The need for an adequate fault injector for CAN has leaded to the proposal or use of different testing tools. However, all these aids present important limitations. The most important one is that none of them can generate local errors independently in the signals received and transmitted by each one of the nodes and, thus, they cannot force specific an complex scenarios; for example, the scenarios leading to inconsistencies between nodes. To overcome these limitations, we have designed and implemented sfiCAN, a star-based physical fault-injection infrastructure for CAN. In sfiCAN the bus topology CAN relies on is substituted by a star topology whose central element is a hub. More specifically, sfiCAN is based on the architecture of CANcentrate, a CAN-compliant simplex star topology previously proposed is logically equivalent to a CAN bus and transparent from the nodes point of view. Moreover, note that each node is connected to the hub by means of a dedicated link containing an uplink and a downlink. The main advantage of this architecture is that the hub becomes a central element that has a privileged view of the communication, knowing the stream transmitted and received by each node, bit by bit. This allows sfiCAN to use the hub to inject errors independently in every single bit in the transmitted and/or received signal of each node and, thus, to force complex error scenarios. Moreover, the use of the hub as a centralized element yields other important testability advantages.