Expect to go back to the original for matters of substance, otherwise a 1-bit error in your recollection could make you suffer (e.g. changing the sign in an equation).  

Concentrate only on remembering the major author(s) and affiliation (easy way to mention it to someone else), how you would find it again, and the key new stuff (for you) which you should find  in the abstract.  If there's an important equation/algorithm/method etc. then you will want to write it down, but that's not so much *remembering a paper* as learning something new *from the starting point of a paper*.

**Edit:** I should note that I vastly prefer reading papers on paper and can therefore make margin notes and file particularly interesting ones.