FeatherBox is a compact Data Processing Unit for artificial intelligence applications on-orbit. It's small size (0.5U) makes it compatible with most satellite platforms. Rather than sending large sets of raw data down to ground stations, FeatherBox can process sensor data from multiple sources and only send down the relevant information. Processing data on orbit reduces downlink costs and improves response times to observable events.
Integrated into FeatherBox is a machine learning system capable of performing 4 trillion operations per second (4 TOPS). If more processing is required, the system can scale by pipelining models across multiple processors. Precision thermal management is handled within the enclosure, meaning there is little to no need for additional external thermal control systems.
Combined with FeatherWare, it provides a complete artificial intelligence payload for on-orbit image processing, enabling autonomous decision making in space.

DATASHEET
Dimensions: 96 mm x 96 mm x 50 mm
Mass: 1.4 kg
Power: 9 W (typ), 22 W peak (tens of microseconds)
Unlock the potential of AI in Space
Combining Cutting-Edge Technology with Reliability
FeatherBox comprises a commercial off the shelf computing module integrated with Exo-Space's rad-tolerant motherboard. The device is a fully-integrated Linux system, designed to accelerate TensorFlow Lite models. FeatherBox leverages the fastest, most power efficient computer products on the market while also maintaining the reliability needed for space applications.
Protection against the harsh environment of space
Multi-layered structural shielding to minimize total ionizing dose and single event effects
Various software redundancies and mitigations including error detection and correction code for volatile and non-volatile memory
Voltage and current monitoring with automatic power cycle capabilities
APPLICATIONS

Autonomous Rendezvous Proximity Operations and Docking

Satellite Inspection and Servicing

Earth Observation Image Processing

Data Compression and Encryption
Technical Specifications
Interfaces
Connector
15 pin D-Sub Micro-D connector
UART
Data rates up to 5 Mbps
Ethernet
Data rates up to 1000 Mbps
Non-Standard
Additional interfaces/protocols upon request: SpaceWire, QPSK, etc
Properties
Mass
1.4 kg
Size
96 mm x 96 mm x 50 mm
Power Supply
5V
Power Consumption
9 W (typ), 22 W peak (tens of microseconds)
Performance
CPU
Quad Cortex-A53
GPU
Integrated
ML Accelerator
Coprocessor: 4 TOPS
RAM
4GB LPDDR4 with ECC
Flash Memory
32GB SLC NAND Flash (EDAC)
AI/ML Model Type
TensorFlow Lite (8MB size)
CORE ADVANTAGES
Compact and Powerful
Powerful machine vision processor and AI accelerator inside a compact 0.5U enclosure.
Custom structural shielding, and error detection and correction codes (EDAC) to protect against the effects of space radiation.
Mission Safety
Single communication and power line connector for easy plug-and-play integration.
Plug and Play