These have been a popular magazine project, and I actually have half
a dozen used commercial ones sitting in my garage waiting for reverse
engineering.  I did not get the programmer box, and their Lithium RAM
batteries are dead except for one, so there is no other choice.

  They all use a similar construction, the units I have happen to have
a 6803 (effectively, the discrete NMOS predecessor of the 68HC11)
controller feeding the array via some 10 or twelve lines - or is it five
or six?

  The array has a set of latched shift registers driving the columns,
and the rows are driven in turn, i.e., it is multiplexed by row.  A
complete set of row data is shifted in whilst the previous row is being
strobed, then the strobe is turned off, the data latched, and the next
row strobe is turned on.

  With 1 in 7 multiplexing, good brightness is obtained.  You will
probably need ULN drivers for the columns making two chips and 8
resistors per 8 columns or 10 chips per 8 7x5 modules.  You may get by
with a ULN driver for the rows as whilst they draw a heavy current (the
full supply current in fact), only one is ever enabled at once.

  The "plenty of I/O lines" requirement is debatable.  Three to select
the row via an HC138, one for serial data and one to latch makes five
lines to drive the display, add I2C bus, serial port and maybe Microwire
for temperature.  Using an I2C EEPROM for data storage, since it is only
read repeatedly, should work.  You might as well initialise it and use
it for the font also (i.e., download the font from the PC along with
the message!)  I have a suspicion you could do it with a PIC16F84.

  Cheers,
        Paul B.