Many types of physical unclonable function (PUF) structures have been proposed in the last decade. The responses generated from the conventional PUF are vulnerable to attack. In this chapter, the transient effect of ring oscillator structure has been used. This works on two loops with complex loops containing NOT gates and NAND gates. Response prediction of these loops is a very difficult task for the adversary. Many machine learning algorithms may produce the responses with higher accuracies. This study provides new masked PUF architectures that are more secure and invulnerable to modeling attacks. Hence, in this chapter, masking-based configurability design on various PUF structures is introduced. This will be helpful for resource-constrained machines. For different sizes of challenge-response pair, machine learning techniques need to be changed, but prediction accuracy by the attacker should be low. By using this kind of masked PUF structure, 54.7% uniqueness can be obtained, and 97.5% reliability can be achieved. Machine learning accuracy is 70.7% with SVM and 63.67% accuracy in LR.