White Paper

Tackling AI and Multi-Domain Operations in Real Time with PCIe Gen 4.0

Issue link: https://read.uberflip.com/i/1359091

Contents of this Issue

Navigation

Page 1 of 6

WHITE PAPER Tackling workloads in real time with PCIe Gen 4.0 mrcy.com 2 mrcy.com 2 Until just a few years ago, defense industr y technology designers could count on continual leaps in raw semiconductor processing power as they created new generations of systems that ser ved as the processing backbone for critical operations. Then, as individual chip compute power slowed, designers moved to advanced processing models, with multiple specialized components working together to process data streams. This design approach is now essential for high-end per formance in edge ser vers, a class of processing systems crucial for agile electronic defense response. As today 's deployed server designs continue to demand faster processing of more data, they are encountering a new bottleneck—data-path bandwidth. PCIe (peripheral component interconnect express) Gen 3.0, a component interconnect mainstay since 2010, can no longer keep up with system requirements. Throughput has stagnated. Fortunately, a straightforward upgrade path is now viable. PCIe Gen 4.0, with double the bandwidth, is supported by new, rugged, and deployable server-class technology. TODAY'S NEED FOR MORE INTER-COMPONENT BANDWIDTH The complexity of modern defense systems makes interconnect bandwidth increasingly critical to overall system performance. Field-deployed processing subsystems receive input from multiple types of sensor systems, including radar, Lidar, visual imagery, EO/IR and EW. This data is received, transferred and analyzed for comparison and analysis purposes such as identifying artifacts in a current image that were not present in prior images or presenting information to the proper decision-maker for quick response. Each sensor interface is a data path that must be able to keep pace with the data source. To process data streams, a subsystem will employ multiple components of various types – CPUs, GPUs, FPGAs and DSPs. Moving data quickly from one component to the next is critical to overall computing performance. At a higher level, deployed servers must communicate with each other and with displays that form the human interface. For example, an attack submarine's complex electronic warfare system acquires multiple types of sensor data, combines and analyzes that data in real time, and dynamically generates a threat warning to aid in rapid response. A mobile, tactical command center does the same thing, receiving sensor input from both vehicles and individual combatants, then creating a holistic view of the battlefield from that data. Even single-purpose systems, such as air defense radars, are employing huge arrays of specialized processors, all with interconnects, as they prepare to counter the next generation of weapons technology. AI AND 5G DEMAND EVEN MORE BANDWIDTH Looking beyond current requirements, computing systems deployed in the field need to support AI functions, as defense systems use deep learning and inference engine tools in the field to extract critical information from a torrent of incoming sensor data. Applying AI effectively to real-time applications will demand high bandwidth and flexible data movement without bottlenecks. For example, backwards-chaining inference engines work from a conclusion/an assumption and look for supportive data to draw a deduction. Low- latency access to a range of data types is essential to the operation of real-time systems using this technique. 5G communications technology has also come in to play. A host of units on the battlefield are starting to transmit continual streams of sensor data over 5G networks. 5G has the bandwidth to keep up with a new generation of sensors, but edge computing must also keep pace with 5G to process, analyze and archive those fast data streams to deliver agile communications and time-sensitive responses. Data collection on the electronic battlefield has exploded to levels unimaginable just a decade ago, in terms of both complexity and sheer volume. In parallel, adversaries have added increasingly sophisticated capabilities, steadily decreasing the timeframes available for effective response. Future multi-domain operations aim to rapidly synchronize and integrate services and technologies across the six warfighting domains to deter and defend on any battlefield. An Attack Sub is a Complex System for Acquiring Sensor Data

Articles in this issue

Links on this page

view archives of White Paper - Tackling AI and Multi-Domain Operations in Real Time with PCIe Gen 4.0