Virtex-4 RocketIO MGT User Guide www.xilinx.com 191UG076 (v4.1) November 2, 2008RChapter 8Low-Latency DesignIntroductionThe Virtex®-4 RocketIO™ Multi-Gigabyte Transceiver (MGT) has the flexibility tominimize latency in the PCS.The RocketIO transceiver contains two buffers, one for the transmitter and one for thereceiver. The RX buffer is a ring buffer that supports clock correction and channel bonding.Specifically, these buffers allow for phase differences between PCS RXCLK andRXUSRCLK, as shown in Figure 8-1, or between PCS TXCLK and TXUSRCLK, as shown inFigure 8-2.Some applications do not require the full feature set of the PCS, and a user can choose tominimize latency from/to the serial pins (RXP/RXN, TXP/TXN). For these applications,the MGT includes parallel clock synchronization features that align the PMA-generatedparallel clock (PMA XCLK0) to a parallel clock sourced from the PCS (PCS XCLK), makingit possible to bypass the RX buffer and TX buffer. The transmitter and receiver data pathsinclude PCS bypass muxes that allow PCS features to be bypassed in these clocking modes.Refer to Figure 8-1 and Figure 8-2 for the buffer bypass muxing options selected by theRXDATA_SEL and TXDATA_SEL attributes respectively.For the RX, latency reduction is achieved by using the RX Buffer (RX Low Latency BufferedMode) or by bypassing the RX Buffer (RX Low Latency Buffer Bypass Mode). The lattermode supports the low-latency data path bypass options selected via RXDATA_SEL:• RXDATA_SEL = 00 — full data path• RXDATA_SEL = 01 — data directly from PMA interface• RXDATA_SEL = 10 — data directly from alignment block• RXDATA_SEL = 11 — data directly from 8B/10B decoderFor the TX, latency reduction is achieved by bypassing features of the PCS (TX LowLatency Buffered Mode) or by bypassing the TX Buffer and/or PCS features (TX LowLatency Buffer Bypass Mode). These reduced-latency bypass options are selected viaTXDATA_SEL:• TXDATA_SEL = 00 — full data path• TXDATA_SEL = 01 — data directly from fabric interface• TXDATA_SEL = 10 — data directly from output of 8B/10B encoder
