I am looking for a way to reduce a 11 byte unique ID to 32 bits or fewer. I am using an Atmel AVR microcontroller that has the ID number burned in at the factory, but because it has to be transmitted very often in a very low power system I want to reduce the length down to 4 bytes or fewer.

The ID is guaranteed unique for every microcontroller. It is made up of data from the manufacturing process, basically the coordinates of the silicone on the wafer and the production line that was used. They look like this:

304A34393334-16-11001000 314832383431-0F-09000C00

Obviously the main danger is that by reducing these IDs they become non-unique. Unfortunately I don't have a large enough sample size to test how unique these numbers are. Having said that because there will only be tens of thousands of devices in use and there is secondary information that can be used to help identify them (such as their approximate location, known at the time of communication) collisions might not be too much of an issue if they are few and far between.

Is something like MD5 suitable for this? My concern is that the data being hashed is very short, just 11 bytes. Do hash functions work reliably on such short data?