Artificial intelligence (AI) is beginning to impact defense-industry surveillance and analysis systems, prompting a transformation in board and module designs. Suppliers of these systems are modifying design parameters as military planners move toward standardization.
A range of military suppliers that develop products ranging from sensors and processing boards down to connectors examined changing design requirements at the recent Embedded Tech Trends conference, sponsored by the VITA standards organization. Many speakers detailed the rising interest in AI, which is being explored as a way to analyze the huge volumes of data gathered by remote sensors. Image processing and AI are typically processed by highly-parallel graphical processing units (GPUs), which are playing a larger role in boards and modules.
“AI is ideal for drones and other autonomous systems,” said Emil Kheyfets, military product line director at AiTech Defense Systems. “Image processing tasks generally require heavy-duty calculation power. These kinds of calculations are too much for a conventional CPU [central processing unit], so general-purpose GPUs like Nvidia’s processors are being used. Complete AI systems still require a regular processor to make a decision or take an action based on the analyzed information.”
Though machine learning is viewed as a solution for many complex jobs like analyzing hours of video imagery, implementing it is no simple task. Systems must be trained to recognize individual objects, like a car or truck, so it can be time consuming to collect images from agencies. Issues like this make it difficult for AI and system suppliers to develop workable business plans and build systems that implement them.
“We’re seeing a lot of AI interest, but it’s still a challenge to see where AI fits,” said Devon Yablonski, principal product manager, Sensor Processing, at Mercury Systems. “In defense, there are intriguing opportunities. But there’s a lack of training data, much of which is secret, and there’s a lack of understanding of business models. Many companies are struggling or have already failed.”
The impact of AI ripples throughout the supply chain. The large volumes of data from sensors is driving changes in networks used in military vehicles. In many systems, networking loads are rising as massive amounts of data are transferred to the neural networks that run AI programs.
Dense networks, often called fabrics, have extensive wiring that links systems and memory. That’s prompting changes that change the relationship between connectors and the FR4 material used in printed circuit boards. Wires can carry faster signals than FR4, so companies are beginning to use so-called flyover schemes, in which wire/connector combinations carry high-speed signals across circuit boards, for example between memories and GPUs.
“AI and machine learning are fueling growth for connectors,” said Burrell Best, signal integrity architect at Samtec Inc. “Faster systems require advanced fabrics and attached storage. As we get down the road with faster fabrics, signal integrity really matters. As FR4 becomes prohibitive, cabling is trending towards ultra-thin gauge cables and low loss dielectrics, which provide superior signal integrity performance over printed circuit board materials.”
Though many design teams are dealing with increased data transfer requirements, that’s not always the case. Military architects sometimes use distributed architectures, putting intelligence in sensors. Incoming data can then be analyzed so that only potentially useful data is sent over networks.
“AI can reduce data transfer and data storage requirements by storing only data that requires attention or action,” AiTech’s Kheyfets said. “That can save hundreds of Gbytes per camera over the course of a mission.”
As systems get more complex, military leaders are making a push to use more standards so they can build systems using commercial off-the-shelf (COTS) systems. Early this year, a tri-service memo signed by the Secretaries of the Air Force, Navy and Army endorsed a number of standards such as Modular Open Systems Approach (MOSA). It directed agencies to include MOSA-supporting standards in all requirements, programming and development activities. That’s an exciting step for board and module suppliers, who feel the market will grow as standardization frees up revenue.
“There are acquisition dollars behind these standards,” said Mike Hackert of NAVAIR. “The Army’s Joint Battle Command Platform program is very big, it has 125,000 platforms. All three services are focused on the same endgame, lower overall costs.”
Hackert noted that the adoption of standards impacts spending and costs throughout the long lifetime of military hardware. Standards make it easier to carry spare parts while also letting engineers migrate to newer technology over time.
“90% of our cost comes after we acquire technology. If we have a chassis for standard modules, we can change technologies out much faster than if there are no standards. We’ll be able to save money using standard modules,” Hackert said.Continue reading »