Physical Unclonable Functions (PUF) is a kind of hardware security primitive which is vulnerable to machine learning attacks. This work proposed a structural adjustment method based on Arbiter PUF (APUF), with the Strict Avalanche Criterion (SAC) as a guideline, improves the SAC property by leading multiple arbiters to obfuscation circuits at intermediate positions; quantitative analysis of SAC property by optimization method leads to the best placement of arbiters, which improves the utilization of resources. FPGA experiment results show that the proposed method significantly improves PUF’s resistance to machine learning attack such as logistic regression, evolutionary strategies and deep neural networks, and requires lower hardware resources overhead compared to other similar APUF based schemes with the same attack resistance.