System-on-chip, data processing method thereof, and neural network device

A data conversion method of a System-on-Chip, the method comprising:

• dividing first data obtained from a first memory into a plurality of pieces of division data;
• assigning a plurality of tags to the plurality of pieces of division data, each of the plurality of tags comprising a coordinate value for a corresponding piece of division data;
• obtaining a transform matrix using affine transform matrices of single-axis rotation, based on a given transformation condition;
• transforming coordinate values included in the plurality of tags for the plurality of pieces of division data using the obtained transform matrix;
• and
• storing the plurality of pieces of division data in a second memory, based on the transformed coordinate values corresponding to the plurality of pieces of division data.

A data padding method of a System-on-Chip, the method comprising:

• assigning a plurality of tags to a plurality of pieces of division data into which first data is divided, each of the plurality of tags comprising a coordinate value for a corresponding piece of division data;
• determining whether each of the plurality of pieces of division data requires padding, based on the coordinate value included in each of the plurality of assigned tags;
• generating padding data for a piece of division data, which is determined to require padding, from among the plurality of pieces of division data;
• transforming a coordinate value of the piece of division data, which is determined to require padding, based on a type of padding;
• and
• storing the generated padding data in a memory using the transformed coordinate value.