Method and apparatus for hierarchical wafer quality predictive modeling

1. A method for performing enhanced wafer quality prediction in a semiconductor manufacturing process, the method comprising steps of:

obtaining, with a processor, data including at least one of tensor format wafer processing conditions, historical wafer quality measurements and prior knowledge relating to at least one of the semiconductor manufacturing process and wafer quality;

building, with said processor, a hierarchical prediction model including at least the tensor format wafer processing conditions; and

predicting, with said processor, wafer quality for a newly fabricated wafer based at least on the hierarchical prediction model and corresponding tensor format wafer processing conditions.

2. The method of claim 1 , wherein the step of building the hierarchical prediction model comprises:

decomposing higher-order weight tensors into lower-order tensors;

approximating the lower-order tensors based at least in part on the prior knowledge; and

minimizing a prediction error and optimizing an approximation of the lower-order tensors as a function of the prediction error.

3. The method of claim 2 , wherein the step of decomposing higher-order weight tensors into lower-order tensors is performed using a canonical polyadic decomposition.

4. The method of claim 2 , wherein the step of decomposing higher-order weight tensors into lower-order tensors comprises approximating each of the higher-order weight tensors by a rank-R tensor with a canonical polyadic decomposition, where R is indicative of a number of sources from which domain knowledge is obtained.

5. The method of claim 4 , wherein R is greater than one.

6. The method of claim 2 , wherein the step of optimizing an approximation of the lower-order tensors is performed using a block coordinate decent methodology.

7. The method of claim 2 , further comprising updating at least one of the tensor format wafer processing conditions, historical wafer quality measurements and prior knowledge as a function of an approximation of the lower-order tensors.

8. The method of claim 1 , wherein the step of building a hierarchical prediction model comprises minimizing a loss function L(y n , χ n , ) summed over all training samples, where y n is a response variable for a regression, χ n is a K th -order tensor, where K is an integer indicative of a number of modes of the tensor χ n , n is a training sample number, and is a weight tensor.

9. The method of claim 1 , wherein the step of building the hierarchical prediction model is performed using tensor inputs without converting the tensor inputs to corresponding vectors.

10. The method of claim 1 , further comprising utilizing an output generated by at least one of the step of predicting wafer quality and the hierarchical prediction model to perform advanced process control (APC) relating to the semiconductor manufacturing process.

11. The method of claim 1 , further comprising utilizing an output generated by at least one of the step of predicting wafer quality and the hierarchical prediction model to perform wafer quality fault detection and classification (FDC).

12. The method of claim 1 , further comprising utilizing an output generated by at least one of the step of predicting wafer quality and the hierarchical prediction model to determine actual metrology sampling frequency.

13. The method of claim 1 , wherein the prior knowledge relating to at least one of the semiconductor manufacturing process and wafer quality comprises information from a plurality of sources.

14. The method of claim 1 , wherein the step of building the hierarchical prediction model comprises utilizing rich information embedded in a tensor structure of the tensor format wafer processing conditions.

15. The method of claim 1 , wherein the step of building the hierarchical prediction model comprises representing feature correspondence in one or more steps of the semiconductor manufacturing process.