In the ATLAS experiment readout of front-end detector and trigger electronics is achieved via the so-called RODs (ReadOut Drivers). These receive data via dedicated point-to-point links, process these data and output the resulting data via other dedicated point-to-point links to the ROS (ReadOut System). The event data is pushed through the sub-detector specific RODs, built from custom hardware and which also may provide support for calibration and control. The functionality of the ROS is detector independent. Event data received from the RODs is buffered and forwarded upon request to the Higher-Level Trigger farm via a commodity Ethernet network. For new detector and trigger systems to be installed after LHC Run 2 a new approach will be followed: a new system, the FELIX (Front-End Link eXchange) system will interface to links connecting to front-end detector and trigger electronics. The links are bi-directional and typically will make use of GBT link technology to transfer LHC clock and fixed latency trigger accept signals and associated reset signals to the front-end electronics via dedicated “e-links” (of which up to 42 can be time-multiplexed on a single GBT link). For this purpose FELIX interfaces to the current TTC system. Other e-links will be used for transferring control and configuration commands to the front-end electronics sent to the FELIX system via a switched network built using standard technology (either Ethernet or Infiniband) to which the FELIX system is connected. Responses to these commands and event data will flow via e-links into the FELIX system and will be forwarded via the network to server PCs. For off-detector trigger systems instead of the GBT protocol typically simple 8B/10B coding will be used for data transfers toward the FELIX system, with links making use of standard FPGA transceivers. These links may run at higher clock rates than the standard GBT links. The functionality currently provided by the RODs and also additional control, monitoring, calibration and configuration functionality will be implemented in software running on the server PCs connecting to the network.