A High-Density MAC and CIDR Table Processor
iFlow™ Address Processor Product Brief
www.siliconaccess.com
Figure 1: iFlow Address Processor Block Diagram
DESCRIPTION
As netwo rk dat a rates ha ve in cre as ed to 1 0 g iga bi ts per
second and higher, performance demands on the
address lookup function have soared to more than 50
million loo ku ps per second. Software-based solu tio ns
cannot keep up, forcing designers to employ expensive
and power-hungry , terna r y-CAM s (TCAM s). The iFl ow
Address Processor (iAP) provides a complete
subsystem-on-a-chip solution for address lookups that
solves many of the problems associated with standard
TCAMs and TCAM-based coprocessor products.
The iAP is a high-density address and flow search
engine for exact or longest-prefix matches on
packetized data. With a peak rate of 65 million lookups
per second, the iAP ca n p erfo r m m or e th an tw o l oo kup s
per minimum length packet at OC-192 speeds. An on-
chip table size of 256K IPv4 entries, combined with
integrated memory, delivers an unmatched level of
integration.
FEATURES
BLOCK DIAGRAM
• Single-chip Layer 2/3/4 lookup coprocessor that
integrates the equivale nt of a 9-Mbit TCAM, tw o levels
of associa ted me mory, statis tic s eng ine and interf ac e
FPGA/ASIC
• 65 mill ion look ups per sec ond with determinis tic resul t
latency and no key-size penalties
• Supports 16 ind ependent ta bles with conf igurab le key
sizes
• Expandable to as many as eight iAPs providing up to
two million IPv4 table entries
• Supports MAC, N-tuple flows, IPv4 and IPv6 (with
VPN tag) lookups without requiring bank
segmentation
• Two lev els of i nteg rate d as s oci ate d da ta enables on e
lookup to provide the search result plus up to 48 bits
of per-entry associated data and 256 bits of per-hop
associated data
• 64 concurre nt look up cont exts supp ort multi -thr eaded
NPUs such as the iFlow Packet Processor (iPP)
• On-the-fl y modifi cation of ass ociat ed data fields us ing
on-chip ALU (performed in parallel with lookup)
• Table management in hardware (transparent to user)
• Up to one mi llion i nserts o r deletes per sec ond with no
impact on lookup rates
• Innova tive powe r manage ment requ ires les s than 5W
(versus up to 20W required for non-integrated
solutions)
• ECC protection on all memories for unmatched
reliability
• 32-, 64-, or 96-bit, 83-133MHz Zero Bus Turnaround
(ZBT) or Double Cycle Deselect (DCD) synchronous
SRAM interface
• Standard 0. 13 micron, 1.2V CMOS process
• 520-ball BGA package
Servi ce Rate
Manager
Request
Assembler
Memory Result Buffer Memory
Lookup
Table
Per-Hop
Associated
Memory
ZBT or DCD 32-, 64-, 96-bit @ 83-133 MHz
Pipelined Lookup
ALU
Associated
Memory
iFlow Address Processor