White Paper


Issue link: https://read.uberflip.com/i/1370335

Contents of this Issue


Page 4 of 5

WHITE PAPER Bringing commercial technologies to defense in the age of AI mrcy.com 5 For instance, multiple interconnects improve OpenVPX system resilience to shock and vibration but also create poor channel integrity, bottlenecking processor performance. Specialized backplanes and modules are utilized to alleviate signal-channel limitations and deliver steady operation across broad temperature ranges. COOLING COMPONENTS As today 's systems accelerate AI through high-power components, thermal management becomes a multi- constraint problem that often requires trade-offs in performance, size or cost. Air, liquid, hybrid and other cooling methods enable smaller form factor Intel Xeon- powered packages to operate reliably across wide temperature ranges. Computing guidelines must be weighed against cooling pros and cons to ensure design solutions are suitable for target applications. For example, forced airflow is a cost-effective cooling method that utilizes fans and mechanical design techniques to facilitate airflow through the chassis. However, fans can increase the risk of mechanical failure and generate noise, which may not be suitable for airborne and underwater platforms. In contrast, liquid-cooled solutions employ fuel and dedicated refrigerants but are typically more expensive than their air-cooled counterparts. Mercury systems evaluates a broad range of cooling techniques against platform and budget requirements to deliver mission-critical solutions. MITIGATING RISK WITH SECURE PROCESSING Increased globalization has driven integrated circuit supply chains across borders, creating threats such as counterfeiting, reverse engineering and piracy, which can be unknowingly propagated and exploited to undermine AI systems. Domestic design and manufacturing can tightly control chip-level silicon assembly to box-level subsystem packaging, mitigating risk. Manufacturing and testing components in secure facilities minimizes the danger of back doors, counterfeits and Trojans. 5 ENABLING AI ANYWHERE WITH EDGE-READY HARDWARE The AI era will require innovative edge-ready hardware that accelerates workloads and scales to match AI advancements. To scale Intel-powered solutions from the data center across fog and edge layers and deploy them in the field, servers and computers must become smaller, operate reliably in harsh environments and pass rigorous testing and certifications. HIGH-DENSITY COMPUTING IN A MINIMUM FOOTPRINT Unlike static data centers, mobile edge surface, subsurface, airborne and ground platforms have finite space and power resources. Advanced mechanical and electrical engineering techniques can overcome these constraints by densely packing multiple silicon components and flexible power options in more compact footprints. While a typical data center rackmount server is 29" deep, Mercury 's rackmount servers are as short as 13". Miniaturization approaches such as system-in-package, wafer-stacking, 2.5D and 3D fabrication further shrink components and the form factors that house them by varying degrees. When size and weight really matter, Mercury 's rugged OpenVPX server blades have a footprint reduced by over 90% when compared with commercial blades with similar processing capabilities. Component selection is critical to meeting platform power requirements. High-performance Intel Xeon Scalable processors, for example, typically require more power than other types of processors. These high-end data center CPUs may not be appropriate for AI systems running at the edge — on vehicle batteries, for example. Specialized AC/DC power supplies, redundant power and battery power cases enable systems to run in-theater and on any power source. RUGGEDIZATION FOR EDGE OPERATIONS Rugged computing systems can process information near the data source to drive time-sensitive AI decision- making because they are built to operate in environmental extremes that present shock, vibration, dust, fog, pressure, electromagnetic field, altitude and temperature challenges. Such extreme environments have the potential to impact hardware reliability and performance, requiring further design enhancements to guarantee full-throttle performance. Mercur y utilizes a variety of miniaturization and packaging techniques to deliver SWaP optimized products that range from chip-scale to system scale. SUB-SYSTEMS / SYSTEMS MODULES / LRUS COMPONENTS

Articles in this issue

Links on this page

view archives of White Paper - Bringing_commercial_technology_defense_age_ai_WP