To support the next generation of space missions, improved platform autonomy is in high demand to handle the ever-increasing mission complexity. The MAG-AIA-30101 AI processor targets AI applications in space where high-reliability guarantees are required to ensure mission success. Because of its modern architectural design, the processor supports state-of-the-art AI and machine learning models that are essential for implementing a high-degree of platform autonomy.
MAG-AIA-30101 is a high-performance and fault-tolerant AI co-processor soft IP for integration in space-grade system-on-chip (SoC) solutions. MAG-AIA-30101 enables the robust deployment of state-of-the-art AI models for on-board space applications. As such, MAG-AIA-30101 is a crucial building block for enabling platform autonomy in high-criticality operational scenarios. The soft IP is integrated within a broader open AI compiler ecosystem for improved programmability and productivity.
(int8) up to 10 TOPS (1 OP = 1 MAC)
Size-scalable soft IP (up to 16k MAC units)
Architecture optimized for convolutional workloads
Based on Apache TVM
1W (at full size)
4 MB L1 scratchpad memory
AXI-based loosely coupled architecture
SPARC, RISC-V, ARM, … (ISA agnostic)