Forums

Little thermal thing

Started by George Herold December 11, 2018
A 0.25" x 1.3 " pcb holds a sensor, heater (300 ohm, 1206 or bigger)
 and diode temp. sensor.
I first made it with a common ground for the heater and diode,
but there was to much offset from the heater current*
and I had to add a third wire to sense local ground for the diode.
Which worked great.  About a 2 sec thermal time constant.  

When I cut the ground plane (razor blade on thermals)
thermal time constant was 40 seconds.  

1/2 oz copper vs 1/16" fr4.

But if there are no expected thermal changes on the 
60 second time frame... is there some way faster 
(thermally) is better?   The pcb is wrapped in foam and 
cooled.. 
Just float everyone and be slow what's the problem? 

 






*I think I could mostly fix this, by changing the 'ground' 
of the current source feeding the diode.  
On 12/11/18 8:25 PM, George Herold wrote:
> A 0.25" x 1.3 " pcb holds a sensor, heater (300 ohm, 1206 or bigger) > and diode temp. sensor. > I first made it with a common ground for the heater and diode, > but there was to much offset from the heater current* > and I had to add a third wire to sense local ground for the diode. > Which worked great. About a 2 sec thermal time constant. > > When I cut the ground plane (razor blade on thermals) > thermal time constant was 40 seconds. > > 1/2 oz copper vs 1/16" fr4. > > But if there are no expected thermal changes on the > 60 second time frame... is there some way faster > (thermally) is better? The pcb is wrapped in foam and > cooled..
Two-ounce copper and closer placement.
> Just float everyone and be slow what's the problem?
It trashes your thermal forcing rejection. If your bandwidth goes down by 20x, so does your forcing rejection at low frequencies, unless you're prepared to do fancy lead-lag things that are liable to become unstable if you have a void in your thermal paste, for instance. I have a current product that uses an 0603 thermistor mounted to a baby board in intimate contact with the cold plate. Crucially, the thermistor is on the same side of the board as the cold plate, and shares a common ground plane. The lateral clearance is about 0.040". (I could use a few in series if the cold plate were large enough to worry about gradients.) Its thermal response is dramatically faster than a glass bead thermistor and silver epoxy, and besides, it's way way cheaper and more repeatable. Cheers Phil Hobbs -- Dr Philip C D Hobbs Principal Consultant ElectroOptical Innovations LLC / Hobbs ElectroOptics Optics, Electro-optics, Photonics, Analog Electronics Briarcliff Manor NY 10510 http://electrooptical.net http://hobbs-eo.com
On Tuesday, December 11, 2018 at 8:40:59 PM UTC-5, Phil Hobbs wrote:
> On 12/11/18 8:25 PM, George Herold wrote: > > A 0.25" x 1.3 " pcb holds a sensor, heater (300 ohm, 1206 or bigger) > > and diode temp. sensor. > > I first made it with a common ground for the heater and diode, > > but there was to much offset from the heater current* > > and I had to add a third wire to sense local ground for the diode. > > Which worked great. About a 2 sec thermal time constant. > > > > When I cut the ground plane (razor blade on thermals) > > thermal time constant was 40 seconds. > > > > 1/2 oz copper vs 1/16" fr4. > > > > But if there are no expected thermal changes on the > > 60 second time frame... is there some way faster > > (thermally) is better? The pcb is wrapped in foam and > > cooled.. > > Two-ounce copper and closer placement. > > > Just float everyone and be slow what's the problem? > > It trashes your thermal forcing rejection. If your bandwidth goes down > by 20x, so does your forcing rejection at low frequencies, unless you're > prepared to do fancy lead-lag things that are liable to become unstable > if you have a void in your thermal paste, for instance.
OK I assume that 'thermal forcing rejection' is the same or related to how much gain (delta(power)/delta(temp)) I can use. It's certainly true that with the faster common ground plane I can crank up the gain by a factor of ~20 also. (There's a V^2 term in there... I'm using the term 'gain' a bit loosely.) So my response is trashed (reduced) at all frequency ranges... But if there are no big temperature or load changes so what?
> > I have a current product that uses an 0603 thermistor mounted to a baby > board in intimate contact with the cold plate. Crucially, the > thermistor is on the same side of the board as the cold plate, and > shares a common ground plane. The lateral clearance is about 0.040". > (I could use a few in series if the cold plate were large enough to > worry about gradients.) > > Its thermal response is dramatically faster than a glass bead thermistor > and silver epoxy, and besides, it's way way cheaper and more repeatable.
Nice, I've always used the glass beads shoved into a hole with thermal goop. I never thought about the response time of the bead/ goop. George H.
> > Cheers > > Phil Hobbs > > > > > > -- > Dr Philip C D Hobbs > Principal Consultant > ElectroOptical Innovations LLC / Hobbs ElectroOptics > Optics, Electro-optics, Photonics, Analog Electronics > Briarcliff Manor NY 10510 > > http://electrooptical.net > http://hobbs-eo.com
On 12/12/18 9:30 AM, George Herold wrote:
> On Tuesday, December 11, 2018 at 8:40:59 PM UTC-5, Phil Hobbs wrote: >> On 12/11/18 8:25 PM, George Herold wrote: >>> A 0.25" x 1.3 " pcb holds a sensor, heater (300 ohm, 1206 or bigger) >>> and diode temp. sensor. >>> I first made it with a common ground for the heater and diode, >>> but there was to much offset from the heater current* >>> and I had to add a third wire to sense local ground for the diode. >>> Which worked great. About a 2 sec thermal time constant. >>> >>> When I cut the ground plane (razor blade on thermals) >>> thermal time constant was 40 seconds. >>> >>> 1/2 oz copper vs 1/16" fr4. >>> >>> But if there are no expected thermal changes on the >>> 60 second time frame... is there some way faster >>> (thermally) is better? The pcb is wrapped in foam and >>> cooled.. >> >> Two-ounce copper and closer placement. >> >>> Just float everyone and be slow what's the problem? >> >> It trashes your thermal forcing rejection. If your bandwidth goes down >> by 20x, so does your forcing rejection at low frequencies, unless you're >> prepared to do fancy lead-lag things that are liable to become unstable >> if you have a void in your thermal paste, for instance. > OK I assume that 'thermal forcing rejection' is the same or related > to how much gain (delta(power)/delta(temp)) I can use. > It's certainly true that with the faster common ground plane I can > crank up the gain by a factor of ~20 also. (There's a V^2 term in there... > I'm using the term 'gain' a bit loosely.)
The forcing rejection depends on the loop gain, just like the load regulation of a linear voltage regulator.
> > So my response is trashed (reduced) at all frequency ranges... > But if there are no big temperature or load changes so what?
Then you don't really need a temperature controller. ;) If your system has a bandwidth of 0.01 Hz when it could have 0.2 Hz, for instance, the rejection of thermal forcing from somebody opening the lab door (say a 100s event) is reduced from 26 dB to essentially zero.
>> I have a current product that uses an 0603 thermistor mounted to a baby >> board in intimate contact with the cold plate. Crucially, the >> thermistor is on the same side of the board as the cold plate, and >> shares a common ground plane. The lateral clearance is about 0.040". >> (I could use a few in series if the cold plate were large enough to >> worry about gradients.) >> >> Its thermal response is dramatically faster than a glass bead thermistor >> and silver epoxy, and besides, it's way way cheaper and more repeatable. > > Nice, I've always used the glass beads shoved into a hole with thermal goop. > I never thought about the response time of the bead/ goop.
I was a bit worried about the low thermal diffusivity and high thermal mass of the glass epoxy smearing out the response, but with a short path in 1-oz copper it doesn't seem to be a problem, and it sure is easier to build and to repair. Use lock washers to make sure the thermal contact stays good. Cheers Phil Hobbs -- Dr Philip C D Hobbs Principal Consultant ElectroOptical Innovations LLC / Hobbs ElectroOptics Optics, Electro-optics, Photonics, Analog Electronics Briarcliff Manor NY 10510 http://electrooptical.net http://hobbs-eo.com
On Tue, 11 Dec 2018 17:25:25 -0800, George Herold wrote:


> When I cut the ground plane (razor blade on thermals) > thermal time constant was 40 seconds. >
Ugh! Yes, the thermal conductivity of fiberglass laminate is REALLY low. So, the 1 Oz copper has at least 20 X more conductivity than the substrate. One possible fix is to run a copper plane on the back side of the board. Jon
On Wednesday, December 12, 2018 at 2:43:03 PM UTC-5, Jon Elson wrote:
> On Tue, 11 Dec 2018 17:25:25 -0800, George Herold wrote: > > > > When I cut the ground plane (razor blade on thermals) > > thermal time constant was 40 seconds. > > > Ugh! Yes, the thermal conductivity of fiberglass laminate is > REALLY low. So, the 1 Oz copper has at least 20 X more conductivity than > the substrate. One possible fix is to run a copper plane on the back > side of the board. > > Jon
Right, I'm going to spin some more boards and try that. (ground plane on back side.) It is a rather shockingly bad time constant without that little piece of copper. GH.
On Wed, 12 Dec 2018 11:55:47 -0800 (PST), George Herold
<gherold@teachspin.com> wrote:

>On Wednesday, December 12, 2018 at 2:43:03 PM UTC-5, Jon Elson wrote: >> On Tue, 11 Dec 2018 17:25:25 -0800, George Herold wrote: >> >> >> > When I cut the ground plane (razor blade on thermals) >> > thermal time constant was 40 seconds. >> > >> Ugh! Yes, the thermal conductivity of fiberglass laminate is >> REALLY low. So, the 1 Oz copper has at least 20 X more conductivity than >> the substrate. One possible fix is to run a copper plane on the back >> side of the board. >> >> Jon > >Right, I'm going to spin some more boards and try that. >(ground plane on back side.) > >It is a rather shockingly bad time constant without that little piece >of copper. > >GH.
If I understand your geometry (and I probably don't) you could put the heater and the sensor on opposite sides of a thin board with a lot of copper on each. That would maximize the amount of FR4 heat conduction and minimize the heater-to-sensor tau. The overall board tau depends on the mass of the rig and the heat flow out to the world. That could be slow even if the board were solid diamond. -- John Larkin Highland Technology, Inc picosecond timing precision measurement jlarkin att highlandtechnology dott com http://www.highlandtechnology.com
On Wednesday, December 12, 2018 at 3:26:10 PM UTC-5, John Larkin wrote:
> On Wed, 12 Dec 2018 11:55:47 -0800 (PST), George Herold > <gherold@teachspin.com> wrote: > > >On Wednesday, December 12, 2018 at 2:43:03 PM UTC-5, Jon Elson wrote: > >> On Tue, 11 Dec 2018 17:25:25 -0800, George Herold wrote: > >> > >> > >> > When I cut the ground plane (razor blade on thermals) > >> > thermal time constant was 40 seconds. > >> > > >> Ugh! Yes, the thermal conductivity of fiberglass laminate is > >> REALLY low. So, the 1 Oz copper has at least 20 X more conductivity than > >> the substrate. One possible fix is to run a copper plane on the back > >> side of the board. > >> > >> Jon > > > >Right, I'm going to spin some more boards and try that. > >(ground plane on back side.) > > > >It is a rather shockingly bad time constant without that little piece > >of copper. > > > >GH. > > If I understand your geometry (and I probably don't) you could put the > heater and the sensor on opposite sides of a thin board with a lot of > copper on each. That would maximize the amount of FR4 heat conduction > and minimize the heater-to-sensor tau.
Yeah that might work, a 'ground' plane on each side. It could be a little smaller then too.. that would help. George H.
> > The overall board tau depends on the mass of the rig and the heat flow > out to the world. That could be slow even if the board were solid > diamond. > > > > > > > -- > > John Larkin Highland Technology, Inc > picosecond timing precision measurement > > jlarkin att highlandtechnology dott com > http://www.highlandtechnology.com