Hhhmm...I was hoping some COMSOL user would come along and answer this seemingly simple issue.
I am not an FEA modeler, but I think I can offer a couple of ideas that might work.
If you build your model from the bottom up, i.e., creating points, then lines, then surfaces and then volumes.
You may think of building your SiO2 and Aluminum pieces separately but that is not correct as you will then need to figure out how to glue them together; instead, make sure you build the pieces by re-using the common surface.
In other words, instead of having 4 points from one piece on one end and another 4 points from the other, with 2 of the points being concurrent...you just really need 6 points...using the top 2 and the middle 2, you create lines and a surface for the end of one bar and using the bottom 2 and the middle 2 (again), you create the remaining lines and another surface for the end face of the other bar.
Proceed similarly with the rest of the points, create a single set of lines and surface for the middle common wall.
Create the two volumes re-using the common wall
I think after this, your model should behave as you expect...I hope.
On the flip side, if you are the kind of user that make models via volumes and boolean operations, I think you may be able to build a single volume as large as your bimetal piece, then come and split the piece right down the middle making sure to request something during the operation so that you do not end up with two separate bodies but with 2 pieces sharing a common wall...maybe you can find something along those lines ?
gsal