Detection system, compensation method, and computer readable medium for semiconductor surface morphology

A detection system used for semiconductor surface morphology, comprising:

• an optical measurement device for scanning a target detection region with an illuminating light beam and obtaining a light interference state formed by a reflection light from the target detection region;
• a morphological measurement device for obtaining an actual height of the target detection region;
• and
• a computation processing device coupled to the optical measurement device and the morphological measurement device and configured to perform:
• a first training step, performed on condition that a plurality of standard objects under test are each provided to the optical measurement device, wherein the computation processing device obtains a light interference state from the optical measurement device, and transforms the light interference state into an abnormal signal of each of the standard objects under test, the abnormal signal having a variable spectral state, and transforms the light interference state into a normal signal of each of the standard objects under test, the normal signal having an invariable spectral state, and uses a plurality of abnormal eigenvalues corresponding to the abnormal signals and a plurality of normal eigenvalues corresponding to the normal signals as input vectors, and providing the input vectors to a first neural network model for undergoing training;
• a second training step, performed on condition that the standard objects under test are each provided to the optical measurement device, wherein the computation processing device cooperates with the morphological measurement device to obtain an actual height of surface morphology of each of the standard objects under test, and obtains a light interference state from the optical measurement device, transforming the light interference state into the abnormal signals having variable spectral states of the standard objects under test, a corresponding abnormal eigenvalue, and a corresponding abnormal height difference, wherein the computation processing device obtains a light interference state from the optical measurement device, and transforms the light interference state into the normal signals having invariable spectral states of the standard objects under test, a corresponding normal eigenvalue, and a corresponding normal height difference, wherein the computation processing device uses the abnormal signals, normal signals, abnormal eigenvalues, normal eigenvalues, abnormal height difference and normal height difference as input vectors, and providing the input vectors to a second neural network model for undergoing training, wherein the abnormal height difference and the normal height difference are differences between a detected height of each signal and the actual height of the signal;
• an analysis step, performed on condition that an object under test exists, wherein the computation processing device allows the optical measurement device to obtain a plurality of interference signals formed by a reflection light within the target detection region of the object under test, and transforms the interference signals into a plurality of spectral signals, the computation processing device provides the to-correct spectral signal to the first neural network model and the second neural network model when at least one to-correct spectral signal having variability exists in the spectral signals to obtain a first feature information that includes waveform class information and a second feature information that includes compensation degree information, and the computation processing device combines the first feature information and the second feature information to generate a compensation information according to an abnormal feature contained in the to-correct spectral signal;
• and
• a compensation step, wherein the computation processing device corrects according to the compensation information, a to-correct height corresponding to the to-correct spectral signal to become a corrected height to correctly create surface morphology of the target detection region of the object under test.