looking for high value resistors

[alan]

Hello people

I am looking for high value resistors (1G, 10G) that have these
additional properties:

low capacitance (I can't find any specs for this)
reasonably linear resistance vs voltage
low excess noise
Easily available in the US
low tempco
(whatever other characteristic might be important)

What type of resistor would have the best properties?

Using this for a high gain, medium speed (<1kHz) current amplifier for
STM and stuff.

 

[Country Loon]

Hello people

I am looking for high value resistors (1G, 10G) that have these
additional properties:

low capacitance (I can't find any specs for this)
reasonably linear resistance vs voltage
low excess noise
Easily available in the US
low tempco
(whatever other characteristic might be important)

What type of resistor would have the best properties?

Using this for a high gain, medium speed (<1kHz) current amplifier for
STM and stuff.

Never heard of such a resistor but if it exists it is bound to be noisy as
the noise is something like

Vn=sqrt(kTRB)

and R is the resistance.


Tom

 

[Winfield Hill]

Country Loon wrote...

alan wrote ...

I am looking for high value resistors (1G, 10G) that have these
additional properties: [ snip ] Using this for a high gain,
medium speed (<1kHz) current amplifier for STM and stuff.

Never heard of such a resistor but if it exists it is bound to
be noisy as the noise is something like Vn=sqrt(kTRB) and R
is the resistance.

Sorry, high-value resistors are much better for trans-resistance
amplifiers. That's because of Ohm's law, i_n = v_n / R, so the
effective Johnson current noise seen at the input node due to the
feedback resistor is sqrt(kTB/R), which gets better for large R.
That's the right way to use the Johnson-noise math for currents.

 

[Martin Brown]

I am looking for high value resistors (1G, 10G) that have these
additional properties:

low capacitance (I can't find any specs for this)
reasonably linear resistance vs voltage
low excess noise
Easily available in the US
low tempco
(whatever other characteristic might be important)

What type of resistor would have the best properties?

Using this for a high gain, medium speed (<1kHz) current amplifier for
STM and stuff.

They are used in Faraday detector current amplifiers for mass
spectrometry. ISTR they are something of a variable product, very tricky
to make and you may need to select from a batch to get adequate
specimens. Could be even more of an issue if you intend to hit them with
large voltages - some have a tendency to hold charge like a battery.

Circuit capacitance(s) can be an issue at 1G & 10G. Usually fiddled out
downstream with an empirical compensation circuit.

Regards,

 

[Daniel Haude]

On Wed, 13 Oct 2004 03:38:24 -0700,

I am looking for high value resistors (1G, 10G) that have these
additional properties:

We always buy them from ELTEC. Expensive stuff, but very good, very low
capacitance (the chip-only type).

low capacitance (I can't find any specs for this)

ELTEC goes to great lengths in their data sheets explaining why they don't
want to give specs, but if you put a pistol against their head they'll say
something like .3pF. With 1G that gives you nominally about 6kHz BW.

reasonably linear resistance vs voltage

No can do. Haven't found any good ones for that, and with Eltec's, all
bets are off beyond 1V (!). That's what they say and I've found it
confirmed.

Using this for a high gain, medium speed (<1kHz) current amplifier for
STM and stuff.

Try to go to 100M instead of GOhm range. Noise gets a bit worse, but you
have less BW and nonlinearity issues. Of course all this depends on the
current you're going to measure.

Good luck,
--Daniel

 

[Frank Miles]

Hello people

I am looking for high value resistors (1G, 10G) that have these
additional properties:

low capacitance (I can't find any specs for this)
reasonably linear resistance vs voltage
low excess noise
Easily available in the US
low tempco
(whatever other characteristic might be important)

What type of resistor would have the best properties?

Using this for a high gain, medium speed (<1kHz) current amplifier for
STM and stuff.

I don't know enough about your application -- it may well prohibit it --
but have you considered using a feedback capacitance? If you can get
away with periodically resetting it, you can make a much cleaner small
feedback capacitor than you can buy in such an extremely high resistance.
Furthermore, if you need to push bandwidth, resistors are really nasty
because the parasitic C in this G-ohm class is far from a single lumped
capacitance -- its more distributed (at least those than I have measured).
This comes from the serpentine pattern of the element.

When we have had to use such values, we've usually gotten them from IMS:
http://www.ims-resistors.com/

HTH...

-frank
--

 

[alan]

On Wed, 13 Oct 2004 03:38:24 -0700,



We always buy them from ELTEC. Expensive stuff, but very good, very low
capacitance (the chip-only type).

Their web page seems to not exist at the moment. Where can I get more
info? And why would you want a chip type? I thought you would want a
bigger size resistor.

ELTEC goes to great lengths in their data sheets explaining why they don't
want to give specs, but if you put a pistol against their head they'll say
something like .3pF. With 1G that gives you nominally about 6kHz BW.

I found a glass encapsulated Welwyn 1G resistor in our lab, and I
measured the cap to be .04pF. Maybe I should be more careful about
measuring? Coz this value seems way low compared to yours. I found
another Dale/Vishay 1G resistor that looks like a "regular" 1G resistor,
and that has a cap of about .2pF.

No can do. Haven't found any good ones for that, and with Eltec's, all
bets are off beyond 1V (!). That's what they say and I've found it
confirmed.

I thought resistors designed for high voltage were supposed to be good
about this. The Welwyn ones are less than a few ppm per volt.

Try to go to 100M instead of GOhm range. Noise gets a bit worse, but you
have less BW and nonlinearity issues. Of course all this depends on the
current you're going to measure.

ACtually, I want minimal noise at a few hundred Hz to take good spectra.
The setpoint is like 100pA. The best I can do with resistors is
actually run 10G and divide the output by 10. Frank mentioned using
capacitors for feedback, which is an idea I have in the back of my head,
but I'm not sure yet how to implement it while still keeping the DC part
of the signal.

Now that I think about it, you can't ever cheat the input noise of
whatever input stage you use. In this case, it's the LT1793, with input
noise 1fA/rtHz, voltage noise is 8nV/rtHz*200Hz*(30 to 80pF input
cap)=.8fA/rtHz. Unless there is a better op-amp. (And no, I don't
really want to deal with discrete transistors unless it is really really
simple)

Noise of 1G resistor is 4fA/rtHz and 10G is 1.3fA/rtHz.

 

[Winfield Hill]

alan wrote...

Their web page seems to not exist at the moment. Where can I get more
info? And why would you want a chip type? I thought you would want a
bigger size resistor.


I found a glass encapsulated Welwyn 1G resistor in our lab, and I
measured the cap to be .04pF. Maybe I should be more careful about
measuring? Coz this value seems way low compared to yours. I found
another Dale/Vishay 1G resistor that looks like a "regular" 1G resistor,
and that has a cap of about .2pF.

Your measurements are very much in line with my own, which were made
very carefully!

I thought resistors designed for high voltage were supposed to be good
about this. The Welwyn ones are less than a few ppm per volt.

That's correct, I'm not sure what Daniel is talking about.

Actually, I want minimal noise at a few hundred Hz to take good spectra.
The setpoint is like 100pA. The best I can do with resistors is
actually run 10G and divide the output by 10. Frank mentioned using
capacitors for feedback, which is an idea I have in the back of my head,
but I'm not sure yet how to implement it while still keeping the DC part
of the signal.

Dan suggests 100M instead of 1000G, and I suggest 100G instead of 10G,
if you can get it...

Now that I think about it, you can't ever cheat the input noise of
whatever input stage you use. In this case, it's the LT1793, with input
noise 1fA/rtHz, voltage noise is 8nV/rtHz*200Hz*(30 to 80pF input cap)
=0.8fA/rtHz. Unless there is a better op-amp.

8nV isn't very good. The LT1792 is 6nV or less, and the AD743 is
3.6nV typical (one takes their worst-case value with a grain of salt).
The opamp current noise can be reduced by operating it at lower supply
voltage and / or by cooling.

(And no, I don't really want to deal with discrete transistors unless
it is really really simple).

The circuitry is simple, getting the JFETs isn't so simple.

Noise of 1G resistor is 4fA/rtHz and 10G is 1.3fA/rtHz.

Right, and 100G is 0.4fA per root Hz.

 

[Daniel Haude]

On Fri, 15 Oct 2004 04:27:08 -0700,

I found a glass encapsulated Welwyn 1G resistor in our lab, and I
measured the cap to be .04pF. Maybe I should be more careful about
measuring? Coz this value seems way low compared to yours. I found
another Dale/Vishay 1G resistor that looks like a "regular" 1G resistor,
and that has a cap of about .2pF.

Sorry, I lost a digit there: .03pF is more like it.

--Daniel

 

[Daniel Haude]

On 15 Oct 2004 05:59:52 -0700,

That's correct, I'm not sure what Daniel is talking about.

Just about what ELTEC says in their data sheet. Maybe I should switch
suppliers.

Dan suggests 100M instead of 1000G, and I suggest 100G instead of 10G,
if you can get it...

If you want to have "a few hundred Hz" of bandwith with a 100G feedback
resistance, you must also do frequency compensation because the 3dB point
of such a circuit (w/ .02pF stray C) will be at about 40 Hz. That's the
infamous "R-C-R trick" or "Keithley circuit" which has been discussed at
great detail in this group earlier this year.

But to the point of STM/STS: I don't think you're very well off with an
100G STM amplifier because it limits the max currnt at which you can do
any measurement to about 100pA -- and although I made a lot of measurments
on semiconductors with such and lower currents, I still like to be able to
fry my tip with several nA against a metal surface. A low-gain amplifier
will of course impose a higher low limit of usable stabilization currents.
Much of this depends on your specific application.

With STS, all boils down to how much time you want to spend on your
measurement. You can use a high-gain, low-noise, low-bandwidth amplifier,
but that'll require low modulation frequencies and in turn long lock-in
integration times. A low-gain amp will yield a higher bandwidth allowing a
higher modulation frequency, but at the cost of more noise -- which you'll
have to filter again with a long time constant. I haven't done the math; I
wonder if the two cases (10G @ 200Hz vs. 1G @ 2kHz) are equivalent. If so,
use the smaller R and have a more versatile amp.

The most important point in hogh-resolution STS is to keep the noise
across the gap low, because this you can't filter with the lock-in.

For the < 100ueV energy resolution that can be achieved with a jellybean
transresistance amplifier comprising a OPA627BM and a 1G feedback resistor
looking into almost 1 nF of input capacitance, check out an upcoming
article in the Review of Scientific Instrumentation (I can give you the
exact rev once the issue is out in November).

--Daniel

 

[alan]

On 15 Oct 2004 05:59:52 -0700,



Just about what ELTEC says in their data sheet. Maybe I should switch
suppliers.

Do you have a capacitance bridge or something?

If you want to have "a few hundred Hz" of bandwith with a 100G feedback
resistance, you must also do frequency compensation because the 3dB point
of such a circuit (w/ .02pF stray C) will be at about 40 Hz. That's the
infamous "R-C-R trick" or "Keithley circuit" which has been discussed at
great detail in this group earlier this year.

yeah, I should look into that sometime

But to the point of STM/STS: I don't think you're very well off with an
100G STM amplifier because it limits the max currnt at which you can do
any measurement to about 100pA -- and although I made a lot of measurments
on semiconductors with such and lower currents, I still like to be able to
fry my tip with several nA against a metal surface. A low-gain amplifier
will of course impose a higher low limit of usable stabilization currents.
Much of this depends on your specific application.

just switch the gain.

With STS, all boils down to how much time you want to spend on your
measurement. You can use a high-gain, low-noise, low-bandwidth amplifier,
but that'll require low modulation frequencies and in turn long lock-in
integration times. A low-gain amp will yield a higher bandwidth allowing a
higher modulation frequency, but at the cost of more noise -- which you'll
have to filter again with a long time constant. I haven't done the math; I
wonder if the two cases (10G @ 200Hz vs. 1G @ 2kHz) are equivalent. If so,
use the smaller R and have a more versatile amp.

The time you want to spend per spectra sets how wide you have to open
the badwidth on the lock-in.
Given that you have a fixed setpoint current, you're better off running
the higher gain since the spectral noise density is lower and thus the
S/N is better. I don't understand what you said above about using a
higher bandwidth and then filtering it down. Isn't that the same as
using a lower bandwidth in the first place?

For the < 100ueV energy resolution that can be achieved with a jellybean
transresistance amplifier comprising a OPA627BM and a 1G feedback resistor
looking into almost 1 nF of input capacitance, check out an upcoming
article in the Review of Scientific Instrumentation (I can give you the
exact rev once the issue is out in November).

heh, you need to have better wiring

 

[alan]

alan wrote...

8nV isn't very good. The LT1792 is 6nV or less, and the AD743 is
3.6nV typical (one takes their worst-case value with a grain of salt).
The opamp current noise can be reduced by operating it at lower supply
voltage and / or by cooling.

assuming I'm not going to cool the op-amp, it looks like I can only
reduce the AD743's current noise by approx 4x. That's still not as good
as the LT1793. The LT1792's current noise seems to be worse as well. At
200Hz and 80pF, the LT1793's voltage noise is still pretty good.

The circuitry is simple, getting the JFETs isn't so simple.

ok, what's the circuit?

Right, and 100G is 0.4fA per root Hz.

what, am I supposed to run this in a 100V amplifier and divide the
output by 10?

 

[alan]

I found a glass encapsulated Welwyn 1G resistor in our lab, and I
measured the cap to be .04pF. Maybe I should be more careful about
measuring? Coz this value seems way low compared to yours. I found
another Dale/Vishay 1G resistor that looks like a "regular" 1G resistor,
and that has a cap of about .2pF.

Actually, I just looked at Welwyn's web page, and for what I believe to
be the same resistor (it's their only glass coated one) they quote a
capacitance of .2pF. So I don't know what's up with that.

They also have some planar resistors that they claim to be low
capacitance, but they won't give any values

 

[Winfield Hill]

alan wrote...

what, am I supposed to run this in a 100V amplifier and
divide the output by 10?

That's what Keithley does in some of their models. :>)

 

[Daniel Haude]

On Mon, 18 Oct 2004 04:05:52 -0700,

just switch the gain.

Hah! Not so easy to do in the first stage. What would you propose? I
already threw the reed relay out of my amp.

The time you want to spend per spectra sets how wide you have to open
the badwidth on the lock-in.
Given that you have a fixed setpoint current, you're better off running
the higher gain since the spectral noise density is lower and thus the
S/N is better.

Yes, but a higher gain (as long as we're only talking about the standard,
two-component transresistance amp) automatically lowers your bandwidth,
forcing you to use low mod frequencies and long integration times.

I don't understand what you said above about using a
higher bandwidth and then filtering it down.

A higher-bandwidth, lower-gain amplifier permits higher mod frequencies at
the cost of more noise, still requiring you to use long integration.

Isn't that the same as
using a lower bandwidth in the first place?

That's what I said -- if in fact it is the same (which I suspect but don't
know) a low-gain high-BW amplifier has some advantages: higher currents
for tip preparation, higher scan speeds in topo mode, easier to obtain
feedback resistors. My hunch is that the optimum resistance, all things
considered, is between 100 and 1000M.

heh, you need to have better wiring

In this setup there's no way to get to the STM through less than seven
meters of wiring...

--Daniel

 

[Guy Macon]

Hah! Not so easy to do in the first stage. What would you propose? I
already threw the reed relay out of my amp.

Why?

 

[alan]

On Mon, 18 Oct 2004 04:05:52 -0700,
in Msg. <[email protected]>


Hah! Not so easy to do in the first stage. What would you propose?

A jumper. Coz otherwise I'd turn the gain knob on my Keithley 427.

Yes, but a higher gain (as long as we're only talking about the standard,
two-component transresistance amp) automatically lowers your bandwidth,
forcing you to use low mod frequencies and long integration times.


A higher-bandwidth, lower-gain amplifier permits higher mod frequencies at
the cost of more noise, still requiring you to use long integration.

I don't know how you take your spectra or lock-in measurements, but I
don't think the mod frequency makes a difference in how "fast" you can
take data, given that it is above some minimal level. The bandwidth on
the lock in determines this, because it sets how fast the output of the
lock in can fluctuate. Given that this bandwidth is now fixed, the low
gain amplifier is always worse since the spectral noise density is
higher, so the lock-in always ends up integrating more noise. The only
way to get less noise is to decrease the bandwidth, but that will make
your measurement slower.

That's what I said -- if in fact it is the same (which I suspect but don't
know) a low-gain high-BW amplifier has some advantages: higher currents

No, the signal to noise goes as sqrt(gain) for a given measurement speed.

I just realised that maybe your situation is that you need to take data
so fast that you use almost the entire amplifier bandwidth for the
lockin bandwidth? In that case, I think the procedure should be:

Determine how fast you want to take spectra. This sets the lockin time
constant/bandwidth.
Set the gain as high as possible such that you still have bandwidth to
do this. This maximizes the signal to noise.
Measure and publish.

In this setup there's no way to get to the STM through less than seven
meters of wiring...

that's still 140pF per meter...

 

[Daniel Haude]

On Mon, 18 Oct 2004 13:39:11 -0700,

I don't know how you take your spectra or lock-in measurements, but I
don't think the mod frequency makes a difference in how "fast" you can
take data, given that it is above some minimal level. The bandwidth on
the lock in determines this, because it sets how fast the output of the
lock in can fluctuate. Given that this bandwidth is now fixed, the low
gain amplifier is always worse since the spectral noise density is
higher, so the lock-in always ends up integrating more noise.
Correct.

The only
way to get less noise is to decrease the bandwidth, but that will make
your measurement slower.
Correct.

No, the signal to noise goes as sqrt(gain) for a given measurement speed.
Correct.

I just realised that maybe your situation is that you need to take data
so fast that you use almost the entire amplifier bandwidth for the
lockin bandwidth? In that case, I think the procedure should be:

Determine how fast you want to take spectra. This sets the lockin time
constant/bandwidth.
Set the gain as high as possible such that you still have bandwidth to
do this. This maximizes the signal to noise.

Correct. This way I've ended up at 1G.

Measure and publish.

Ummm... correct.

that's still 140pF per meter...

Correct, but do you know of any UHV compatible coax with less than 1mm
diameter that has significantly lower capacitance?

--Daniel

 

[alan]

On Mon, 18 Oct 2004 13:39:11 -0700,



Correct, but do you know of any UHV compatible coax with less than 1mm
diameter that has significantly lower capacitance?

if it has to be less than 1mm the entire 7 meters, then desert
cryogenics has one that is 94pF/m.