Evaluating an object

A nontransitory computer program product that stores instructions for evaluating an object, the instructions comprising the steps of:

• acquiring an image of an area of a reference object;
• wherein the area comprises multiple instances of a structure of interest, and the structure of interest is of a nanometric scale;
• determining multiple types of attributes from the image;
• reducing a number of the attributes to provide reduced attribute information;
• generating guidelines, based on the reduced attribute information and on reference data, for evaluating the reduced attribute information;
• and
• evaluating an actual object by implementing the guidelines.

A method for measuring depths and/or heights of structures on a substrate, the method comprising:

• acquiring, by a charged particle beam system, a backscattered electron image of an area of a reference object, wherein the area comprises multiple instances of a structure of interest, the structure of interest is positioned in a recess having an aspect ratio that exceeds ten, and the structure of interest is of a nanometric scale;
• determining multiple types of attributes from the image;
• reducing a number of the attributes to provide reduced attribute information;
• generating guidelines, based on the reduced attribute information and on reference data, for evaluating the reduced attribute information, wherein generating the guidelines includes training a neural network to output data that approximates the reference data when fed with the reduced attribute information;
• and
• evaluating an actual object by implementing the guidelines.