Fabricated
AF-A1
Standalone AI acceleration for radiation-tolerant designs.
A standalone AI accelerator for systems that already have a host processor and need dedicated radiation-tolerant inference.

Applications
Use it when a payload already has a host processor and needs dedicated onboard AI.
- Streaming transient-signal analysis at the sensor
- Anomaly detection on telemetry and sensor data
- Sensor data triage next to an existing flight computer
At a glance
- Type
- AI accelerator
- Throughput
- 30 GOPS
- Evaluate
- Via dev board
- Hardening
- Rad-tolerant
Specifications
Core
- Type
- Standalone AI accelerator
- Host interface
- Parallel
- Hardening
- TMR + ECC-protected memory
A pure accelerator.
Nearly the whole die is inference hardware.
Performance
| Parameter | Min | Typ | Max | Unit |
|---|---|---|---|---|
| AI throughput | 30 GOPS (projected) | |||
| Efficiency | 100-300 GOPS/W (projected) | |||
| Core clock | -- | 11 | -- | MHz |
Array-scale parallelism.
Thousands of operations per cycle at modest clock rates.
Clock & supply
| Parameter | Min | Typ | Max | Unit |
|---|---|---|---|---|
| Core supply (VDD) | 1.62 | 1.8 | 1.98 | V |
| I/O supply (VDDIO) | 1.62 | 3.3 | 3.6 | V |
| Active current | Characterization in progress | |||
Supply current will be released after characterization.
Package
- Signal I/O
- 73 pads
Radiation
Radiation tolerance by architecture.
AF-A1 uses triple modular redundancy on critical logic, ECC-protected memory, and memory scrubbing.
TID and single-event test results will be published here as characterization completes.
Ordering
AF-A1 is evaluated on its dev board. Register interest and we'll contact you as boards come available.
Evaluation hardware for AF-A1
Specifications are preliminary and subject to change. Orders are subject to export screening.
