Neural network accelerator configured to perform operation on logarithm domain

A neural network accelerator comprising:

• a log converter configured to convert real number data on a real domain into log data on a logarithm domain;
• and
• an operation circuit configured to convert a multiplication operation on the real domain and a division operation on the real domain for the real number data into logarithmic addition and logarithmic subtraction for the log data and to convert addition on the real domain for the real number data into Jacobian logarithmic addition for the log data,
• wherein the operation circuit includes a Jacobian logarithmic adder configured to perform the Jacobian logarithmic addition,
• wherein the Jacobian logarithmic adder includes:
• a maximum value determiner configured to output a maximum value based on a first magnitude component corresponding to first log data and a second magnitude component corresponding to second log data;
• a sign determiner configured to output a sign component corresponding to the maximum value among a first sign component corresponding to the first log data and a second sign component corresponding to the second log data, as an output sign component;
• an offset operator configured to quantize a difference between the first magnitude component and the second magnitude component and to output an output offset based on the first sign component, the second sign component, and the quantization result;
• and
• a magnitude operator configured to calculate an output magnitude component of an output data based on the maximum value and the output offset.