R
Functional Description
Register Cascade Feature
In the Spartan-3E family, one of the IOBs in a differential
pair can cascade either its input or output storage elements
with those in the other IOB of the differential pair. This is
intended to make DDR operation at high speed much sim-
pler to implement. The new DDR connections that are avail-
able are shown in Figure 1 (dashed lines), and are only
available for routing between IOBs and are not accessible to
the FPGA fabric. Note that this feature is only available
when using differential I/O.
Q
Q
D
D
D1
PAD
To Fabric
D2
IDDRIN2
IQ2
Q
D
ICLK1
ICLK2
IDDR2
As a DDR input pair, the master IOB registers incoming
data on the rising edge of ICLK1 (= D1) and the rising edge
of ICLK2 (= D2), which is typically the same as the falling
edge of ICLK1. This data is then transferred into the FPGA
fabric. At some point, both signals must be brought into the
same clock domain, typically ICLK1. This can be difficult at
high frequencies because the available time is only one half
of a clock cycle assuming a 50% duty cycle. See Figure 4
for a graphical illustration of this function.
ICLK1
ICLK2
d
d+1 d+2 d+3 d+4 d+5 d+6 d+7 d+8
PAD
D1
d
d+2
d+1
d+4
d+3
d+6
d+5
d+8
d+7
D2
d-1
DS312-2_22_030105
In the Spartan-3E device, the signal D2 can be cascaded
into the storage element of the adjacent slave IOB. There it
is re-registered to ICLK1, and only then fed to the FPGA
fabric where it is now already in the same time domain as
D1. Here, the FPGA fabric uses only the clock ICLK1 to pro-
cess the received data. See Figure 5 for a graphical illustra-
tion of this function.
Figure 5: Input DDR Using Spartan-3E Cascade Feature
ODDR2
As a DDR output pair, the master IOB registers data coming
from the FPGA fabric on the rising edge of OCLK1 (= D1)
and the rising edge of OCLK2 (= D2), which is typically the
same as the falling edge of OCLK1. These two bits of data
are multiplexed by the DDR mux and forwarded to the out-
put pin. At some point in the FPGA fabric, the signal D2
must be brought into the clock domain OCLK2 from the
domain OCLK1. This can be difficult at high frequencies,
because the time available is only one half a clock cycle.
See Figure 6 for a graphical illustration of this function.
Q
Q
D
D
D1
PAD
To Fabric
D2
In the Spartan-3E device, the signal D2 can be cascaded
via the storage element of the adjacent slave IOB. Here, it is
registered by OCLK1 and then forwarded to the master IOB
where it is re-registered to OCLK2, selected as usual by the
DDR multiplexer, and then forwarded to the output pin. This
way the data for transmission can be processed using just
the clock OCLK1 in the FPGA fabric. See Figure 7 for a
graphical illustration of this function.
ICLK2
ICLK1
ICLK1
ICLK2
PAD
D1
d
d+1 d+2 d+3 d+4 d+5 d+6 d+7 d+8
d+2 d+4 d+6
d
d+8
d+7
D2 d-1
d+1
d+3
d+5
DS312-2_21_021105
Figure 4: Input DDR (without Cascade Feature)
6
www.xilinx.com
DS312-2 (v1.1) March 21, 2005
Advance Product Specification
XILINX [ XILINX, INC ]
