diodes for breakfast Jan 1, 2006

[kell]

I was putting diodes in a pot of boiling water this morning.
No, it's not a hangover cure.
I was just trying to get an idea of whether led's
have the same temperature coefficient as "regular" diodes.
So I made a series of 1N4148's, put 8 mA through it,
and measured voltage drop at room temperature and boiling.
Did the same, at 8 mA, with a string of cheap pink led's.
I assumed a change from room temp to boiling of 75 degrees centigrade
(didn't have a thermometer).
The 1N4148's and the led's came out to have very nearly
the same tempco, they were both between 2.00 and 2.05
volts per degree centigrade per diode. Also pretty close to the tempco
figures I've seen cited in the literature of about 2.1 or 2.2 mV/C.
I then put 19 mA through the 1N4148's and got a much lower
tempco of 1.56 mV/deg C. I'm wondering if this is because
(A) 19 mA was enough to warm up the diodes, throwing everything off,
(B) tempco really varies that much with current, or
(C) my kitchen experiment was too sloppy.

 

[John Popelish]

I was putting diodes in a pot of boiling water this morning.
No, it's not a hangover cure.
I was just trying to get an idea of whether led's
have the same temperature coefficient as "regular" diodes.
So I made a series of 1N4148's, put 8 mA through it,
and measured voltage drop at room temperature and boiling.
Did the same, at 8 mA, with a string of cheap pink led's.
I assumed a change from room temp to boiling of 75 degrees centigrade
(didn't have a thermometer).
The 1N4148's and the led's came out to have very nearly
the same tempco, they were both between 2.00 and 2.05
volts per degree centigrade per diode. Also pretty close to the tempco
figures I've seen cited in the literature of about 2.1 or 2.2 mV/C.
I then put 19 mA through the 1N4148's and got a much lower
tempco of 1.56 mV/deg C. I'm wondering if this is because
(A) 19 mA was enough to warm up the diodes, throwing everything off,
(B) tempco really varies that much with current, or
(C) my kitchen experiment was too sloppy.

(D) There is an additional resistive drop in series with the diode
junction that has a significant but different tempco than the
junction, but its drop becomes significant only at higher current.

I would try several lower currents to see if the tempco is stable
below some limiting current. Do you insulate the diodes or are they
in contact with the water?

 

[kell]

(D) There is an additional resistive drop in series with the diode
junction that has a significant but different tempco than the
junction, but its drop becomes significant only at higher current.

I would try several lower currents to see if the tempco is stable
below some limiting current. Do you insulate the diodes or are they
in contact with the water?

I just dunked the diodes right in the water. They were soldered
together
in a string with a current limiting resistor, which I kept out of the
water.

From a practical standpoint I don't have to worry, because in my

application
(I'm building a voltage regulator), the diode string will have very
low current, probably no more than a mA.
Thanks for the heads up about series resistance, I didn't think about
that.

 

[Klaus Kragelund]

What about the conductance of the water? Won't that lead to unreliable
results?

Regards

Klaus

Better to place them i your kitchen oven

 

[Winfield Hill]

kell wrote...

I was putting diodes in a pot of boiling water this morning.
No, it's not a hangover cure.
I was just trying to get an idea of whether led's
have the same temperature coefficient as "regular" diodes.
So I made a series of 1N4148's, put 8 mA through it,
and measured voltage drop at room temperature and boiling.
Did the same, at 8 mA, with a string of cheap pink led's.
I assumed a change from room temp to boiling of 75 degrees
centigrade (didn't have a thermometer).
The 1N4148's and the led's came out to have very nearly
the same tempco, they were both between 2.00 and 2.05 (milli)
volts per degree centigrade per diode. ...

A nice simple experiment. It should be noted that the LEDs are
effectively better than the diodes as a reference, because they
have a much higher voltage drop, and hence a much lower % change
with temperature. Also, some people use an LED as the reference
for a current source, using a transistor and resistor, wherein
the transistor's tempco nearly cancels the LED's tempco. Cool.

 

[Winfield Hill]

Winfield Hill wrote...

kell wrote...

A nice simple experiment. It should be noted that the LEDs are
effectively better than the diodes as a reference, because they
have a much higher voltage drop, and hence a much lower % change
with temperature. Also, some people use an LED as the reference
for a current source, using a transistor and resistor, wherein
the transistor's tempco nearly cancels the LED's tempco. Cool.

It should be further noted that the LED zero-tempco current-source
trick can't be done with a stack of diodes to bias the transistor,
which is quite popular and commonly seen, because each diode in
the stack adds another -2mV/C to the tempco. Furthermore, the Vbe
reference trick, wherein two resistors are used with a transistor
to make an arbitrary reference voltage to bias a current-source
transistor, also suffers from a multiplied tempco. The only low-
cost part (or set of parts) that works is one LED. Some, like me,
find it slightly bizzarre to use an LED as a voltage reference,
and shy away from using this trick. Others consider it elegant.

 

[John Popelish]

I just dunked the diodes right in the water. They were soldered
together
in a string with a current limiting resistor, which I kept out of the
water.
application
(I'm building a voltage regulator), the diode string will have very
low current, probably no more than a mA.
Thanks for the heads up about series resistance, I didn't think about
that.

I think you can see the effect of non ideal diode resistance in the
data sheet curve of voltage drop versus current at the current where
the voltage no longer follows a logarithmic shape, but tends toward a
linear one.

For instance, lay a straight edge along th line in figure 1 of:
http://www.junun.org/MarkIII/datasheets/1N4148.pdf
It looks to me that the curve deviates from a log starting at about 10
mA. The ideal diode formula, including the tempco is more likely to
be accurate along the straight (logarithmic) part of that curve.

If the experiment is worth doing, I think you should try to keep the
diodes dry during the test. Either place them in the oven with a
thermometer beside in a glass baking dish, or put them in a test tube
or something similar for the dunk test. This will let you take data
points at much lower current (around the range your application will
use), without having to worry if leakage current through the water is
involved.

 

[Ken Smith]

A nice simple experiment. It should be noted that the LEDs are
effectively better than the diodes as a reference, because they
have a much higher voltage drop, and hence a much lower % change
with temperature. Also, some people use an LED as the reference
for a current source, using a transistor and resistor, wherein
the transistor's tempco nearly cancels the LED's tempco. Cool.

They also make a cool over current clamp:


!
!/ c
---+----!
! !\ e
V !
--- /
! \
! /
! !

 

[Phil Hobbs]

Winfield Hill wrote...



It should be further noted that the LED zero-tempco current-source
trick can't be done with a stack of diodes to bias the transistor,
which is quite popular and commonly seen, because each diode in
the stack adds another -2mV/C to the tempco. Furthermore, the Vbe
reference trick, wherein two resistors are used with a transistor
to make an arbitrary reference voltage to bias a current-source
transistor, also suffers from a multiplied tempco. The only low-
cost part (or set of parts) that works is one LED. Some, like me,
find it slightly bizzarre to use an LED as a voltage reference,
and shy away from using this trick. Others consider it elegant.

The nicest thing about forward-biased diodes as voltage references is that
you can get pretty low noise--subtracting a 0.7V Vbe from 1.7V LED makes a
1-V reference that's a good 15 or 20 dB quieter than a bandgap, at the price
of lower accuracy and higher drift, which is sometimes a big win.

The other thing it would be nice to know is whether the logarithmic slope of
V vs I in the LEDs is closer to kT/e than in the 1N4148s. It's one of the
niggling annoyances of electronics that diodes don't obey the diode equation,
but diode-connected transistors do!

Cheers,

Phil Hobbs

 

[Bob Monsen]

A nice simple experiment. It should be noted that the LEDs are
effectively better than the diodes as a reference, because they
have a much higher voltage drop, and hence a much lower % change
with temperature. Also, some people use an LED as the reference
for a current source, using a transistor and resistor, wherein
the transistor's tempco nearly cancels the LED's tempco. Cool.

Well, I of course rush off to simulate this, and run into a problem: The
LED models I have all exhibit a positive Vf tempco. I tried the standard
ltspice LED models, plus some other models I picked up from the ltspice
group on yahoo.

Here is one of the models (although they all exhibit the same symptom.)

*Typ RED GaAs LED: Vf=1.7V Vr=4V If=40mA trr=3uS
..MODEL RED_LED D (IS=93.2P RS=42M N=3.73 BV=4 IBV=10U
+ CJO=2.97P VJ=.75 M=.333 TT=4.32U
+ Vpk=1.7V Iave=40mA type=LED)

I wonder if there is some simple correction I can make to the model?

--
Regards,
Bob Monsen

Zero is the number of objects that satisfy a condition that is never
satisfied. But as never means "in no case", I do not see that any progress
has been made.
- Poincare

 

[Robert Latest]

["Followup-To:" header set to sci.electronics.design.]
On 1 Jan 2006 09:06:10 -0800,

transistor, also suffers from a multiplied tempco. The only low-
cost part (or set of parts) that works is one LED. Some, like me,
find it slightly bizzarre to use an LED as a voltage reference,
and shy away from using this trick. Others consider it elegant.

I've done it many times and I really like it. Also, for red LEDs, the
voltage drop across the current-setting resistor is approx. 1V, making
the math really easy ;-)

robert

 

[Bob Masta]

The nicest thing about forward-biased diodes as voltage references is that
you can get pretty low noise--subtracting a 0.7V Vbe from 1.7V LED makes a
1-V reference that's a good 15 or 20 dB quieter than a bandgap, at the price
of lower accuracy and higher drift, which is sometimes a big win.

The other thing it would be nice to know is whether the logarithmic slope of
V vs I in the LEDs is closer to kT/e than in the 1N4148s. It's one of the
niggling annoyances of electronics that diodes don't obey the diode equation,
but diode-connected transistors do!

Another nice thing about LEDs instead of Zeners is that the knee
is *much* sharper at low currents.

Best regards,


Bob Masta
dqatechATdaqartaDOTcom

D A Q A R T A
Data AcQuisition And Real-Time Analysis
www.daqarta.com
Home of DaqGen, the FREEWARE signal generator

 

[Winfield Hill]

Bob Masta wrote...

Another nice thing about LEDs instead of Zeners is that the knee
is *much* sharper at low currents.

http://www.picovolt.com/win/elec/comp/diode/diode-curves.html

Yes, here again (it's been a few years since I posted this link)
are some old diode, transistor and LED forward-voltage plots I
made over a 1pA to 10mA range. Sorry they cut off above 1.25 V,
but you can see the red LED is much sharper (110mV/decade) than
Ebers-Moll predicts (60mV/decade) over the 0.1nA to 0.2uA range.

I'll have to install Corel's Quattro-Pro spreadsheet on my new
"updated" WinXp computer, so I can look at the orginal file to
see what the LED did at higher currents.

 

[Rich Grise]

kell wrote...

A nice simple experiment. It should be noted that the LEDs are
effectively better than the diodes as a reference, because they
have a much higher voltage drop, and hence a much lower % change
with temperature. Also, some people use an LED as the reference
for a current source, using a transistor and resistor, wherein
the transistor's tempco nearly cancels the LED's tempco. Cool.

Well, (a), I wonder about how photosensitivity might affect these
readings and (b), I wonder about nonlinearity - i.e., did kell
also dip them into a container of ice water and check their cold
Vf? It might not be a straight line, you know.

Of course, a real thermometer for ambient would help a lot.

Cheers!
Rich

 

[Rich Grise]

Winfield Hill wrote...

It should be further noted that the LED zero-tempco current-source
trick can't be done with a stack of diodes to bias the transistor,
which is quite popular and commonly seen, because each diode in
the stack adds another -2mV/C to the tempco. Furthermore, the Vbe
reference trick, wherein two resistors are used with a transistor
to make an arbitrary reference voltage to bias a current-source
transistor, also suffers from a multiplied tempco. The only low-
cost part (or set of parts) that works is one LED. Some, like me,
find it slightly bizzarre to use an LED as a voltage reference,
and shy away from using this trick. Others consider it elegant.

Get a black magic marker. ;-)

Cheers!
Rich

 

[kell]

I was putting diodes in a pot of boiling water this morning.
No, it's not a hangover cure.
I was just trying to get an idea of whether led's
have the same temperature coefficient as "regular" diodes.
So I made a series of 1N4148's, put 8 mA through it,
and measured voltage drop at room temperature and boiling.
Did the same, at 8 mA, with a string of cheap pink led's.
I assumed a change from room temp to boiling of 75 degrees centigrade
(didn't have a thermometer).
The 1N4148's and the led's came out to have very nearly
the same tempco, they were both between 2.00 and 2.05
volts per degree centigrade per diode. Also pretty close to the tempco
figures I've seen cited in the literature of about 2.1 or 2.2 mV/C.
I then put 19 mA through the 1N4148's and got a much lower
tempco of 1.56 mV/deg C. I'm wondering if this is because
(A) 19 mA was enough to warm up the diodes, throwing everything off,
(B) tempco really varies that much with current, or
(C) my kitchen experiment was too sloppy.

I tried that experiment again and couldn't get the led's to settle down
to a steady voltage
when I had them in the water. I decided to try a different way of
testing it.
I put a series of led's in parallel with a series of an equal number of
1N400X diodes,
tied to a common ground at the bottom of the strings with a small
current fed to
the top of each string. I attached one lead from a voltmeter to the
top of each string
of diodes. Now when I heated the strings of diodes with a hair drier,
IF the led's had
the same tempco as the diodes, the voltage on the meter would stay
unchanged,
because the two strings, having an equal number of elements, would have
equal temperature-related drops.
As it turned out, the voltage did change.
Now to find out the temperature coefficient of the led's, I need to
find a ratio between
the number of led's and diodes where the tempcos cancel out. For
example, if 7 led's
balance out with 6 1N400X diodes and show no voltage change when heated
with the
hair drier, I could figure the led tempco would be six sevenths the
diode tempco, which
is known.
This is about the most reliable way I could think of to really pin down
the led tempco;
if I put them in the oven it would just be a mess.

 

[Winfield Hill]

Winfield Hill wrote...

Bob Masta wrote...

http://www.picovolt.com/win/elec/comp/diode/diode-curves.html

Yes, here again (it's been a few years since I posted this link)
are some old diode, transistor and LED forward-voltage plots I
made over a 1pA to 10mA range. Sorry they cut off above 1.25 V,
but you can see the red LED is much sharper (110mV/decade) than
Ebers-Moll predicts (60mV/decade) over the 0.1nA to 0.2uA range.

I'll have to install Corel's Quattro-Pro spreadsheet on my new
"updated" WinXp computer, so I can look at the orginal file to
see what the LED did at higher currents.

The red LED shows a steady +120mV/decade I-V slope over four
decades from 0.1nA to 1uA, where Vf is about 1.34V, then the
I-V slope *decreases* to +60mV/decade for four more decades
to 10mA, the highest value I measured.