Satellite operators are looking at new brain-like processing power that promises to deliver greater intelligence capabilities and significantly reduce energy consumption, according to a leader at the U.S. Navy Research Laboratory's Naval Center for Space Technology.
The center is tasked with improving the Navy's autonomous systems in space and is investigating applications in this technology — what's called neuromorphic processors. The technology would offer major opportunities for AI and allow the Defense Department to advance pattern recognition, autonomous tasking, and reduction in energy and latency, all of which are critical for robotic applications where there is a need for a robot to immediately respond to constant, real-world change.
"We [the satellite community] are interested in using AI to control robots and other autonomous systems," Steven Meier, the center's director, told GovCIO Media and Research on GovCast. "This is a challenge, though, because AI and [machine learning] types of systems tend to be very brittle. They break, often catastrophically, when encountering situations it wasn't designed for."
Meier explained the necessity to introduce more modernization to these technologies as they become more widely used.
"As AI and ML is used today, we have to download all the data, analyze it on the ground using substantial computing resources. And it isn't really feasible to fly computers that big on most spacecraft. AI/ML systems take up too much space, require too much power, and generate far too much heat. So someday soon, we hope to fly computers and AI/ML algorithms that can recognize objects or phenomena of interest on board and just download the pertinent data directly to scientists and warfighters," Meier added.
The lab has been leading an effort to develop neuromorphic hardware elements based on a new building element for electrical circuits called "memristors," also known as memory resistors. If successful, memristors would revolutionize electronics and end the era of transistors made of silicon.
"In terms of what we're pushing forward at the state of the art at NCST to help facilitate this for satellite systems, or others, are some developing novel types of computers that … work more like a brain than a regular computer, like your laptop," Meier said. "We have ... several of these processors that are very unique, and they, we feel like, are the future of AI/ML type systems."
The Air Force Research Laboratory is also working on Neuromorphic Intelligent Computing Systems, a development program using neuromorphic technologies to advance capabilities like artificial intelligence and machine learning for edge computing.
While significant strides have been made around neuromorphic computing in the past several decades, there are a number of challenges that the satellite community would need to overcome before adopting and mass-producing the technology, including the general availability of neuromorphic hardware and developing algorithms that are capable of mimicking the human brain.