Connecting the Internet of Things

Crossfield® has experience in the specification and integration of high performance computing (HPC) and high performance embedded computing (HPEC) systems for real-time signal processing and simulation applications. Using both commercial off-the-shelf components and custom designed hardware, we deliver state-of-the-art performance for a broad spectrum of applications, such as hardware-in-the-loop simulation and signals intelligence.

Embedded Systems

Crossfield develops embedded systems for the most technically challenging applications in the world, be it ground, air, sea, or undersea. We integrate next-generation processing, networking, and storage technologies into VITA standard form factors to provide modular systems to our customers. Our VITA 46/48 VPX modules are designed to fit VITA 65 OpenVPX profiles to create systems that are both scalable and upgradable. VITA 66 standardizes pluggable optical connectors for the VPX backplane, enabling ultra-fast optical links and overcoming the performance limitations of electrical backplanes.

Crossfield is developing HPEC system components that use serial switch fabrics and hybrid electical-optical interconnects between heterogenous processors, non-volatile storage, and instrumentation modules.

An optical switch fabric enables HPEC modules to communicate with each other at higher data rates (25 Gbps per lane today, with 50 Gbps per lane in development). It also enables payload modules from multiple chassis to easily communicate with remote sensors and transducers over single-mode and multi-mode fiber optics.
Stacks Image 1436
Stacks Image 1388

Heterogeneous Processing

Graphics Processing Units (GPUs), Field Programmable Gate Arrays (FPGAs) and other types of coprocessors currently provide 1-10 TFLOPS/TIOPS of raw signal processing performance per device. With this tremendous performance boost, physics-based algorithms can be used to provide high fidelity simulations and analysis. The different types of processors are tailored to different types of algorithms with varying computational intensity.

NVIDIA® CUDATM and OpenCL programming languages are used to implement parallel signal processing algorithms on graphics processors and general purpose processor, while Verilog and VHDL are commonly used to implement DSP logic in FPGAs, with more advanced FPGAs now providing support for the OpenCL programming language.

Crossfield is using these frameworks to implement real-time simulation and signal processing algorithms on heterogenous processors. In addition, Crossfield is developing techniques to integrate our instrumentation gateways with HPCs in a tightly-coupled real-time environment.

Networking Technologies

HPC systems are configured with 10/40/100 Gigabit Ethernet and QDR/FDR/EDR InfiniBand switch fabrics. Open Fabrics Enterprise Distribution (OFED) software provides a common software protocol stack to access these networks.

Crossfield uses PeerDirectTM Remote Direct Memory Access (RDMA) data transfer protocols to maximize throughput and minimize latency between HPC processors and our Instrumentation Gateways.

* PeerDirect is a trademark of Mellanox Technologies
Stacks Image 1722
Stacks Image 2201

Storage & Memory Technologies

Non-Volatile Memory Express (NVMe) is the new high-performance interface for Solid State Disks (SSDs) that defines optimized storage protocols for PCI Express and network fabrics such as iWarp, InfiniBand and FibreChannel.

Hybrid Memory Cube (HMC) is competing with DDR4 SDRAM and other new memory technologies for use in next generation servers and signal processors. HMC defines a high-speed serial link and protocol for memory transactions.

These storage and memory technologies provide an order-of-magnitude improvement in power-bandwidth over previous generation components. Crossfield is applying these technologies in our current and next generation embedded systems and instrumentation gateway products.

Advanced Technologies

Crossfield holds advanced technology patents in the areas of Multipath RDMA data communications and fused floating-point units. Multipath RDMA provides an order-of-magnitude increase in network bandwidth through a switch fabric. Fused floating-point units save area and power consumption compared to a discrete floating-point operations.
Stacks Image 1828