This paper proposes a CRB minimization algorithm for reconfigurable intelligent surface (RIS)-aided millimeter-wave (mmWave) integrated sensing and communications (ISAC) systems. The goal is to enhance sensing accuracy while maintaining satisfactory communication performance. The CRB, which represents the lower bound of the variance of an unbiased estimator, is derived as a metric for sensing performance. A joint optimization problem is formulated to minimize the CRB while ensuring the achievable data rate meets a threshold. The problem is transformed into a more tractable form using the Fisher information matrix (FIM), and a double-layer loop algorithm based on penalty concave-convex procedure (penalty-CCCP) and block coordinate descent (BCD) is proposed to solve the complex non-convex problem. Successive convex approximation (SCA) and second-order cone (SOC) constraints are used to handle non-convexity in hybrid beamforming optimization. Manifold optimization (MO) is applied to optimize the analog beamforming and phase shifts. Numerical results show that the proposed algorithm effectively minimizes the CRB and improves sensing performance compared to other baselines. Additionally, the hybrid beamforming algorithm achieves approximately 96% of the sensing performance of the full digital approach with a limited number of RF chains. The algorithm is shown to converge and achieve good performance in terms of both sensing and communication metrics.This paper proposes a CRB minimization algorithm for reconfigurable intelligent surface (RIS)-aided millimeter-wave (mmWave) integrated sensing and communications (ISAC) systems. The goal is to enhance sensing accuracy while maintaining satisfactory communication performance. The CRB, which represents the lower bound of the variance of an unbiased estimator, is derived as a metric for sensing performance. A joint optimization problem is formulated to minimize the CRB while ensuring the achievable data rate meets a threshold. The problem is transformed into a more tractable form using the Fisher information matrix (FIM), and a double-layer loop algorithm based on penalty concave-convex procedure (penalty-CCCP) and block coordinate descent (BCD) is proposed to solve the complex non-convex problem. Successive convex approximation (SCA) and second-order cone (SOC) constraints are used to handle non-convexity in hybrid beamforming optimization. Manifold optimization (MO) is applied to optimize the analog beamforming and phase shifts. Numerical results show that the proposed algorithm effectively minimizes the CRB and improves sensing performance compared to other baselines. Additionally, the hybrid beamforming algorithm achieves approximately 96% of the sensing performance of the full digital approach with a limited number of RF chains. The algorithm is shown to converge and achieve good performance in terms of both sensing and communication metrics.