Method of predicting characteristic of semiconductor device and computing device performing the same
To predict characteristics of a semiconductor device, a simulation current-voltage curve of the semiconductor device is generated using compact models where each compact model generates simulation result data by performing a simulation based on device data. The simulation result data indicate characteristics of semiconductor devices corresponding to the device data. The compact models respectively corresponding to process data and semiconductor products. Simulation reference points on the simulation current-voltage curve are extracted. Basic training data corresponding to a combination of the simulation reference points and the simulation current-voltage curve are generated. A deep learning model is trained based on the basic training data such that the deep learning model outputs a prediction current-voltage curve. A target prediction current-voltage curve is generated based on the deep learning model and target reference points corresponding to the target semiconductor product. The deep learning model is a generative adversarial network.
1. A method of predicting characteristics of a target semiconductor device included in a target semiconductor product, the method being performed by at least one processor executing program codes, the program codes stored in computer readable media, the method comprising:
generating a simulation current-voltage curve using a plurality of compact models, the plurality of compact models respectively corresponding to a plurality of process data and a plurality of semiconductor products, and each of the plurality of compact models is configured to perform a simulation based on device data such that simulation result data, indicating characteristics of semiconductor devices corresponding to the device data, is generated;
extracting a plurality of simulation reference points on the simulation current-voltage curve;
generating basic training data corresponding to a combination of the plurality of simulation reference points and the simulation current-voltage curve;
training a deep learning model based on the basic training data such that the deep learning model is configured to output a prediction current-voltage curve; and
generating a target prediction current-voltage curve based on the deep learning model and a plurality of target reference points corresponding to the target semiconductor product,
wherein the deep learning model is a generative adversarial network.
2. The method of claim 1 , wherein the deep learning model is further configured to output a discrimination value based on the simulation current-voltage curve and the target prediction current-voltage curve such that the discrimination value indicates a similarity between the simulation current-voltage curve and the prediction current-voltage curve.
3. The method of claim 2 , wherein
the discrimination value approaches 0 as the prediction current-voltage curve further deviates from the simulation current-voltage curve, and
the discrimination value approaches 1 as the prediction current-voltage curve further approaches to the simulation current-voltage curve.
4. The method of claim 3 , wherein the deep learning model is trained such that the discrimination value approaches 0.5.
5. The method of claim 1 , wherein
the target semiconductor device is a transistor, and
the simulation current-voltage curve indicates a change of a drain current of the transistor based on a change of a gate voltage of the transistor with respect to a drain voltage of the transistor.
6. The method of claim 5 , wherein
a number of the plurality of target reference points is equal to a number of the plurality of simulation reference points, and
a combination of the drain voltage and the gate voltage corresponding to each of the plurality of target reference points is equal to a combination of the drain voltage and the gate voltage corresponding to each of the plurality of simulation reference points.
7. The method of claim 1 , wherein the target semiconductor product is a new semiconductor product that is not included in the plurality of semiconductor products corresponding to the plurality of compact models.
8. The method of claim 1 , wherein
the plurality of process data is based on a process-group combination of a process type, the process type indicating a manufacturing process of each of the plurality of semiconductor products and a product group in which each of the plurality of semiconductor products is included, and
the target semiconductor product corresponds to a new process-group combination that is not included in the process-group combinations of the plurality of semiconductor products.
9. The method of claim 1 , further comprising:
retraining the deep learning model based on uncertainty data,
wherein the uncertainty data indicates uncertainty of the target prediction current-voltage curve, and
the deep learning model is configured to further output the uncertainty data.
10. The method of claim 9 , wherein the uncertainty data include a model uncertainty value indicating the uncertainty of the prediction current-voltage curve caused by insufficiency of the basic training data.
11. The method of claim 10 , wherein retraining the deep learning model includes:
comparing the model uncertainty value with a model reference value;
generating addition training data using the plurality of compact models when the model uncertainty value is larger than the model reference value; and
retraining the deep learning model based on the addition training data,
wherein the addition training data is different from the basic training data.
12. The method of claim 11 , wherein retraining the deep learning model further includes:
determining an addition data range corresponding to a range of the device data such that the model uncertainty value is larger than the model reference value, and
wherein the addition training data correspond to a combination of the plurality of simulation reference points and the simulation current-voltage curve.
13. The method of claim 11 , wherein the deep learning model is further trained based on the addition training data.
14. The method of claim 9 , wherein the uncertainty data include a data uncertainty value indicating the uncertainty of the prediction current-voltage curve caused by noises in the basic training data.
15. The method of claim 14 , wherein retraining the deep learning model includes:
comparing the data uncertainty value with a data reference value;
producing measurement data by measuring the characteristics of the target semiconductor device when the data uncertainty value is larger than the data reference value;
correcting at least one of the plurality of compact models based on the measurement data;
generating updated training data using the corrected at least one compact model; and
retraining the deep learning model based on the updated training data.
16. The method of claim 15 , wherein retraining the deep learning model further includes:
determining a measurement data range corresponding to a range of data such that the data uncertainty value is larger than the data reference value, and
wherein the measurement data are provided by measuring characteristics of the target semiconductor device corresponding to the device data included in the measurement data range.
17. The method of claim 15 , wherein the deep learning model that has been trained based on the basic training data is initialized, and the initialized deep learning model is trained based on the updated training data.
18. A method of predicting characteristics of a semiconductor device included in a target semiconductor product, the method being performed by at least one processor executing program codes, the program codes stored in computer readable media, the method comprising:
generating a simulation current-voltage curve of the semiconductor device using a plurality of compact models, the plurality of compact models respectively corresponding to a plurality of process data and a plurality of semiconductor products, each of the plurality of compact models configured to perform a simulation based on device data such that simulation result data, indicating characteristics of semiconductor devices corresponding to the device data, is generated;
extracting a plurality of simulation reference points on the simulation current-voltage curve;
generating basic training data corresponding to a combination of the plurality of simulation reference points and the simulation current-voltage curve;
training a deep learning model based on the basic training data such that the deep learning model is configured to output a prediction current-voltage curve and uncertainty data indicating uncertainty of the prediction current-voltage curve;
generating a target prediction current-voltage curve based on the deep learning model and a plurality of target reference points corresponding to the target semiconductor product; and
retraining the deep learning model based on the uncertainty data,
wherein the deep learning model is a generative adversarial network.
19. A computing device comprising:
a computer readable medium storing program codes and a plurality of compact models, the plurality of compact models respectively corresponding to a plurality of process data and a plurality of semiconductor products, each of the plurality of compact models configured to perform a simulation based on device data such that simulation result data, indicating characteristics of semiconductor devices corresponding to the device data, is generated; and
at least one processor configured to, when executing the program codes,
generate a simulation current-voltage curve of a semiconductor device using the plurality of compact models,
extract a plurality of simulation reference points on the simulation current-voltage curve,
generate basic training data corresponding to a combination of the plurality of simulation reference points and the simulation current-voltage curve,
train a deep learning model based on the basic training data such that the deep learning model is configured to output a prediction current-voltage curve, and
generate a target prediction current-voltage curve based on the deep learning model and a plurality of target reference points corresponding to a target semiconductor product,
wherein the deep learning model is a generative adversarial network.
20. The computing device of claim 19 , wherein the deep learning model is further configured to output a discrimination value based on the simulation current-voltage curve and the prediction current-voltage curve such that the discrimination value indicates similarity between the simulation current-voltage curve and the target prediction current-voltage curve.