Methods and apparatus are provided for burst transfers of data between DDR memories and embedded processors during training of the PHY interface in an embedded system. An embedded system comprises an embedded processor having at least one cache controller; a memory, wherein the memory has an atomic memory access that comprises a plurality of clock edges; and a memory controller having a physical interface to convert digital signals between the embedded processor and the memory, wherein the cache controller executes a training process to determine a delay through the physical interface for each of the plurality of clock edges using a burst transfer of data. The burst transfer comprises reading a data pattern from the memory and storing the data pattern in one or more registers in the embedded processor.