As a side note, I'd get rid of the $i and use what perl offers: push. (I'd also use an array rather than a reference to the array, but that's minor.)

my @_raa;
while (<DATA>) {
    chomp;
    my @a=split(','); # use Text::CSV_XS here.
    push @{$_raa[$_]}, $a[$_] for 0..$#a;
};
[download]

And then, if you do use Text::CSV_XS to extract the fields, use it to merge the fields back together, since the extraction will remove quotes and escapes, so you'll need to push it back with quotes and escapes (which Text::CSV_XS does for you automatically). The difference between my code and yours is that if some rows have more columns than others, we'll do different things. Yours will have some undef's in those places, while mine will silently ignore them, possibly shifting things around incorrectly. What to do in this scenario wasn't spec'd by the OP, so it's hard to tell which one is right.