Device and method for improving simulator parameter

1 . A device comprising:

a memory configured to store an evolution pool and an input tensor-score pair; and

a processor configured to

sample a generator network in the evolution pool,

initiate a critic network to train the sampled generator network in a back- propagation manner,

generate an input tensor from the trained generator network,

initiate a simulator, trained to output a simulator parameter for a design of a semiconductor device, to perform a black box operation on the input tensor and to output an output score as a result of the black box operation, and to store the input tensor and the output score in the memory as the input tensor-score pair,

store information used in the black box operation in the evolution pool when the output score is greater than a minimum score stored in the memory, and

sort the stored information, and update the evolution pool such that only a preset number of information is left in the evolution pool.

2 . The device of claim 1 , wherein the critic network includes a neural network model trained to simulate an input-output relationship of the black box operation.

3 . The device of claim 2 , wherein the processor is configured to train the critic network by updating the critic network such that a predicted score calculated through the black box operation on the input tensor is closer to the output score, based on the input tensor-score pair stored in the memory.

4 . The device of claim 1 , wherein the processor is configured to train the critic network and update the evolution pool until a number of iterations thereof reaches a preset maximum number.

5 . The device of claim 4 , wherein the processor is configured to determine the generator network generated based on the trained critic network when each of a number of training times of the critic network and a number of updating times of the evolution pool reaches the preset maximum number,

wherein the processor is configured to output input tensor data to the simulator, and

wherein the input tensor data is based on the determined generator network.

6 . The device of claim 1 , wherein the updating of the evolution pool includes:

sorting the stored information based on a level of the output score,

leaving only the preset number of information in a descending order from a maximum score, and

deleting a remaining portion of the information.

7 . The device of claim 1 , wherein the updating of the evolution pool includes:

recalculating and rearranging scores of the stored information every set period, and

leaving only information satisfying a preset condition and deleting a remaining portion of the information.

8 . The device of claim 1 , wherein the updating of the evolution pool includes:

sorting the stored information in an oldest order based on a time at which the information is stored,

leaving only information having a storing time thereof after a preset timing, and

deleting a remaining portion of the information.

9 . The device of claim 1 , wherein the information stored in the evolution pool includes the generator network, the input tensor, the output score, and a storing time of the information, and

the generator network, the input tensor, the output score, and the storing time are stored in association with each other.