Poly Resistor Matching on an IC

[QQ]

Hi,

I am designing a bandgap reference on an IC and I am finding it hard to
tune the resistors so that all the resistances can be broken into unit
resistors AND the bandgap curve is as expected.

So my question is : If I place one resistor, say l=37u, w=5u and
another one of l=38u, w=5u close to each other (ie. their lengths
differ by only 1u) will their resistance per unit square (for a given
IC) match? Will the only error be due to possible dimensional mismatch
or will the resistance per unit square change too, due to the length
difference?

Thanks
QQ

 

[Chris Jones]

Hi,

I am designing a bandgap reference on an IC and I am finding it hard to
tune the resistors so that all the resistances can be broken into unit
resistors AND the bandgap curve is as expected.

So my question is : If I place one resistor, say l=37u, w=5u and
another one of l=38u, w=5u close to each other (ie. their lengths
differ by only 1u) will their resistance per unit square (for a given
IC) match? Will the only error be due to possible dimensional mismatch
or will the resistance per unit square change too, due to the length
difference?

Thanks
QQ

That should match ok, but if you really wanted to make it out of unit
resistors, then you could write a little program to exhaustively search
through all series and parallel and series-parallel combinations of a small
number of resistors, to find a couple of networks which give you very close
to the right ratio. Your unit resistor is pretty big, it might be easier
to find networks of unit resistors if your unit resistor were smaller and
you were to use more of them.

Make sure you account for metal resistance as this can be bigger than the
resistor mismatch.

If you want to have two different value resistors that match reasonably well
then it may be preferable to keep the length of both resistors the same and
get the different values by changing the width, trying to keep the number
of contacts proportional to the width. This should keep the contact
resistance roughly in proportion to the resistance of the middle section.

Chris