A stuckcell may be defined as a defect giving rise to 1/12 unmaskable
occupancy for the affected channel. (A similar defect gives rise to 1/8
occupancy - near enough the same for the purposes of this discussion.)

The relevant considerations are as follows:

   * How many stuckcells must be present on a module in order to introduce
     deadtime to the SCT optolinks?
   * How many stuckcells may be suppressed by ROD?
   * How many stuckcells must be present in one ROD in order to introduce
     deadtime to the SLINK?

From the first consideration, we get a limit to the number of stuckcells
that may be present on a single module. The second and third points, taken
together, impose a limit to the number of stuckcell defects that may be
safely connected to each ROD.

Deadtime is introduced to the optolinks when the occupancy approaches 2%. A
stuckcell defect will affect on average only 1 event in 12, so the mean
background occupancy introduced by a single stuckcell (1/12) * (1/768). This
is two orders of magnitude lower than the limit imposed by deadtime
considerations in the optolinks: definitely not a problem!

From the second consideration, we do not yet have a firm answer regarding
the number of stuckcell defects that may be suppressed within a ROD, however
this may in principle be done. For now, let's assume that none of these
defects may be suppressed in ROD: we may still get a pleasant surprise!

Now for the S-LINK. The readout system has been designed to tolerate 2% mean
occupancy without introducing deadtime, hence this is also the occupancy
which may be tolerated by the S-LINK. Hence if we control the number of
stuckcells allowed on a single module to keep within the limit imposed by
the SCT optolinks, we are also safe from the viewpoint of the SLINK,
irrespective of anything clever that may or may not be done to supress this
background within the ROD.

Given that the number of stuckcell defects we have seen is relatively small,
here is my suggestion:

   * Allow up to two STUCKCELL defects per side of a module. This may be
     restated as: allow up to 2/(12*768) unmaskable occupancy per side of a
     module. Chips should be changed only once this limit has been exceeded.