Curtiss-Wright’s Defense Solutions introduced the CHAMP-XD3, its highest performance, security-enhanced, 3U OpenVPX digital signal processing (DSP) processing module. Based on the just-announced Intel XeonD-1700 processor, the SOSA-aligned payload card represents a “quantum leap” for sensor data processing capability in size, weight, and power (SWaP) constrained applications.
The CHAMP-XD3 combines a 10-core Intel Xeon D-1700 processor for DSP processing with a Xilinx MPSoC FPGA, which supports Curtiss-Wright’s Enhanced TCOTS framework, aligned with the compute-intensive payload profile as defined by the SOSA Technical Standard. This rugged, conduction-cooled module is designed to handle the largest processing tasks characteristic of multi-mode / synthetic aperture radars (SAR), modern signal intelligence (SIGINT), electro-optical/infrared (EO/IR), and EW applications.
To ensure wire-speed transfers of sensor data, the card provides three channels of DDR4-2400 SDRAM memory (up to 48 Gbytes), delivering more than 50% additional memory bandwidth compared to previous generation solutions. In addition, the CHAMP-XD3 features up to 160Gbytes of memory via dual NVMe SSD devices to support demanding virtualization, high-bandwidth, and redundant non-volatile storage requirements. The dual SSD’s deliver twice the memory bandwidth of prior solutions and provide support for Intel’s Virtual RAID on CPU (VROC), which improves RAID performance by shifting control from the host bus adapter to the CPU itself. To eliminate I/O bottlenecks, the module’s backplane supports a wide-band,16-lane Gen3 PCIe expansion plane, in addition to its 40 GbE data plane and dual 10-GbE Control Plane interfaces.
This powerful DSP processor module is engineered for the most demanding compute-intensive sensor processing applications. To meet the needs of emerging cognitive electronic warfare (EW) requirements and accelerate artificial intelligence (AI) / machine learning (ML) functions, the CHAMP-XD3 features Intel’s AVX-512 instructions to support extensive floating-point calculations. With AVX-512, Intel doubles the width of the floating-point engine, significantly speeding sensor, ML, and deep learning inference processing, as well as machine vision and smart video analytics applications. AVX-512 also enables Intel Deep Learning Boost, which is based on optimized vector neural network instructions (VNNI).
The CHAMP-XD3’s Xilinx MPSoC FPGA features a quad-core Arm A53 and a dual-core R5 processor and can be used for enhanced TrustedCOTS (eTCOTS) security functionality. An optional MPSoC FPGA toolkit is available for applications that require greater design versatility for adding security IP. This toolkit enables the integration of advanced security IP, such as Raytheon’s Night Cover product suite and Idaho Scientific’s Immunity cryptography cores. The MPSoC FPGA can also support co-processing and general-purpose I/O requirements. The module’s Microsemi SmartFusion2 FPGA has a dedicated intelligent platform management controller (IPMC) core, with HOST 3.0 extensions, that support system, and health monitoring, as well providing enhanced security capability that adds additional protection of critical control logic.
Software support for the CHAMP-XD3 DSP engine includes a comprehensive BIOS and Linux board support package (BSP) for ease of integration and solution development.
The CHAMP-XD3 is available in both SOSA-aligned and enhanced IO configurations.
The CHAMP-XD3 integrates seamlessly with Curtiss-Wright’s broad range of CMOSS/SOSA- aligned system building blocks, including VPX3-687 VICTORY data/control plane Ethernet switch; VPX3-663 PCIe Gen 3 / 10G Ethernet hybrid switch; VPX3-673A A-PNT/Radial Clock card; VPX3-1260 9th Gen Intel Xeon-E processor-powered general-purpose single-board computer; VPX3-E320 ruggedized universal software-defined radio peripheral; VPX3-4935 NVIDIA Quadro Turing-power GPU-coprocessing engine; 8-slot 3U OpenVPX CMOSS/SOSA-aligned enclosure; CMOSS/SOSA Starter Kit for ground combat vehicles with VPX3-1260, VPX3-673A, and VPX3-687 modules;