Board mounting stress, LTC2986

[slugrustle]

I'm looking to make a thermocouple reader using the Analog Devices LTC2986 measurement IC, which I would like to use because it does almost all of the hard parts in making the measurement.

While reading the LTC2986 datasheet, I noticed two statements indicating that mechanical stresses on the IC can introduce errors.

Long-term stability will also be affected by differential stresses between the IC and the board material created during board assembly.

The mechanical stress of soldering the LTC2986 to a PC board can cause the output voltage reference to shift and temperature coefficient to change. These two changes are not correlated. For example, the voltage may shift but the temperature coefficient may not. To reduce the effects of stress-related shifts, mount the reference near the short edge of the PC board or in a corner.

I've seen similar concepts mentioned in the datasheet for TI's REF54.

The long-term stability value is tested in a typical setup that reflects standard PCB board manufacturing practices. The boards are made of standard FR4 material, the board does not have special cuts or grooves around the devices or go through burn-in process to relieve the mechanical stress of the PCB. These conditions reflect real world use case scenario and common manufacturing techniques.

Can someone point me to an article or app note that outlines best practices for preventing or relieving mechanical stresses on an IC package?  I would like to do my best here, even if I can't quantify the magnitude of the effect.

For what it's worth, my normal practice is to use 4-layer FR-4 boards and to solder by hand with tin-bismuth solder.  I use this because it behaves mostly like leaded solder while being lead free.  The bonus is that it has a lower melting point than even eutectic tin-lead solder.  Maybe this assembly method results in lower stresses compared to reflow with normal lead free solder? 

[slugrustle]

I've found a little bit of information in various app notes about MEMS sensors, but nothing very comprehensive.  At the very least, one of them recommended against manual soldering due to the uneven heating.

Another mentioned that consistent volume of thermal paste on all pads is important and went on to recommend a certain process control methodology and alignment between stencil and PCB of 25µm.  I'm not going to be able to swing that one.

Makes me wonder if using something like MG Chemicals 9410 conductive adhesive might be a nice way to mount an IC like this without introducing much mechanical stress.  Solder all of the other parts, clean the board, carefully apply this adhesive, carefully place the IC, then bake at 90°C for 1 hour and ramp the temperature down slowly.

[Kleinstein]

The main point effected by the board stress should be the voltage reference. The reference is anyway not that good and if critical an external reference would likely be more effective than much extra effort to reduce board stress.

Thermocouples are usually not very accurate anyway. So for this application the board stress should be less of an issue. RTD and similar resistive sensors are anyway ratiometric and do no depend on the reference voltage.  Besides thermal board stress, there is also the humidity effect of a plastic case and FR4 PCB.

[Andreas]

Can someone point me to an article or app note that outlines best practices for preventing or relieving mechanical stresses on an IC package?

The question is: what kind of board stress?

Against the "interenal" board stress there is not much you can do. (Epoxy PCB is swelling with air humidity).
Except for selecting a board material with the same CTE as the IC-Package (only practical possible when both are ceramics)
or put the sensitive part of the cirquit (usually the voltage reference and eventually the A/D converter) on a weak (polyimide) part of the pcb (flex pcb).

Against external stress (by mounting or vibration) you can put cutouts around the
See for example in newer voltage reference data sheets like the LTC6655:
https://www.analog.com/en/products/ltc6655.html

with best regards

Andreas

[slugrustle]

The main point effected by the board stress should be the voltage reference. The reference is anyway not that good and if critical an external reference would likely be more effective than much extra effort to reduce board stress.

Point taken.  I'll look at external references.  I'm still curious about how to reduce board stress even if it's not the major error term, if only to know how to do it.

Thermocouples are usually not very accurate anyway. So for this application the board stress should be less of an issue. RTD and similar resistive sensors are anyway ratiometric and do no depend on the reference voltage.  Besides thermal board stress, there is also the humidity effect of a plastic case and FR4 PCB.

One of the motivations to make this thermocouple reader is to reduce the errors in the reader and its cold junction compensation as much as possible.  I think a lot of commercial units take the same philosophical approach that you outlined, and it kind of bugs me—as much as it also makes sense. 

I wouldn't mind using Rogers 4350B or Teflon...

[slugrustle]

The question is: what kind of board stress?

Against the "interenal" board stress there is not much you can do. (Epoxy PCB is swelling with air humidity).
Except for selecting a board material with the same CTE as the IC-Package (only practical possible when both are ceramics)
or put the sensitive part of the cirquit (usually the voltage reference and eventually the A/D converter) on a weak (polyimide) part of the pcb (flex pcb).

Against external stress (by mounting or vibration) you can put cutouts around the
See for example in newer voltage reference data sheets like the LTC6655:
https://www.analog.com/en/products/ltc6655.html

LTC6655's datasheet has more info, thanks.  It also mentions app note 82, which has good information as well.

The humidity effect is interesting, and I was not aware.  LTC6655's datasheet says conformal coating can help delay the effect, so perhaps CTE matching and ceramics are not the only method.  The articles I found on MEMS devices also seemed to say that improved soldering processes can help with this, or at least hurt less.  It's odd to think that CTE has to do with humidity?  I thought CTE matching was solely about stresses from temperature changes. 

The datasheets also mention stress due to soldering, and maybe reflow with lower temperature solder (Indium alloys look promising) or using thermally conductive adhesive could help there.

[Kleinstein]

A relatively simple way (as suggested by Andreas )to reduce PCB stress is with a lexible PCB and this way the PCB much weaker than the chip.
To avoid stress from mounting a large PCB can be using cut outs and some care where the mounting holes are.

The CTE is not directly linked to humidity, though humidity in plastics can change the properties, maybe also the CTE. A temperature rise with locally lower the relative humidity level and this can give an additional relatively slow settling effect for instruments only power up from time to time. The delay of days and possibly weeks can make the humidity effect hard to test and correct.
There is limited need to really fight thermal stress with parts in a plasic case, as the humidity effect may well be larger than the normal thermal effect.

The main issues I see with thermcouples that I see are the connectors. Some of the relative expensive Lemo connectors sometimes used are not that great, as they have copper contracts. I had better experiance with the relatively cheap color codes plastic ones with proper TC material contacts.  Another issue can be thermal contact at the cold junction. The wires need good thermal contact to no effect the junctions. So reserve some area to the cold junctions and thermal clamping of the wires going there. The LTC2986 is already made for a separate cold junction sensor and not a build in sensor as some of the simper solutions. This helps to keep termal gradients small around the cold junction.
The common type K thermocouples also have some slight magnetic effect that can limit the accuracy. The curie temperature is well within the normal use temperature for those TCs. Another point is having alloys to accurately match the standard curves. The absolute accuracy is just a bit limited, no matter how good the cold junction, amplifier and ADC. They can still be OK for repeatabilty, getting back to the same temperature of checking small differences.

[AnalogTodd]

The question is: what kind of board stress?

Against the "interenal" board stress there is not much you can do. (Epoxy PCB is swelling with air humidity).
Except for selecting a board material with the same CTE as the IC-Package (only practical possible when both are ceramics)
or put the sensitive part of the cirquit (usually the voltage reference and eventually the A/D converter) on a weak (polyimide) part of the pcb (flex pcb).

Against external stress (by mounting or vibration) you can put cutouts around the
See for example in newer voltage reference data sheets like the LTC6655:
https://www.analog.com/en/products/ltc6655.html

LTC6655's datasheet has more info, thanks.  It also mentions app note 82, which has good information as well.

The humidity effect is interesting, and I was not aware.  LTC6655's datasheet says conformal coating can help delay the effect, so perhaps CTE matching and ceramics are not the only method.  The articles I found on MEMS devices also seemed to say that improved soldering processes can help with this, or at least hurt less.  It's odd to think that CTE has to do with humidity?  I thought CTE matching was solely about stresses from temperature changes. 

The datasheets also mention stress due to soldering, and maybe reflow with lower temperature solder (Indium alloys look promising) or using thermally conductive adhesive could help there.

The slotting suggested in app note 82 is a good way to reduce board stress on the package. Basically, when the board flexes, it flexes the package and that creates stress on the internal IC that can cause shifts. Cut where most of that flexing can occur and you get better results.

Humidity is an interesting effect that a lot of people never really got into until it was noticed while checking long term drift on references. The references in plastic packages seemed to have huge fluctuations even though in a controlled temperature chamber with very little airflow. When someone put a humidity meter next to the chamber and tracked the drift with humidity, things suddenly matched up. The issue is that the plastic used in these packages is not perfect; it will allow small amounts of moisture into it and when it does, it expands and contracts. So when humidity rises, the plastic swells slightly and puts added stress on the die. When it drops, the moisture exits and stress is reduced. That's why the best references are in metal cans or hermetic packages.

[slugrustle]

The main issues I see with thermcouples that I see are the connectors. Some of the relative expensive Lemo connectors sometimes used are not that great, as they have copper contracts. I had better experiance with the relatively cheap color codes plastic ones with proper TC material contacts.

I'm thinking to use the Omega engineering PCC-SMP-E-5 connectors, which are specifically made for Type E only.

Another issue can be thermal contact at the cold junction. The wires need good thermal contact to no effect the junctions. So reserve some area to the cold junctions and thermal clamping of the wires going there. The LTC2986 is already made for a separate cold junction sensor and not a build in sensor as some of the simper solutions. This helps to keep termal gradients small around the cold junction.

Plan is to mount an Ametherm ACC-104 thermistor on the negative terminal of each thermocouple connector using Parker Chromerics CIP35 thermal paste.

The common type K thermocouples also have some slight magnetic effect that can limit the accuracy. The curie temperature is well within the normal use temperature for those TCs. Another point is having alloys to accurately match the standard curves. The absolute accuracy is just a bit limited, no matter how good the cold junction, amplifier and ADC. They can still be OK for repeatabilty, getting back to the same temperature of checking small differences.

I'm using type E thermocouples made with special limits of error wire.  Not that it's perfect either, granted.  I do want to get a welder with argon purge eventually.  My thermocouple welder right now just has a carbon/graphite block and no purge system.

[Kleinstein]

In principle the connector is OK, but it has the problem that the cold junction is directly at the connector/PCB interface. To avoid much thermal current via the plug, this wants thermal shielding of the connector plugged in. So one would ideally have the connectors not at the edge of the boad, so that the plug is more or less in contact with the PCB. However this makes connecting a bit more difficult.

For the cold juction sensor one may as well have it directly PCB monted and have the main thermal contact via the pins. Not sure of one really need the thermal paste, maybe better some thermal shield to cover also the top side.

[slugrustle]

For the connectors I'm looking to use (Omega PCC-SMP-E-5), the spot weld between the constantan of the themocouple side and the copper of the PCB side occurs above the PCB.  That's where I was going to put the cold junction temperature sensor.  They even have a version of this connector with a clip there for mounting TO-92 case sensors / diodes.

I was wondering about some kind of foam or thermal insulation around that point.

[Kleinstein]

The relatively thick pins would still make a rather good contact to the PCB, especially with a good size pad that may be a good idea for mechanical stability.
The temperature sense may have enough space between the contact welding spot and the PCB.

If thermal insulation is used, this should be not just local, but more like include both sides of the connector. A main path to bring a high flow in is via the connector and maybe cable.

[slugrustle]

Ahhhh, I think I missed your point and am starting to understand it.  You're talking about avoiding thermal gradients near / around the connector, the cold junction, and the PCB traces due to thermal flux from the external thermocouple connector and wire.

Since this is a personal project (if I go through with it), and the plan is to galvanically isolate each channel, it would be beneficial for both thermal gradient and isolation reasons to have the thermocouple side connector be inside the instrument enclosure behind a door of some kind.

Excellent point.