EK100
Low-power AI computing for satellites.
The EK100 expands what small satellites can do: they compress data, filter data, and detect phenomena with machine learning onboard, while using minimal electrical power.
Technical specs
| Processing cores | 1 × RV64IMAC, 625 MHz, 16 KB L1 iCache; 4 × RV64GC, 625 MHz, 32 KB L1 iCache |
|---|---|
| Memory | 4 GB of 32-bit DDR4; 8 or 16 GB of 64-bit DDR4 |
| Storage | 512 Mb SPI serial NOR flash; 16 GB eMMC, scalable to 256 GB |
| Interfaces | UART, CAN, SPI, I2C, LVDS, RS422/485, Ethernet, USB 2.0. SpaceWire available on request. |
| Supply voltage | 5 V |
| Power consumption | About 15 W |
| Operating temperature | −30°C to 85°C |
| Form factor | PC/104 |
| Software | 64-bit Linux and bare-metal applications. Machine-learning accelerator with support for TensorFlow, Caffe, MXNet, PyTorch, and DarkNet. Orchestration framework for updating and pipelining models in orbit. |
| Radiation tolerance | TID beyond 30 krad |