Internet of things (IoT) devices become ubiquitous, interconnected platforms for everyday tasks, which dictate a growing demand for low-cost security primitives. Physically Unclonable Functions (PUFs) are one of the promising solutions for low-cost key storage and device authentication, where strong PUFs [1-5] are suitable for authentication due to the exponentially large challenge-response pairs (CRPs) space. Early strong PUFs were vulnerable to machine learning (ML) attacks [3], [4], while [1], [2], [5] introduce various nonlinear entropy cells to enhance resilience. However, they all suffer from low energy efficiency because many trivial entropy cells need to be activated for sufficient nonlinearity. Besides, with many enabled cells, a small number of challenge bits flipping only imposes a very small probability for the change on the final response, resulting in a poor standard deviation on their Hamming Weight (HW).