RTAX-S/SL RadTolerant FPGAs
v5.4 1-5
Embedded Memory
As mentioned earlier, each core tile has either three (in a
smaller tile) or four (in the regular tile) embedded SRAM
blocks along the west side, and each variable-aspect-
ratio SRAM block is 4,608 bits in size. Available memory
configurations are: 128x36, 256x18, 512x9, 1kx4, 2kx2 or
4kx1 bits. The individual blocks have separate read and
write ports that can be configured with different bit
widths on each port. For example, data can be written in
by eight and read out by one.
In addition, every SRAM block has an embedded FIFO
control unit. The control unit allows the SRAM block to
be configured as a synchronous FIFO without using core
logic modules. The FIFO width and depth are
programmable. The FIFO also features programmable
ALMOST-EMPTY (AEMPTY) and ALMOST-FULL (AFULL)
flags in addition to the normal EMPTY and FULL flags. In
addition to the flag logic, the embedded FIFO control
unit also contains the counters necessary for the
generation of the read and write address pointers as well
as control circuitry to prevent metastability and
erroneous operation. The embedded SRAM/FIFO blocks
can be cascaded to create larger configurations.
The FIFO control unit was not implemented with SEU-
hardened registers. Designs requiring high SEU tolerance
should implement the FIFO control unit from hardened
core logic.
SRAM structures are inherently susceptible to upsets
caused by high-energy particles encountered in space.
High-energy particles can cause an SRAM cell to change
state, resulting in the loss or corruption of a valuable
data bit. Actel has enhanced the SEU tolerance of the
embedded SRAM within RTAX-S/SL by employing the use
of two upset-mitigation techniques:
• Actel has developed Error Detection and Correction
(EDAC) IP for use with RTAX-S/SL. EDAC can be
accomplished by the use of SmartGen-generated
Error Correcting Codes (ECC) IP, which employs the
use of shortened Hamming Codes
• A background memory-refresher, or scrubber
circuitry, which has been embedded into the EDAC IP.
The embedded scrubber circuitry periodically
refreshes memory in the background to ensure that
no data corruption occurs while the memory is not in
use.
The use of EDAC IP combined with the embedded
memory scrubber circuitry, gives the RTAX-S/SL an SEU
radiation performance level of better than 10-10 errors/
bit-day. See the application note Using EDAC RAM for
RadTolerant RTAX-S/SL FPGAs and Axcelerator FPGAs.
I/O Logic
The RTAX-S/SL family of FPGAs features a flexible I/O
structure, supporting a range of mixed voltages with its
bank-selectable I/Os: 1.5 V, 1.8 V, 2.5 V, and 3.3 V. In all,
RTAX-S/SL FPGAs support at least 14 different I/O
standards (single-ended, differential, voltage-
referenced). The I/Os are organized into banks, with
eight banks per device (two per side). The configuration
of these banks determines the I/O standards supported
(see "User I/Os" on page 2-12 for more information). All
I/O standards are available in each bank.
Each I/O module has an input register (InReg), an output
register (OutReg), and an enable register (EnReg)
(Figure 1-7 on page 1-6). An I/O Cluster includes two I/O
modules, four RX modules, two TX modules, and a buffer
(B) module.
By design, all user flip-flops in the RTAX-S FPGAs are
immune to SEUs including the following three registers
located in every I/O cell buffer: InReg, OutReg, and
EnReg.
Routing
The RTAX-S/SL hierarchical routing structure ties the logic
modules, the embedded memory blocks, and the I/O
modules together (Figure 1-8 on page 1-6). At the lowest
level, in and between SuperClusters, there are three local
routing structures: FastConnect, DirectConnect, and
CarryConnect routing. DirectConnects provide the highest
performance routing inside the SuperClusters by
connecting a C-cell to the adjacent R-cell. DirectConnects
do not require an antifuse to make the connection and
achieve a signal propagation time of less than 0.1 ns.
FastConnects provide high-performance, horizontal
routing inside the SuperCluster and vertical routing to
the SuperCluster immediately below it. Only one
programmable connection is used in a FastConnect path,
delivering a maximum routing delay of 0.4 ns.
CarryConnects are used for routing carry logic between
adjacent SuperClusters. They connect the carry-logic FCO
output of one C-cell pair to the carry-logic FCI input of
the C-cell pair of the SuperCluster below. CarryConnects
do not require an antifuse to make the connection and
achieve a signal propagation time of less than 0.1 ns.
The next level contains the core tile routing. Over the
SuperClusters within a core tile, both vertical and
horizontal tracks run across rows or columns,
respectively. At the chip level, vertical and horizontal
tracks extend across the full length of the device, both
north-to-south and east-to-west. These tracks are
composed of highway routing that extend the entire
length of the device (segmented at core tile boundaries)
as well as segmented routing of varying lengths.