Neural processing unit capable of reusing data and method thereof

A neural processing unit (NPU) comprising:

• a mode selector configured to select a first mode or a second mode;
• and
• a processing element (PE) array including a plurality of PE rows and a plurality of PE columns,
• wherein a first area of the processing element array assigned for the first mode is larger than a second area of the processing element array assigned for the second mode,
• wherein the first area includes at least part of PE rows from the plurality of PE rows,
• wherein the first area and the second area overlap with each other, and
• wherein the plurality of PE rows of the second area are configured to perform a depth-wise convolution operation in response to the second mode.

A neural processing unit (NPU) comprising:

• a mode selector configured to select a first mode or a second mode;
• and
• a processing element array including a plurality of PE rows and a plurality of PE columns,
• wherein a first area of the processing element array assigned for the first mode is larger than a second area of the processing element array assigned for the second mode,
• wherein the first area and the second area overlap with each other,
• wherein a number of the plurality of PE rows of the first area is larger than a number of the plurality of PE rows of the second area, and
• wherein the plurality of PE rows of the second area is configured to perform a depth-wise convolution operation in response to the second mode.