i've heard of it before. i'll be honest: i don't much like haskell. for example, this makes me crazy (haha):

  mmult mA mB = result
    where
      mBT      = transpose mB
      dot a b  = sum $ zipWith (*) a b
      result   = map (\ar -> dot ar <$> mBT) mA

either way, this is along the right lines. however, what FPGAs can you target? any xilinx chips?

it's really gonna take xilinx or altera to innovate in this space, but that is unlikely any time soon.