According to one embodiment, a memory controller includes a decoder configured to perform approximate maximum likelihood decoding, the decoder including: an initial value generation unit configured to calculate first data on the basis of a received word read from a non-volatile memory; a storage unit configured to store the first data and a predetermined number of second data; an update unit configured to calculate new second data by using the predetermined number of second data stored and update the storage unit; an arithmetic unit configured to output an addition result of the first data and the latest second data as decoded word information; and a selection unit configured to select a decoded word with the maximum likelihood on the basis of a plurality of the decoded word information.