Neuromorphic Computing: Making Space Smart

Published Mar 19, 2024

Eric Gallo, Sr. Principal, Future Technologies R&D, Accenture Labs

Space technology is advancing, and companies that never considered themselves space companies are now relying on satellite data and other space services to operate and grow their business. In fact, the immense global interest in satellite data and images has seen a growing number of new companies emerge, several of which we’ve invested in or aligned with through our Accenture Ventures “Project Spotlight” initiative, such as Open Cosmos, Pixxel, Planet, and SpiderOak.

To optimize these operations, reduce costs, and improve effectiveness, there is a growing demand to make space equipment, such as satellites, ‘smart’ by leveraging AI technologies. However, implementing AI in space faces challenges such as limits on the size, weight, and power (SWaP) of computing hardware. There has been a major research and engineering focus on developing low-power AI hardware at the edge to provide onboard intelligence under SWaP constrained conditions. While custom hardened electronics are still preferred for deep space missions, many are adopting commercial off-the-shelf components for shorter missions, offering space technologies the ability to implement these Edge AI capabilities.

Small satellites face bandwidth limitations, but technologies like edge processing and cognitive radio can optimize these limited communication resources. The amount of data collected in space is increasing rapidly. According to NSR, space data traffic is expected to reach 566 exabytes over the next decade, with satellite communications making up 530 EB of that amount. The increasing congestion in space and the substantial data flow to ground stations necessitate improvements in efficiency. Edge AI can segment data in-situ into usable/unusable data to filter the information sent back to Earth, reducing bandwidth and data center processing.

The rise of neuromorphic computing, processors and sensors modeled after the neurons in our brains, are uniquely suited to provide edge processing in space. Neuromorphic technologies are low latency, extremely low power and even capable of learning while in use. Neuromorphic computing can be added to a satellite or other equipment with minimal impacts on system power requirements or mechanical design, making them an ideal low SWaP, adaptable solution for space applications, with the potential to revolutionize satellites and beyond.

Edge computing in space

Until recently, edge processing in space was uncommon. In 2017, the HPE Spaceborne Computer-2 was successfully deployed on the ISS to address data processing bottlenecks. However, its size and power requirements make it unsuitable for smaller satellites or instruments. In 2020, the Phi-Sat became the first satellite to demonstrate edge AI in Earth observation, reducing transmitted data by 30% by processing images prior to transmission to Earth. The satellite used an Intel Movidius Myriad processor designed for low-power edge intelligence, as used in drones, smart surveillance cameras and other energy-constrained applications on Earth. Now there are several vendors offering edge processing electronics for space applications, but the main challenge has been limited physical space, power, and cooling resources. Neuromorphic technologies offer a cost-effective solution by adding intelligence at extremely low power and financial costs, increasing satellite capabilities without adding weight or size.

Why is neuromorphic computing the solution to making space technology smarter?

Neuromorphic computing emulates neurons and synapses in the brain to process information, offering low power, low latency, and data sparsity benefits for edge computation. Accenture Labs demonstrated a 3-5x reduction in system power using BrainChip’s Akida neuromorphic accelerator for audio recognition tasks compared to CPU and GPU. Other companies—including BrainChip, Intel, and Synsense—have developed digital neuromorphic silicon chips, with BrainChip, and Synsense offering commercial processors. These chips are still in v1 and v2, with increasing performance expected as they mature. Startups like EDGX and Neurobus are exploring neuromorphic computing hardware and software designs for space. Intel’s Loihi processor has already been validated in space, and neuromorphic cameras are operating on the ISS as part of the Falco Neuro Project. BrainChip’s Akida recently launched on SpaceX Falcon 9. These efforts are laying the groundwork for widespread implementation of neuromorphic technologies in space applications.

A few potential neuromorphic applications have already been identified and are being investigated by universities, space agencies, start-ups and Accenture Labs in collaboration with Accenture Space Innovation:

1. Optimization of satellite communications using cognitive radios that adapt to changing conditions, ensuring efficient, robust and reliable data transmission.

2. Enabling real-time decision making on satellites, facilitating rapid signaling in emergencies and natural disasters, supporting latency-sensitive applications, and enhancing satellite health analysis and space situational awareness.

3. Filtering and sorting collected images before transmission to earth and enabling real-time focus on areas affected by flooding, deforestation, algal blooms, or other environmental disasters.

4. Enabling longer range missions by reducing reliance on earthbound communications, enhancing data quality, and mitigating potential dependencies.

5. Implementing energy-efficient robotic controls with rapid learning and adaptation to the environment.

6. Integrating real time health monitoring for satellites and for astronauts.

7. Increasing efficiency and effectiveness of experiments in space by directing resources, identifying optimal conditions, and controlling input, thereby optimizing results as companies explore the effects of zero gravity on materials and devices.

8. Learning and adapting to a satellite’s behavior which can then be used for anomaly detection as the hardware ages, allowing for early identification of potential issues or malfunctions.

9. Mitigating damage to hardware during severe space weather, detecting and shutting down components to prevent radiation damage.

10. Improving space situational awareness using neuromorphic cameras and processors to detect objects and identify threats rapidly and efficiently.

11. Enhancing specific satcom applications such as imaging radar (synthetic aperture radar), interference detection, congestion forecasting, jamming classification.

Neuromorphic technologies offer additional functional benefits, such as the ability to load and unload networks onto a processor for multiple tasks on the same hardware, enabling a satellite to switch between scanning for forest fires over land and investigating currents over the ocean. Open-source designs allow developers to create applications specific to their use case and deploy them on neuromorphic hardware, facilitating the use of a single CubeSat by multiple groups.

The technology also excels at leveraging multi-modal input for decision-making and can adapt and change its data processing approach in real-time. Accenture is actively exploring the applications of neuromorphic computing in space and collaborating with academic researchers and commercial neuromorphic companies to evaluate and demonstrate its potential for enhancing edge intelligence and creating new products and applications. In collaboration with BrainChip, we are evaluating their Akida platform, the first commercially available neuromorphic chips, to identify industries and applications that can benefit from this technology.

BrainChip’s Akida has demonstrated its power in various applications, including automotive safety, industrial maintenance, vision solutions, and healthcare devices. It is an ideal choice for investigating and designing edge processing in space, providing sufficient computing power for image streams while consuming less than 1 W of energy. At Accenture, we are using the Akida and other platforms to validate these benefits and create proofs of concept for space tech innovations.

Neuromorphic computing offers crucial benefits for the space industry, including low power consumption, minimal latency processing, and efficient edge intelligence. It has the potential to revolutionize space technology, leading to advancements in exploration, communication, and situational awareness. As our activities in space expand, the use of neuromorphic computing enables faster and more efficient operations, making it a crucial component in the journey of space exploration.