2-133
Performance
Device Performance
The XC3000 families of FPGAs can achieve very high
performance. This is the result of
• A sub-micron manufacturing process, developed and
continuously being enhanced for the production of
state-of-the-art CMOS SRAMs.
• Careful optimization of transistor geometries, circuit
design, and lay-out, based on years of experience
with the XC3000 family.
• A look-up table based, coarse-grained architecture
that can collapse multiple-layer combinatorial logic
into a single function generator. One CLB canimple-
ment up to four layers of conventional logic in as little
as 2.7 ns.
Actual system performance is determined by the timing of
critical paths, including the delay through the combinato-
rial and sequential logic elements within CLBs and IOBs,
plus the delay in the interconnect routing. The ac-timing
specifications state the worst-case timing parameters for
the various logic resources available in the XC3000-
families architecture. Figure 25 shows a variety of ele-
ments involved in determining system performance.
Logic block performance is expressed as the propagation
time from the interconnect point at the input to the block to
the output of the block in the interconnect area. Since
combinatorial logic is implemented with a memory lookup
table within a CLB, the combinatorial delay through the
CLB, called T
ILO
, is always the same, regardless of the
function being implemented. For the combinatorial logic
function driving the data input of the storage element, the
critical timing is data set-up relative to the clock edge
provided to the flip-flop element. The delay from the clock
source to the output of the logic block is critical in the timing
signals produced by storage elements. Loading of a logic-
block output is limited only by the resulting propagation
delay of the larger interconnect network. Speed perfor-
mance of the logic block is a function of supply voltage and
temperature. See Figure 26.
Interconnect performance depends on the routing re-
sources used to implement the signal path. Direct inter-
connects to the neighboring CLB provide an extremely fast
path. Local interconnects go through switch matrices
(magic boxes) and suffer an RC delay, equal to the
resistance of the pass transistor multiplied by the capaci-
tance of the driven metal line. Longlines carry the signal
across the length or breadth of the chip with only one
access delay. Generous on-chip signal buffering makes
performance relatively insensitive to signal fan-out; in-
creasing fan-out from 1 to 8 changes the CLB delay by only
10%. Clocks can be distributed with two low-skew clock
distribution networks.
The tools in the XACT Development System used to place
and route a design in an XC3000 FPGA (the Automatic
Place and Route [APR] program and the XACT Design
Editor)automatically calculate the actual maximum worst-
case delays along each signal path. This timing informa-
tion can be back-annotated to the design’s netlist for use
in timing simulation or examined with X-DELAY, a static
timing analyzer.
Actual system performance is applications dependent.
The maximum clock rate that can be used in a system is
determined by the critical path delays within that system.
These delays are combinations of incremental logic and
routing delays, and vary from design to design. In a
synchronous system, the maximum clock rate depends on
the number of combinatorial logic layers between re-
synchronizing flip-flops. Figure 27 shows the achievable
clock rate as a function of the number of CLB layers.
Figure 25. Primary Block Speed Factors. Actual timing is a function of various block factors combined with routing
factors. Overall performance can be evaluated with the XACT timing calculator or by an optional simulation.
CLBCLB IOBCLB
PAD
(K)
LogicLogic
CKO
T
CLOCK
Clock to Output
Combinatorial Setup
T
CKO
T
ILO
T
ICK
(K)
PAD
IOB
T
PID
T
OKPO
OP
T
X3178
