ABSTRACT
In the previous chapter, we introduced Physically Unclonable Function (PUF) circuits, discussed some machine learning based attacks on them. In this chapter, we take a different approach to modelling PUF circuits. One of the features of modelling PUF circuits through machine learning techniques is that except for arbiter PUFs [220], other PUFs cannot be modelled very satisfactorily in a way to suggest which machine learning to apply to model them. Thus, we should concentrate on developing techniques that model an arbitrary given PUF instance accurately and with little computational effort. This observation suggests heuristic techniques which are effective in estimating input-output relationships when the nature of the data is discrete, and the relationship is either unknown or exceedingly complex.
