LT1991

In general, you should try to avoid using external resistors at all with a precision differential amplifier that already provides internal R's. Any external R mismatch at the inputs will reduce performance: it will add additional offset and drift, increase gain error, and hurt CMRR. 

To avoid those performance reductions, you'd have to buy very expensive 1.78MOhm R's for 0.1% accuracy matching.

On-chip resistors will always beat off-chip in terms of tolerance: when you buy the part, you get R's with 0.08 or 0.12% mismatch at worst, at no extra cost. Since 1.78MOhm is also so much bigger than the internal 450k resistors themselves, its error will dominate over the part's internal R's. Finally, the node where the 1.78M meets the input will be very sensitive, and a potential path for leakage error.

My question for you is why the 1.78M is needed at all? I think it is possible to achieve the gains you want with just the existing part, as long as you connect the correct inputs together. Unless you have a very specific input current limitation or super high input voltage that you're trying to measure, external giant R's should be unnecessary.

Please try designing with our LT1991 web tool:

 http://beta-tools.analog.com/toolbox/LT1991

Using the tool, if you select a gain of 1.25, Single-Ended (S-E) Non-Inverting, you can achieve that just by tying the input to three pins, another one to GND, and tying another two pins together for the output. This tool will also generate a downloadable LTspice file of the design it generates that you can simulate.

By using the existing internal R's and just tying pins together externally, you can avoid all of the bad side effects of external R mismatch and exposed sensitive input nodes that I mentioned above.

Even if accuracy isn't that important to you, at the very least, going with the tool's design is two fewer reliability risks and two fewer components to add to a BOM.

In general, you should try to avoid using external resistors at all with a precision differential amplifier that already provides internal R's. Any external R mismatch at the inputs will reduce performance: it will add additional offset and drift, increase gain error, and hurt CMRR. 

To avoid those performance reductions, you'd have to buy very expensive 1.78MOhm R's for 0.1% accuracy matching.

On-chip resistors will always beat off-chip in terms of tolerance: when you buy the part, you get R's with 0.08 or 0.12% mismatch at worst, at no extra cost. Since 1.78MOhm is also so much bigger than the internal 450k resistors themselves, its error will dominate over the part's internal R's. Finally, the node where the 1.78M meets the input will be very sensitive, and a potential path for leakage error.

My question for you is why the 1.78M is needed at all? I think it is possible to achieve the gains you want with just the existing part, as long as you connect the correct inputs together. Unless you have a very specific input current limitation or super high input voltage that you're trying to measure, external giant R's should be unnecessary.

Please try designing with our LT1991 web tool:

 http://beta-tools.analog.com/toolbox/LT1991

Using the tool, if you select a gain of 1.25, Single-Ended (S-E) Non-Inverting, you can achieve that just by tying the input to three pins, another one to GND, and tying another two pins together for the output. This tool will also generate a downloadable LTspice file of the design it generates that you can simulate.

By using the existing internal R's and just tying pins together externally, you can avoid all of the bad side effects of external R mismatch and exposed sensitive input nodes that I mentioned above.

Even if accuracy isn't that important to you, at the very least, going with the tool's design is two fewer reliability risks and two fewer components to add to a BOM.

Hello Chang

Thanks for your valuable response.