
14
Agere Systems Inc.
Preliminary Data Sheet
July 2001
1.0-1.25/2.0-2.5/3.125 Gbits/s Backplane Interface
ORCA
ORT82G5 FPSC Eight-Channel
ORT82G5 Overview
(continued)
The ORT82G5 FPSC combines 8 channels of high-
speed full duplex serial links (up to 3.125 Gbits/s) with
400k usable gate FPGA. The major functional blocks in
the ASB core are two quad-channel serializer-deserial-
izers (SERDES) including 8b/10b encoder/decoder and
dedicated PLLs, XAUI or fibre-channel link-state-
machine, 4-to-1 or 1-to-4 MUX/deMUX, multichannel
alignment FIFO, microprocessor interface, and 4k x 36
RAM blocks.
Serializer and Deserializer (SERDES)
The SERDES block is a quad transceiver for serial data
transmission, with a selectable data rate of 1.0
—
1.25 Gbits/s, 2.0
—
2.5 Gbits/s, or 3.125 Gbits/s. It is
designed to operate in Ethernet, fibre channel,
Firewire
,
or backplane applications. It features high-
speed 8b/10b parallel I/O interfaces, and high-speed
CML interfaces.
The quad transceiver is controlled and configured with
an 8 bit microprocessor interface through the FPGA.
Each channel has dedicated registers that are readable
and writable. The quad device also contains global reg-
isters for control of common circuitry and functions.
For complete SERDES description, please refer to the
Macrocell Data Sheet,
LU6X14FT1.0-1.25/2.0-2.5/
3.125 Gbits/s Serializer and Deserializer
.
8b/10b Encoding/Decoding
The ORT82G5 facilitates high-speed serial transfer of
data in a variety of applications including Gbit Ethernet,
fibre channel, serial backplanes, and proprietary links.
The SERDES provides 8b/10b coding/decoding for
each channel. The 8b/10b transmission code includes
serial encoding/decoding rules, special characters, and
error detection.
In the receive direction, the user can disable the 8b/10b
decoder to receive raw 10 bit words which will be rate
reduced by the SERDES. If this mode is chosen, the
user must bypass the multichannel alignment FIFOs. In
the transmit direction, the 8b/10b encoder must always
be enabled (version II will allow it to be disabled).
Clocks
The SERDES block contains its own dedicated PLLs
for transmit and receive clock generation. The user pro-
vides a reference clock of the appropriate frequency.
The receiver PLLs extract the clock from the serial
input data and retime the data with the recovered clock.
MUX/DeMUX Block
The purpose of the MUX/deMUX block is to provide a
wide, low-speed interface at the FPGA portion of the
ORT82G5 for each channel or data lane.
The interface to the SERDES macro runs at 1/10th the
bit rate of the data lane. The MUX/deMUX converts the
data rate and bit-width so the FPGA core can run at
1/4th this frequency. This implies a range of
25
—
78 MHz for the data in and out of the FPGA.
The MUX/deMUX block in the ORT82G5 is a 4-channel
block. It provides an interface between each quad
channel SERDES and the FPGA logic.
Multichannel Alignment FIFOs
The ORT82G5 has a total of 8 channels (4 per SER-
DES). The incoming data of these channels can be
synchronized in several ways, or they can be indepen-
dent of one other.
For example, all four channels in a SERDES can be
aligned together to form a communication channel with
a bandwidth of 10 Gbits/s.
Alternatively, two channels within a SERDES can be
aligned together; channel A and B and/or channel C
and D.
Optionally, the alignment can be extended across SER-
DES to align all 8 channels.
Individual channels within an alignment group can be
disabled (i.e., power down) without disrupting other
channels.
XAUI or Fibre-Channel Link State Machine
Two separate link state machines are included in the
ORT82G5. A XAUI compliant link state machine is
included in the embedded core to implement the
IEEE
802.3ae v2.1 standard. A separate state machine for
fibre-channel/
Infiband
is also provided.
Dual Port RAMs
There are two independent memory blocks in the ASB.
Each memory block has a capacity of 4k word by
36 bits. It has one read port, one write port, and four
byte-write-enable (active-low) signals. The read data
from the memory block is registered so that it works as
a pipelined synchronous memory block.