Due to the tamperproof and lightweight nature

physical unclonable function are proposed to provide security with low cost for the internet of things. Security of PUF itself has attracted much more attention. Almost all strong PUF can be modeled using machine learning techniques

while the complicated PUF with non-linear structure

which are resistant to machine learning modeling

are vulnerable to side channel attacks. According to the unified symbol rules

the paper presents existing side channel attack methods on strong PUFs

such as reliability analysis

power analysis and fault injection. The principles

performance and applications of side channel/machine learning hybrid attack methods are elaborated and analyzed. In the end

the temporary predicaments and countermeasures of side channel attack on PUF are discussed.