@User - Why wouldn't you just use forward error correction encoding to use all the spare columns?

In some cases fails are due to shorted columns. These errors can be corrected with an ECC scheme, however extra consumption of shorted elements can be a problem (Icc values go out of spec). In these cases the fail column must be physically disabled and a spare column enabled. This type of test is done at wafer testing after a number of burnin cycles to trigger near to fail elements, and redundancy is used to replace them. ECC is used on field to correct additional fail bits. In addition, if you have a shorted row no spare column and ECC scheme can recover the data, since on that specific row address all columns are unreadable, including the spare ones. In this situation you must redirect the row access to a spare row.