Are you at all familiar with the way CRT text displays are generated? This thing can be done similarly. The advantage of a relatively small (80-character) display as opposed to the typical 25x80 (2000-bytes) CRT display is that you can do the timing and multiplexing in your MCU. After all, microcontrollers/microprocessors were developed as a substitute for lots of logic.

Typically, they consist of a counter chain that generates the timing and addressing, a refresh memory that contains the characters to be displayed, a character generator that converts the logical display information contained in the refresh memory into the physical pixels to be displayed. In the matrix LED array, the LED's are the pixels.

In such an arrangement, if you have two successive 'A' characters, you must write them to two separate locations in the refresh memory. The "character generator" function converts each one to the appropriate pixel combination, and the scrolling function is implemented in firmware as well, by moving the pointer into the character generator ahead every "n" frame cycles, where "n" is determined by the rate at which you want to scroll. This can be accomplished by adding an offset to the column number that is being translated in the character generator function.

If you give this some thought, and, perhaps, you look at a few simple CRT display circuits, which may take a bit of work to find, but particularly if you look at "smooth scrolling" circuits, you will see how it's done there, and inherently, you should also be able to see how one might do that in your case.

RE