NURBS are one of the most common methods of representing curves and surfaces in geometric modeling and computer graphics fields whilst numerous computer graphic algorithms are based on the utilization of a simpler representation of the same curves and surfaces: B?zier representations. In this paper, we present a parallel architecture to perform the decomposition of a NURBS curve into its constituent B?zier polynomial pieces. This architecture presents a regular and easily scalable structure, suitable for VLSI implementation, which can be efficiently exploited for the decomposition process. The performance of the proposed architecture is improved by the use of carry save arithmetic, which permits the reduction of the system time cycle.