This paper explores the integration of sensing capabilities into millimeter wave (mmWave) wireless communication networks, focusing on the use of reconfigurable intelligent surfaces (RIS) to enhance both sensing and communication performance. The authors derive the Cramér-Rao bound (CRB) for estimating the direction-of-angles in a semi-self-sensing RIS approach, where sensors are configured at the RIS to receive radar echo signals. To minimize the CRB while maintaining satisfactory communication performance, a joint optimization problem on hybrid beamforming and RIS phase shifts is proposed. The problem is transformed into a more tractable form using the Fisher information matrix (FIM), and a double-layer loop algorithm based on penalty-CCCP and block coordinate descent (BCD) methods is developed to solve the non-convex problem. The proposed algorithm is validated through numerical results, demonstrating improved sensing performance compared to other baselines and achieving approximately 96% of the sensing performance of full digital approaches with limited RF chains.This paper explores the integration of sensing capabilities into millimeter wave (mmWave) wireless communication networks, focusing on the use of reconfigurable intelligent surfaces (RIS) to enhance both sensing and communication performance. The authors derive the Cramér-Rao bound (CRB) for estimating the direction-of-angles in a semi-self-sensing RIS approach, where sensors are configured at the RIS to receive radar echo signals. To minimize the CRB while maintaining satisfactory communication performance, a joint optimization problem on hybrid beamforming and RIS phase shifts is proposed. The problem is transformed into a more tractable form using the Fisher information matrix (FIM), and a double-layer loop algorithm based on penalty-CCCP and block coordinate descent (BCD) methods is developed to solve the non-convex problem. The proposed algorithm is validated through numerical results, demonstrating improved sensing performance compared to other baselines and achieving approximately 96% of the sensing performance of full digital approaches with limited RF chains.