Leopard

Data processing unit for in-orbit Artifical Intelligence

Learn more

Leopard DPU

Leopard is a CubeSat standard-compliant Data Processing Unit which enables mission designers to

  • Apply Artificial Intelligence solutions in space.

  • Support the capturing, managing and processing of data in orbit.

Leopard redefines the current approach to remote sensing. Now, instead of sending huge, unprocessed sets of data to ground stations, Leopard uses Deep Neural Networks to process data on-board and therefore only sends the most important and valuable insights to the ground. By reducing the time and cost of data transfer and processing, it enables you to focus on a rapid response to any detected phenomena.  

Leopard is integrated with the powerful FPGA to accelerate the execution of deep learning algorithms and has a throughput of up to 3 Tera Operations Per SecondA number of hardware and software measures protect the computer against the influence of radiation. 

With its small size (1U form factor), wide voltage range and universal interfaces, it is compatible with most CubeSat platforms. Its scalable and customisable architecture makes it possible to create larger and more powerful versions dedicated to bigger platforms as well. 

Key advantages

Huge processing power in a small form-factor

Powerful FPGA to accelerate the execution of deep learning algorithms

Artificial Intelligence available on board

Processing data directly in orbit using Deep Neural Networks and ML libraries such as Cafe or TensorFlow

Freedom, flexibility and security

Open-source operating system (Linux), support of multiple Neural Network models, single-cell memory and rad-tolerant design

TECHNICAL SPECIFICATION

PROCESSING CORES Zynq UltraScale+ MPSoC ZU2EG/ZU4EG/ZU9EG/ZU15EG
4x XARM A53 CPU 800 – 1000 MHz
ARM R5 lock-step CPU
FPGA architecture for custom function implementation
MEMORY 4-16 GiB DDR4 providing Error Detection and Correction (EDAC)
4-16 GiB SLC flash-based filesystem storage (EDAC)
INTERFACESControl Interfaces: CAN, GPIO
Data Interfaces: High-Speed LVDS, SPI, RS422/485
Additional custom interfaces upon request
SPECIFICATIONSA radiation hardened Payload Controller
Voltage +6.5 to +14 V
Power Consumption 7.5 W to 40 W depending on workload and specified processing speed
Computational Throughput for Neural Networks Processing up to 3 TOPS
Operating Temperature 0 to 90 °C
Radiation Tolerance > 20 kRad (Si)
Thermal interface customisable for satellite architecture
SOFTWARE ECOSYSTEM64-bit Linux
Deep Learning Accelerator fed with Caffe or TensorFlow models
FPGA reconfigurable in orbit
REDUNDANCYPossibility to introduce additional redundancy to each version
FORM-FACTOR1U

Do you want to find out more?
Contact us!

Key applications

Image segmentation and object detection

Signal quality enhancement

Data compression and encryption

Spacecraft autonomy

Optical navigation

Many more!

Contact

Write, call us or meet us directly!