BrainChip Akida 2 (Neuromorphic Edge AI)
Overview
BrainChip Akida 2 is a second-generation neuromorphic AI processor using a brain-inspired event-driven spiking neural network (SNN) architecture. With 1.5W ultra-low power consumption, it targets always-on edge AI inference including keyword spotting, gesture recognition, and sensor analytics.
Core Specifications
| Item | Parameter |
|---|---|
| Architecture | Akida 2 (event-driven SNN) |
| Process | TSMC 28nm |
| Neural Cores | 20+ (configurable) |
| On-Chip Memory | Event-based SRAM |
| Power Consumption | |
| Power Consumption | ~1.5W |
| Form Factor | PCIe card / M.2 module |
| Quantization | 1-4 bit native |
| Programming | MetaTF (CNTK/Keras/TensorFlow) |
Key Features
- Event-driven: computes only on events, not frames
- Ultra-low power: 1.5W for always-on operation
- Native SNN: no conversion from ANN needed
- 1-4 bit precision: natively quantized
- On-chip learning: support for incremental learning
Use Cases
- Always-on keyword spotting
- Gesture recognition
- Vibration analysis / predictive maintenance
- Edge sensor fusion
- Low-power object detection
Related Cards
- IBM NorthPole (Neuromorphic)
- Loihi 2 (Intel Neuromorphic) — 页面待创建