.

Often the thermal impedance and response times of temperature sensors
are 'optimistically' rated in air blasting at 3m/s or stirred oil.

The ADT7320 quotes the junction-to-ambient at 37°C/W in still air, the junction-to-case as 33°C/W. Thermistors are often quoted "in a stirred oilbath", but semiconductor manufacturers have a pretty well-defined set of rules.

BTW, note too that there is a temperature dependency in the power
dissipation, especially in continuous mode. At 1sps & 3V, power goes
down to a fairly constant 150uW, which is more reasonable.

150uW is better than 700uW, but it's not spectacularly less.

Can't complain too much though, it's the sort of thing that keeps food
on the table and the wine fridge fairly well-stocked.

People not thinking about self-heating can also be good for an extra citation ...

Sloman A. W. “Comment on ‘A versatile thermoelectric temperature controller with 10 mK reproducibility and 100 mK absolute accuracy’ [Rev. Sci. Instrum. 80, 126107 (2009)] “, Review of Scientific Instruments 82, 27101- 027101-2 (2011).

There were claiming 100mK absolute accuracy while generating 200mK of self-heating in their thermistor. Clown car stuff.

 

[Jamie]

I used 1K platinum RTDs, 2-wire, and RG174 coax for my cabin automation temp
sensing. The longest run is about 75 feet. I took out the coax resistance by
calculation and got better than 1 deg C. I got compulsive (cabin fever?) and did
an ice-point cal to final trim things up.

I did a simple mux thing with 0.1% resistors, ratiometric, all the math in
PowerBasic.

There are some 24 bit SPI delta-sigma ADCs with two differential channels. Hang
an RTD and a good resistor in series, across the ADC reference supply, and
measure the two voltage drops, and do the math. That will measure RTD resistance
to within a few PPM of the reference resistor. For the cabin, I wasn't *that*
compulsive.

I was pricing some 16 bit SPI DAC's and I thought 55 bucks for an 8 pin
chip was a little steep. So I can imagine what a 24 bit ADC cost...

Jamie

 

[Bill Sloman]

Maybe you need your reading glass prescription upped to a double lens
stack?

I've got Zeiss varifocal lenses, and they work fine, and I can take them off for close work - the short-sighted eye focusses perfectly on stuff that's between four and ten inches away

The problem isn't seeing the symbols, but the way they are organised - or not organised in this particular case - on the schematic.

There's an art to making a schematic easily readable, and you hadn't mastered it back then. Your recent stuff is much better, if still a bit wee\sk on grouping stuff by function.

 

[Bill Sloman]

I was pricing some 16 bit SPI DAC's and I thought 55 bucks for an 8 pin
chip was a little steep. So I can imagine what a 24 bit ADC cost...

You seem to have imagined wrong

http://parametric.linear.com/Analog...54,1367,10002!s_1049,1!gtd_!1049_24:24!1030_2

Newark lists the LTC2402 at $11 one-off, and the LTC2412 at $9.52, though they've got a promotion going on so today's price is only $7.41.

TI has similar parts that are cheaper. The ADS1226 seems to be $7.65 (1 to 9)
from Mouser.

 

[Spehro Pefhany]

On Fri, 12 Apr 2013 18:02:52 -0700 (PDT), George Herold

On Apr 12, 8:46 pm, Jim Thompson <To-Email-Use-The-Envelope-I...@On-My-
Web-Site.com> wrote:
On Fri, 12 Apr 2013 17:43:33 -0700 (PDT), George Herold
[snip]

IC temperature sensors are all crap. Every single one, AFAICT--none
that I know of claims accuracy better than 1 degree C, even the trimmed
ones. Why would that be, if it's so fundamental?

I don't know of a single commercial unit that uses the ratiometric
current method, they're all PTAT's, some trimmed, some not.

Jim, I'm confused (or just ignorant again).. I thought PTAT's used the
current ratio trick.

George H.
[snip]

Different ratio. PTAT's generate a voltage that indeed is ratio'd by
using two devices of differing areas, thus different current
_densities_.

The Jim Williams' technique, using resistors and adapted by me to use
current sources, uses the same device, "measured" at two carefully
ratio'd currents.

The PTAT effect depends on lots of other variables as well.

I'll dredge thru my files and find one I can show.

...Jim Thompson

A very old one I designed for Fairchild...

http://www.analog-innovations.com/SED/PTAT_Demo_SED.pdf

Explanation to follow... I'm off to the 5-year-old grandson's ice
hockey practice... he made the team >:-}

...Jim Thompson

http://www.analog-innovations.com/SED/Killian_Thompson_Age_5.mp4

...Jim Thompson

So, can we count on you incorporating hockey metaphors into your
posts? Say "deke" and "stick-handle", for starters?


Best regards,
Spehro Pefhany

 

[George Herold]

Hi, George. I posted up a link to Linear's AN45 elsewhere
under this topic. See page 7 there. But I take your
experiences here seriously and wanted to think about this,
not at the 'charged gas' theory level but at the higher (and
more usual for an EE) device modeling level.

The two-current pulse method, using say 1X and 10X currents,
depends upon the following:

        dV = (k/q) * ln( 1+Ic/Is ) dT

Although k and q are known, the entire factor that includes
the ln( 1+Ic/Is ) part isn't knowable in advance. But if one
assumes that the +1 term is negligible then the two pulsed
currents results in:

        dV1/dT = (k/q) * ln( Ic1 ) - (k/q) * ln( Is )
        dV2/dT = (k/q) * ln( Ic2 ) - (k/q) * ln( Is )

Subtracting dV1/dT from dV2/dT yields:

        dV1 - dV1 = (k/q) * ln( Ic2/Ic1 ) ) dT

And if the ratio of Ic2/Ic1 is known a priori then the entire
factor, (k/q) * ln( Ic2/Ic1 ) ), is also known. And as a
consequence could be used to measure temperature without
having to calibrate the system. Or so it seems at first
blush.

But it's also the case that the value of the saturation
current, Is, is itself a rather complex function of T:

  Is(T) = Is(Tn) * (T/Tn)^3 * e^( -(q*Eg/k) * (1/T-1/Tn) )

Where Tn is some chosen T(nominal).

In fact, this particular component is what overwhelms the
first equation (which is positive vs temperature) and yields
the usually quoted -2mV/K figure (very approximately.) So, in
fact, Is(T) is the dominant factor in Vbe change over T and
in no possible way is it a simple function of T!

(Even the above Is(T) equation itself is a simplification.
The power ((T/Tn)^3) for example is an approximation and not
strictly true in practice. Same with Eg, which itself is also
taken as a single approximation value.)

Just as a guess, the idea of ln( Is ) being cancelled
entirely out of the equation by ratiometry, even assuming
that the die is at thermal equilibrium, would make me worry a
little. (I accept that pulsing the 1X/10X current change fast
enough or that using low enough currents, like the 1uA and
10uA you mention, would yield a near-equilibrium state.) I'm
just not sure that at this level of modeling, that _Is_
remains dead stable as a modeling parameter when facing a 10X
current change. There is a lot of linearity over orders of
magnitude change, as a broad statement. But exactly how
linear is it when provided a two point ratio a decade apart,
vs device variation?

I wonder that some temperature error is swept under this
Is(T) rug and hidden from the analysis, so to speak. Even
assuming thermal equilibrium. Because it may really be that
Is(V,T), not Is(T), as both the power (^3) and Eg are taken
as simple constants for simplification when they aren't, in
fact, invariant at this level of modeling.

You mention base currents as a possible error. I've ignored
that so far. The equation:

        dV = (k/q) * ln( 1+Ic/Is ) dT

in the diode connected case refers to Ic. The currents
through it, on whole, are (beta+1)/beta times as much. If you
cobble up precision current sources at exactly 1X and 10X,
the ratio of Ic2/Ic2 would still be 10, even though you are
driving Ie, I think. However, beta itself changes vs Ic. So
there is that to account for, if you were only using a beta
level model. But the:

        dV1 - dV1 = (k/q) * ln( Ic2/Ic1 ) ) dT

method doesn't use or rely upon beta. So I'm not imagining a
problem there because (1) the ratio is still 10X and (2) beta
isn't used in the analysis method.

Interesting problem getting past a certain level of accuracy,
though. There must be several papers that go beyond the AN45
app note I'd posted up earlier. I haven't seen one, yet.

Jon- Hide quoted text -

- Show quoted text -

Hey Jon, I was thinking about this, and it seems like the error is
due to the non-ideality factor (NIF) in the equation. Now I've only
read about the NIF in the context of pn junctions, but wouldn't there
be something similar in a diode connected transistor?
Now according to Streetman the NIF arises because of carrier
recombination in the transition region. I know when I looked at the
temperature dependence of some pn diodes (maybe 1n4148's?) that the
NIF was much closer to 2.
So then what transitors would have NIF's close to 1?
I wonder if they list NIF's in the spice models?

George H.

 

[George Herold]

I recently designed a system to measure the temperature of the metal
structure holding parts of a radar antenna up.  The resolution was 0.1
C, and the accuracy was nominally 1.0 C, the purpose being to be able
to calibrate over a range of temperatures with a reasonable number of
temperature sensors.  The cable between cabinet and measurement point
was up to 10 meters long.

First tried thermocouples.  The problem was getting the signal through
standard mil-spec connectors - the junction from chromel-alumel to
brass to copper was not temperature controlled, and things rapidly got
complex.

Then tried thermistors.  The beads were good enough, and the resistance
was high enough to render cable resistance immaterial.  But the ICs
available to put on the circuit board had too few ADC bits, yielding
too coarse a resolution.

Settled on platinum RTDs in a full Kelvin (4-wire) configuration, after
finding a suitable IC from Maxim (or LT?).  This chip is ratiometric,
so we used a low-tempco metal-film resistor as the comparison.  If
people are interested, I'll look up the chip number.

Joe Gwinn- Hide quoted text -

- Show quoted text -

Hi Joe that's neat, I'm a bit confused as to why platinum RTD's were
easier thant thermistors?
I've never used a platinum RTD.

George H.

 

[George Herold]

On Sat, 13 Apr 2013 09:40:38 -0700, Jim Thompson

02:52 -0700 (PDT), George Herold
On Apr 12, 8:46 pm, Jim Thompson <To-Email-Use-The-Envelope-I...@On-My-
Web-Site.com> wrote:
On Fri, 12 Apr 2013 17:43:33 -0700 (PDT), George Herold
[snip]
IC temperature sensors are all crap. Every single one, AFAICT--none
that I know of claims accuracy better than 1 degree C, even the trimmed
ones. Why would that be, if it's so fundamental?
I don't know of a single commercial unit that uses the ratiometric
current method, they're all PTAT's, some trimmed, some not.
Jim, I'm confused (or just ignorant again).. I thought PTAT's used the
current ratio trick.
George H.
[snip]
Different ratio.  PTAT's generate a voltage that indeed is ratio'd by
using two devices of differing areas, thus different current
_densities_.
The Jim Williams' technique, using resistors and adapted by me to use
current sources, uses the same device, "measured" at two carefully
ratio'd currents.
The PTAT effect depends on lots of other variables as well.
I'll dredge thru my files and find one I can show.
                                       ...Jim Thompson
A very old one I designed for Fairchild...
http://www.analog-innovations.com/SED/PTAT_Demo_SED.pdf
Explanation to follow... I'm off to the 5-year-old grandson's ice
hockey practice... he made the team >:-}
                                       ...Jim Thompson

http://www.analog-innovations.com/SED/Killian_Thompson_Age_5.mp4

                                      ...Jim Thompson

So, can we count on you incorporating hockey metaphors into your
posts? Say "deke" and "stick-handle", for starters?

"Keep your stick on the ice!"

George H.

 

[Bill Sloman]

Hi Joe that's neat, I'm a bit confused as to why platinum RTD's were
easier than thermistors?

I've never used a platinum RTD.

They aren't. I think that what Joe Gwinn was saying that he didn't discover24-bit sigma delta A/D converters (mostly they aren't better than 20-bit, but they all produce three eight-bit bytes of output data)until after he'd decided to try the Platinum RTDs.

Pt RTDs give about +0.4% resistance change per degree Kelvin, thermistors about -4%, so they aren't easier to use. Their self-heating isn't usually asbad, and they don't go nuts if you put too much current through them, but it takes even a sophisticated user a while to get conscious of this kind ofproblem.

 

[Jon Kirwan]

<snip>

Hey Jon, I was thinking about this, and it seems like the error is
due to the non-ideality factor (NIF) in the equation. Now I've only
read about the NIF in the context of pn junctions, but wouldn't there
be something similar in a diode connected transistor?

The BJT models are rife with "emission coefficients," which
is what I think you are talking about. In BJTs, my meager
practical experience (small signal BJTs at low frequency and
DC) is they really are very close to 1. Yes, an emission
coefficient, if different than 1, would affect things. But
only the step size. At least, given the simple models I was
discussing earlier.

Now according to Streetman the NIF arises because of carrier
recombination in the transition region. I know when I looked at the
temperature dependence of some pn diodes (maybe 1n4148's?) that the
NIF was much closer to 2.

Diodes quite commonly have emission coefficients > 1. Some as
high as 4 or more, I think. LTspice (I just checked) shows
their single (I'm sure there should be more than one model as
it is a widely sourced part) model having an emission
coefficient of 1.752. Which is, as you say, much closer to 2
than to 1.

So then what transitors would have NIF's close to 1?
I wonder if they list NIF's in the spice models?

Yeah, they do. Forward, reverse, high current, low current,
upside down, ... Go into LTspice, then access the Help.
Search on Q. The just look down the list for emission
coefficients. 5 of them pop up right away. If there is a
modeled diode anywhere, backwards or forwards, there is a
separate emission coefficient for it, I think. Almost
gives Rube Goldberg a run for his money.

If you see step size variations from part to part, but
consistent within a given part, I'd tend to lump that one
onto the simple emission coefficient. It's a quick fix. But
step size variations over T on the same part? There is
something else involved... or the single emission coefficient
isn't enough... which is about the same problem.

It all comes down to teasing out what you can assign to a
priori physics and remove from the modeling by applying a
better physical model and what you are stuck gluing onto an
existing model, which uses some mathematical behavior as a
tool but where there is no real physics underneath it. It can
get pretty hairy.

Assuming there is any additional fruit to be had using these
as temperature sensors, better than the simple models
provide, I think means delving into the deeper nature of the
parts and not getting hung up tinkering on the high level
modeling side. It's like the scenario where you are using a
simple, small-angle pendulum model to understand the behavior
of pendulums made by children. You can argue all day about
tinkering with pasted on parameters to make the model match
experience, but unless you delve deeper into how the kids are
building them and see they are using different sized holes
(and different diameter rods rocking in them) and develop a
better more physically based model, those parameters will
never get you that far. You'll be mired in limited degrees of
freedom pasted to the wrong math models for the physics
involved and never extracate yourself (with any remaining
sanity.)

It would be fun to horse around trying to make really good T
sensors from BJTs, if no one else had already scraped that
bowl clean. But I suspect there have been a number of Ph.D.s
already squandered without achieving a great deal more
utility than we already see.

Did you do a single-point calibratation on your probe, by the
way? Or how did you establish where on the absolute scale
things were at?

Jon

 

[[email protected]]

I was pricing some 16 bit SPI DAC's and I thought 55 bucks for an 8 pin
chip was a little steep. So I can imagine what a 24 bit ADC cost...

We're paying around a buck for eight channel, 24-bit, differential
DACs and about a buck-fifty for four channel ADCs with input
diagnostics. Delta-sigma audio stuff is really cheap.

 

[Bill Sloman]

You, Slowman, are just a wee bit "weak" (*) at following schematics.

I've been better than average at the places I've worked.

My drawings are hierarchical by function, AND by matching
requirements. So it is a rarity that a component in one hierarchical
block has to match something in another hierarchical.

I wonder what that was intended to mean ...

(*) Maybe because you aren't capable of recognizing functions at the
device-level ?>:-}

Probably not. When I started out we did a lot of discrete transistor design. The appendices to my 1970 Ph.D. contain quite a few - all pretty horrible, but they worked.

And your oscillator claims are down-right farcical.

You've got no idea what I'm claiming about my oscillator. You and Mike Terrell seem to be off in fantasy-land, imagining what I might be planning on putting together, and being rude about the products of your own diseased imaginations.

 

[Jamie]

We're paying around a buck for eight channel, 24-bit, differential
DACs and about a buck-fifty for four channel ADCs with input
diagnostics. Delta-sigma audio stuff is really cheap.

I don't think I need 24 bits. I will be returning to work tomorrow and
need to call the person that needed this device to get more details.

This is going to be a hand held device that needs to generate a clean
and high res output analog signal from a laptop's USB port. The laptop
is going to define the generated data, also the unit needs 2 outputs and
2 inputs.. This is all going to terminate to a modular plug.

This will be a QC calibration tool so we must make sure we are using
a good DAC. I think I'll be using a ladder dac, I see issues using DS types.

Jamie

 

[George Herold]

The BJT models are rife with "emission coefficients," which
is what I think you are talking about. In BJTs, my meager
practical experience (small signal BJTs at low frequency and
DC) is they really are very close to 1. Yes, an emission
coefficient, if different than 1, would affect things. But
only the step size. At least, given the simple models I was
discussing earlier.


Diodes quite commonly have emission coefficients > 1. Some as
high as 4 or more, I think. LTspice (I just checked) shows
their single (I'm sure there should be more than one model as
it is a widely sourced part) model having an emission
coefficient of 1.752. Which is, as you say, much closer to 2
than to 1.

Hmm, if emission coefficients go as high as four then it is somehow
different from the non-ideality factor which is between 1 and 2.

Yeah, they do. Forward, reverse, high current, low current,
upside down, ... Go into LTspice, then access the Help.
Search on Q. The just look down the list for emission
coefficients. 5 of them pop up right away. If there is a
modeled diode anywhere, backwards or forwards, there is a
separate emission coefficient for it, I think. Almost
gives Rube Goldberg a run for his money.

If you see step size variations from part to part, but
consistent within a given part, I'd tend to lump that one
onto the simple emission coefficient. It's a quick fix. But
step size variations over T on the same part? There is
something else involved... or the single emission coefficient
isn't enough... which is about the same problem.

It all comes down to teasing out what you can assign to a
priori physics and remove from the modeling by applying a
better physical model and what you are stuck gluing onto an
existing model, which uses some mathematical behavior as a
tool but where there is no real physics underneath it. It can
get pretty hairy.

Assuming there is any additional fruit to be had using these
as temperature sensors, better than the simple models
provide, I think means delving into the deeper nature of the
parts and not getting hung up tinkering on the high level
modeling side. It's like the scenario where you are using a
simple, small-angle pendulum model to understand the behavior
of pendulums made by children. You can argue all day about
tinkering with pasted on parameters to make the model match
experience, but unless you delve deeper into how the kids are
building them and see they are using different sized holes
(and different diameter rods rocking in them) and develop a
better more physically based model, those parameters will
never get you that far. You'll be mired in limited degrees of
freedom pasted to the wrong math models for the physics
involved and never extracate yourself (with any remaining
sanity.)

Grin, sure I'm all about models being close to the real physics.

It would be fun to horse around trying to make really good T
sensors from BJTs, if no one else had already scraped that
bowl clean. But I suspect there have been a number of Ph.D.s
already squandered without achieving a great deal more
utility than we already see.

Did you do a single-point calibratation on your probe, by the
way? Or how did you establish where on the absolute scale
things were at?

For the single point calibration, I measured three diodes, Vf (at
10uA) vs T. The three diodes were picked to have high, average, and
low forward voltages at room temp. I then found that the voltage
difference as a function of temp was mostly linear with T from 77 to
400K. So with a standard curve*, I just measure Vf and T and do a
linear correction on all the points.
The temp points are all based on a ~$400 lakeshore calibrated diode.

George H.

*the standard curve is the data from the middle diode.

 

[George Herold]

RTDs are very accurate and stable and are usable over a wide temp range, things
that termistors aren't. Thinfilm platinum RTDs are fairly cheap, too, much
cheaper than a thermistor of similar accuracy.

RTDs are easy to linearize, too, in hardware or software. I have somewhere
around here a dual-opamp circuit that does 3-wire RTD conditioning with
linearization. In software, a bit of second-order correction is usually all you
need... just a few lines even in assembly.

Thermistors can be good over a narrow temp range where you want a lot of signal,

Yeah I've been using them near room temp. Diode laser and the
permenant magnet for an NMR. The magnet has a nice trick, you leave
the thermal loop open til you start to use it, and then you just ask
the loop to keep the T whereever it is. And you adjust the nmr
frequency to match the magnet. It harldy needs calibrated
thermistors.

George H.

 

[Jon Kirwan]

Hmm, if emission coefficients go as high as four then it is somehow
different from the non-ideality factor which is between 1 and 2.

They go as high as needed for diodes. For BJTs, they are
almost always very close to 1, at least with small signal
types I'm more familiar with, testing out against hand
calculuations. I think they are the same thing, as I've seen
both terms used interchangeably and I don't know of another
meaning for the term you are using.

I think you are just talking about this:

Ic = Is*(e^(Vbe/[n*k*T/q]) - 1)

With 'n' being your ideality thing. Elsewhere, that's just
called the emission coefficient. You can see it called
'ideality' here, under the "Shockley diode equation"
subheading:

http://en.wikipedia.org/wiki/Diode#Shockley_diode_equation

But if you read that paragraph, you will also see it called
what I call it, as well. So I really do think we are talking
about the exact same thing. If not, you need to show me the
equation you are talking about and the factor in it.

The emission coefficient isn't just about recombination, by
the way. It hides several things under a single umbrella,
including carrier generation.

And as if that wasn't enough, things are dramatically
different at different currents, as well. At low currents,
you have recombination of carriers AT THE SURFACE, which must
be accounted for. You have recombination of carriers in the
emitter base space charge layer. And you have the formation
of emitter base surface channels. Each of these (only
ideally, though again) do vary with Vbe and with different
tau's and added to the other shockley equation. At higher
currents the injection of minority carriers into the base
region starts to become significant, relative to the majority
carrier concentration. Since space charge neutrality is
maintained, the total majority carrier concentration
increases by the same amount. The non-ideality factor for
this is usually taken to be 2 and is incorporated into the
EM3 model (by Webster, I believe.) This becomes another one
of those exponentials sitting as a divisor on Is.

As mentioned, Rube Goldberg would be jealous. The underlying
physics is NOT modeled, but instead just higher level
exponential math models applied until the behavior starts to
look "mostly okay." Behavior in the corners is modeled very
badly, for example, because the various exponentials don't
reflect the actual shape there very well. It's good enough
for horseshoes, though, and engineers. Not for physicists.

Grin, sure I'm all about models being close to the real physics.

Well, the spice stuff isn't very physical. If you look at the
differential equations you find that there are multiple
exponentials. The -1 term in the Shockley equation just
mentioned is there so that you calculate out a nice Ic=0 when
Vbe=0. The exponentials are only approximations of a reality
that includes MANY taus, not just one. And as you know well,
there is no single exponential that can model the sum of
multiple ones. And there are some non-exponentials that
result from the integral equations that basically no one in
the non-FAB land bothers to use. I got myself deep into this
some time ago when attempting to model the behavior of a
Hamamatsu diode against actual observation over differing die
temperatures (from -70C to about 80C) and over light flux
varying from femptoamps to microamps. The ONLY way I got
results that were close was to go back to applying dopant
concentrations, gradients, and a vague approximation (luckily
not too hard with photodiode I can examine) of the physical
dimensions. I had thought that a 1D model would be good
enough. It wasn't.

I can see why spice doesn't go there.

For the single point calibration, I measured three diodes, Vf (at
10uA) vs T. The three diodes were picked to have high, average, and
low forward voltages at room temp. I then found that the voltage
difference as a function of temp was mostly linear with T from 77 to
400K. So with a standard curve*, I just measure Vf and T and do a
linear correction on all the points.
The temp points are all based on a ~$400 lakeshore calibrated diode.

George H.

*the standard curve is the data from the middle diode.

Okay. So you had a calibrated diode to work with. I guess my
question came from the fact that while the 1X/10X method
provides some basic removal of the Is parameter for a given
device, it does not (so far as my poor awareness permits me)
provide an absolute position which probably has to come from
somewhere external. Apparently, you had that. But if you had
a method to develop an exact calibration down to absolute
zero without any calibration, then I'd love to have learned
of it. (So would a lot of people with expensive freeze
points, I suppose.)

Jon