Neural processing unit, neural processing system, and application system
Provided is a neural processing unit that performs application-work including a first neural network operation, the neural processing unit includes a first processing core configured to execute the first neural network operation, a hardware block reconfigurable as a hardware core configured to perform hardware block-work, and at least one processor configured to execute computer-readable instructions to distribute a part of the application-work as the hardware block-work to the hardware block based on a first workload of the first processing core.
1. A neural processing unit configured to perform application-work including a first neural network operation, the neural processing unit comprising:
a first processing core configured to execute the first neural network operation;
a hardware block reconfigurable as a hardware core configured to perform hardware block-work, the hardware block being reconfigurable between different types of hardware cores for performing different types of operations, and the different types of operations including different types of neural network operations; and
at least one processor configured to execute computer-readable instructions to
distribute a part of the application-work as the hardware block-work to the hardware block based on a first workload of the first processing core, and
reconfigure the hardware block as a first type of hardware core for performing a first type of operation included in the hardware block-work, the first type of hardware core being among the different types of hardware cores, the first type of operation being among the different types of operations, and the first neural network operation including the first type of operation.
2. The neural processing unit of claim 1 , wherein the at least one processor is configured to execute computer-readable instructions to distribute a part of the first neural network operation as the hardware block-work to the hardware block depending on whether the first workload exceeds a defined value.
3. The neural processing unit of claim 1 , wherein the at least one processor is configured to execute computer-readable instructions to reconfigure the hardware block as a convolutional neural network (CNN) processing hardware core when the hardware block-work includes a CNN operation.
4. The neural processing unit of claim 3 , wherein the CNN operation comprises image recognition processing.
5. The neural processing unit of claim 1 , wherein the at least one processor is configured to execute computer-readable instructions to reconfigure the hardware block as a recurrent neural network (RNN) processing hardware core when the hardware block-work includes an RNN operation.
6. The neural processing unit of claim 5 , wherein the RNN operation comprises voice recognition processing.
7. The neural processing unit of claim 1 , wherein
the first neural network operation comprises a first multiply and accumulate (MAC) operation, and
the at least one processor is configured to execute computer-readable instructions to distribute a part of the first MAC operation as the hardware block-work to the hardware block based on a ratio of the first MAC operation and the first workload.
8. The neural processing unit of claim 1 , wherein
the at least one processor is configured to execute computer-readable instructions to reconfigure the hardware block as a MAC processing hardware core when the hardware block-work includes a MAC operation, and
the hardware block reconfigured as a MAC processing hardware core is configured to perform the MAC operation using a look-up table including quantized weighted data.
9. The neural processing unit of claim 8 , wherein the look-up table comprises one or more result value obtained based on operating input data and the quantized weighted data.
10. The neural processing unit of claim 1 , further comprising:
a second processing core configured to execute a second neural network operation different from the first neural network operation, wherein
the application-work comprises the second neural network operation, and
at least one processor is configured to execute computer-readable instructions to distribute a part of the first neural network operation or a part of the second neural network operation as the hardware block-work to the hardware block based on the first workload and a second workload of the second processing core.
11. The neural processing unit of claim 10 , wherein the at least one processor is configured to execute computer-readable instructions to
distribute the part of the first neural network operation as the hardware block-work to the hardware block in response to an amount of the first workload being larger than an amount of the second workload, and
distribute the part of the second neural network operation as the hardware block-work to the hardware block in response to the amount of the first workload being smaller than the amount of the second workload.