GbE/PCI-X-Optimized Structured ASIC for Appliance and HBA/NIC Markets

The Netcomposer-1 (NC-1) Platform ASIC device is designed to offer a low cost, scalable metal programmable SoC solution targeted at enabling ASSPs and system ASICs requiring high performance acceleration and content inspection in hardware for PCI-X and GbE markets. The applications for NC-1 range from price sensitive SMB gateways to high performance iSCSI/TOE offload Host bus adapters (HBA). Customers will receive their chips at 1/3 the time with ½ the resources, and 1/3 NRE costs of a traditional standard cell 0.13 µm ASIC. While the customer logic/IP provides feature differentiation, the pre-diffused data path and IP system blocks such as the on-chip ARM processor, and memory controllers help facilitate tasks such as exception processing and fast data movement, while preserving valuable engineering and verification time, saving customers time, resources and expenses.

The NC-1's significant degree of functional SoC integration reduces total bill of material (BOM), power dissipation, and footprint. The chip includes a 450 MHz FA626 ARMv4-compliant CPU, a 32 KB L2 cache controller (configurable as on-chip SRAM), an MPCA TM block, 2x 10/100/1000 Mbps Ethernet MAC controllers (GMAC), a 66/133 MHz PCI-X controller supporting both initiator and target modes (backward compatible with the PCI 2.2 specification), a host DMA engine, a data coherency engine (DCE), a 150 MHz DDR-1 SDRAM memory controller, and a combo Static Memory & Flash Controller (SMC). The 68 pins connected to the MPCA block in the chip are shared with the SMC I/F. These pins can be used as programmable I/O by system designers that intend to incorporate their own proprietary memory controller design or applications that require higher densities (e.g., a 3^rd GbE MAC).All system blocks integrated in NC-1 are also available as independent IP blocks through Faraday.

The figure below shows 4 potential ASSP/ASIC applications for NC-1: Single-chip SMB Gateway, iSCSI/TOE HBA (NIC), Secure GbE HBA (NIC), and I-TDM/PWE-3 Emulation Engine.

Four Potential Applications for NC-1

The two GbE MACs (GMAC) support GMII/MII interfaces for connecting to standard 10/100/1000 Ethernet transceivers. For wire-speed processing data can travel either directly into the custom application-specific engine implemented in the MPCA block (Flow-through), or can be stored in the SDRAM and subsequently retrieval by the FA626 CPU for deep packet processing (Store and forward). Once processed through the customer application logic in MPCA, the packets can either be sent through the PCI-X I/F to the host for further analysis and higher upper layer processing, or can be directed out through the same Ethernet port (or second GbE port) towards the network. Both Ethernet MACs are fully compliant with the 802.3 specification supporting Ethernet Type II and IEEE802.3/802.2/SNAP type frames, while also supporting jumbo packets.

NC-1 supports complete IPv4 frame processing, while IPv6 frames are passed on as L2 frames. To enable QoS applications, the NC-1 Ethernet MACs support 2 priority queues (normal or high priority) which are automatically reordered for the packet transmission to save CPU overhead and bus bandwidth. Furthermore, to support high performance flow-through PCI-X processing architectures, the MPCA includes 3 direct–path VCI 2.0 ports to the PCI-X controller to take advantage of the latency in the PCI Read Cycle and enable simultaneous independent in-between Writes. Additionally, the MPCA has 4 VCI ports connected to the M-Hub switch fabric consisting of 2x 64-bit master ports, 2x 32-bit slave ports, 2x 64-bit DMA ports connecting directly to both GMACs and another 64-bit DMA port connecting directly to Host-DMA (HDMA). To enhance the system performance and to minimize the processor interferences for the data transfer, the NC-1 includes a descriptor based “scatter & gather” HDMA. There are 4 sets of DMA channels, each assigned to the CPU, PCI-X, and AHB Bus.