|
Consider making a contact hole
through the CVD oxide that was deposited over the finished
transistors. |
|
 |
Since the source/drain area are as small as
possible, you have to make not only a small hole but
you must align it precisely relative to the gate stack of the
transistors. |
|
 |
Small misalignements would produce a short
circuit between the gate electrode and the Al that will be put in the
contact hole eventually. |
 |
Now cover the whole gate stack with
Si3N4 that will resist whatever etching procedure
you use to remove the oxide in the future contact hole. You gain a lot in your
"process window" for the contact
hole - misalignements don't matter as much any more as shown below. |
|
|
|
|
|
|
|
|
|
 |
That looks pretty good - but there is
a prize to pay: |
|
 |
First of all - how do we make the nitride
encapsulation? And don't forget; Si3N4 must never
come in contact with Si - you always need a thin layer of
buffer oxide underneath
(not shown in the picture). There is quite a bit of added process
complexity! |
|
 |
The topography gets worse. The aspect ratio of
the contact hole - the relation between depth and diameter - increases and with
it the problems of filling it with Al. |
|
 |
Still, starting around 1987, FOBIC
was used and helped to get the next generation onto the market. |
 |
The picture below is a cross section
through a 16 Mbit DRAM memory cell
shown before.
It shows one of the contacts to a transistor (which is connected on the other
side to a trench
capacitor). The FOBIC structure has been outlined; it is clearly
visible. |
|
|
|
|
|
|
|
|
|
© H. Föll