Secure Hash Algorithm (SHA) is the most popular standard of Hash functions. Several security protocols use SHA to provide message integrity, authentication and digital signature. Nowadays, a new technology based on Chaotic Neural Network is used to design Hash functions due to the following important properties of Neural Networks: non-linearity, compression, confusion and diffusion. This paper presents an enhancement of our previously proposed Hash function based on a Chaotic Neural Network (in term of complexity)[1]. The theoretical analysis and the obtained experimental performances demonstrate the efficiency of the implemented structure in terms of strong Collision Resistance and High Message Sensitivity compared with SHA-2.