Embedded AI for military applications

Artificial intelligence is redefining decision-making and responsiveness, creating an environment for a potential shift in global power to those who wield it.

Artificial intelligence (AI)-powered smart machines and other cognitive offensive capabilities are being fielded by nations and non-nations who are seeking to raise their respective global postures and/or degrade others as a new age of warfare begins. The new age of warfare is AI-powered and uses technology that is affordable and widely available.

Realizing that other actors are doggedly seeking AI dominance, military leaders are responding to the danger this situation presents to national defense and global political positioning. As Mary Miller, the Acting Assistant Secretary of Defense for Research and Engineering, said to the House Armed Services Committee in March 2018, “The development of a new strategy for AI is a major priority. AI, machine learning, and ‘human-machine teaming’ are major cornerstones of the ‘third-offset’ strategy that responds to concerns that the U.S. military is at risk of losing its technological edge to potential adversaries, including Russia and China.”

Converging capabilities created AI

For decades, multiple dominant digital technology trends have been converging and now they have come to an inflection point: The continuation of Moore’s Law, big data and cloud processing, the Internet of Things (IoT), and the technology achievements stemming from the development of smart, autonomous vehicles have collectively converged and produced the environment for AI and machine-enabled computing to come of age.

Moore’s law: Historically, scientific and financial algorithms and big-data processing problems in general have always needed more processing capability than was available. The algorithms that seem most useful always need tomorrow’s compute power to provide the processing, required within a reasonable amount of time. However, today’s AI algorithms have intersected with the current processing capability predicted by Moore’s Law and are achieving meaningful results in a useful time.

The cloud-computing megatrend: Cloud companies including Dropbox, Microsoft Azure, Amazon AWS, and others offer places to store data safely off-site that is accessible from anywhere in the world. These data center-powered clouds offer virtual big-data processing capabilities to deliver high-performance analytics with ease. The ability for so many to access these kinds of computing and storage capabilities was unimaginable a few years ago, but today it is available to all with a low-cost subscription. This reality has encouraged the obsession of keeping any data that could be useful, even if in most cases no one knows exactly what to do with it yet.

Internet of Things (IoT): Mobile devices capture temperature, location, audio (voice), and visual image data, to name a few, at the edge. In near-real-time the collected data is transmitted to the cloud or other central repository where big processing resources can analyze the individual data and as a group, offer learned opinions and calculated predictions through AI.

Autonomous vehicles: With big financial bets being placed by Google, Apple, Amazon, Uber, Tesla, and the automobile industry as a whole, the momentum of AI innovation within the autonomous-vehicle domain has become unstoppable. This is producing on-platform processing capabilities that support secure and intrinsically safe AI processing and effector (e.g., avionics, vetronics, countermeasures, etc.) initiation required for autonomous vehicle deployment. AI-enabled vehicles from many enterprises and nations are fundamentally a technology game changer. These capabilities will not remain within the commercial sector, as actors from around the globe will leverage them as force multipliers for both good and bad missions.

Embedding HPC for AI

The hardware challenge for deploying AI capability at the tactical edge, as is often required for defense applications, pivots around making data center technology sufficiently small, light, and power-­efficient. Additionally, these processing resources should be rugged (for survivability) and run cool (for reliability). For system integrity, trust and security must be built in; for critical mission effector actuation determinism and inherent flight safety-worthiness, both must be demonstrated and provable.

Military-grade AI hardware made rugged, smaller, and cooler

As AI applications expand, they will no longer be restricted to the data center. As big processing capability shrinks, uses less power, and is made rugged, AI applications will increasingly power smarter, more capable military platforms and missions.

Many of these physical enablers already exist. Modern military-grade electronic packaging technology is able to shrink compute architectures through system-in-package and wafer-stacking techniques to reduce system volume. High-performance cooling systems efficiently remove the heat generated from these miniaturized, thermally dense, embedded systems enabling reliable, full-throttle deterministic processing. Rugged fabrication techniques enable even the most powerful data center CPUs [central processing units], FPGAs [field-programmable gate arrays], and GPGPU [general-purpose graphics processing unit] processors and accelerators to be ruggedly mounted to their substrate regardless of their native interconnects.

Modern military-grade electronic packaging technology has matured to the point that data center-like capabilities can be shrunk to smaller form factors, including OpenVPX (ANSI/VITA standards), the most widely adopted and supported rugged high-­performance open system compute architecture for defense programs. (Figure 1.)


Figure 1 | Pictured is an example of secure embedded AI processing system building blocks. Embedded rugged AI processing blades, compliant to the 3U OpenVPX open system architecture (ANSI/VITA 65) are available as air-, conduction-, or liquid-cooled blades for deployment on the ground, in the air, or under the sea.

AI made secure and trusted

AI algorithms create a neural network, which may be thought of as a computer data representation of the human brain. This neural network is trained using data to make it “intelligent.” In the case of defense AI systems, the data used for training is often sensitive or classified. These situations call for secure systems. (Figure 2.)


Figure 2 | Secure embedded AI processing system building blocks: Embedded rugged AI processing blades in various deployable packages are compliant to the 6U OpenVPX open system architecture (ANSI/VITA 65). At about a tenth of the size of a data center server, these blades are powered with the same processors and may be air- or liquid-cooled.

Ultimately, the neural network could contain a brain equivalent to someone who has been trained on all military intelligence, assets, strategies, personnel, and anything else known collectively by military and intelligence agencies. Although we are not at this stage, we are moving in that direction. There are already AI-trained brains that contain vast amounts of information or AI-derived information that is highly restricted. It is imperative to protect this information, and the hardware and technology that support it, from all competitors. This is the essence of embedded security and built-in trust, which should be holistic across hardware, firmware, software, facilities, personnel access, and processors used. Security and trust are critical in the modern competitive environment.

AI made ready for flight-safety certification

Vehicle autonomy is a critical use of AI because the algorithms assist in detecting objects to avoid (or follow) and perform actions without human intervention. Commercial and defense mobile platforms have mission computers to control their effectors that are required to prove high levels of reliable, deterministic, and safe operation. Safe operation means that both the hardware and software have been shown and documented to be highly deterministic and reliable. Such systems can be depended upon to act appropriately based upon inputs and application parameters. Varying degrees of reliability and critical function execution are validated through Design Assurance Level (DAL) certification, including DO-254 for hardware and DO-178 for software. These certifications are difficult if not impossible to add on after a module/system has been designed. High levels of demonstrable safety assurance are required before autonomous vehicles are permitted to take to the skies and streets.

What’s coming up for AI?

AI is moving out of the confinements of the data center and becoming embedded in all types of platforms, including those used by the military. Simultaneously, adversaries are investing heavily in AI technology for technical advantage through espionage and offensive operations and missions. These adversaries view AI as more than a force multiplier and tactical differentiator; they see it as a means to tip the balance of power in their favor.

To mitigate these challenges, modern militaries have to develop and deploy better AI-powered capabilities for all manner of defensive and offensive missions. The most effective and practical way to do this is to leverage existing artificial intelligence IP in the commercial domain and make it ready for military applications.

Modern defense prime contractors are developing the capabilities and technologies required to embed AI processing capabilities all the way to the tactical edge. Through system miniaturization and new military-grade packaging, the best commercial data center technologies can be made compatible with defense applications. These companies are building in effector determinism, as defined by flight-safety certification and embedded holistic systemwide security, that enable AI-powered systems to be deployed anywhere. Simply stated, making the best commercial data center AI processing capability rugged, secure, and safe for deployment at the tactical edge will enable the next generation of smarter military missions, offsetting an increasingly sophisticated and similarly equipped set of competitors.

John Bratton is director of product marketing for Mercury Systems’ Sensor and Mission Processing group, which is based in Andover, Massachusetts. John has more than 25 years of experience managing and promoting solutions within the embedded packaging, interconnect, and RF domains. John earned his bachelor’s degree in mechanical engineering from Manchester Metropolitan University, England and is a member of the IMechE and ASME.