 |
 |
Neuromorphic X1 is a compact, ultra-efficient analog in-memory compute (AiMC) IP macro featuring a 32 × 32 1T1R crossbar for low-power matrix operations—ideal for edge AI and embedded IoT deployments.
View Product Page
- Analog in-memory MAC engine in a 32 × 32 1T1R array
- Drop-in, licenseable macro for energy-efficient computation at the edge
- Designed for embedded / IoT AI where energy and latency are critical
Neuromorphic X1 brings in-memory compute to compact silicon form factors, reducing data movement and significantly lowering power consumption for AI workloads.
It is optimized for:
- Always-on AI (keyword spotting, anomaly detection)
- Edge AI in battery-powered devices
- Low-latency on-device processing
At a glance:
- Architecture: 32 × 32 1T1R analog crossbar
- Compute Type: Multiply–Accumulate (MAC) in crossbar array
- Target Applications: TinyML, sensing, embedded AI inference
- Benefits: High energy efficiency, low latency, small area footprint
Figure Explanation:
This diagram illustrates the data write process to the Neuromorphic X1’s 32×32 ReRAM crossbar via a Wishbone interface.
- 1 page = 4 bytes (one wordline)
- Data packets (P1–P5) are transferred from the Wishbone bus to the page buffer
- From the buffer, data is written into the corresponding rows and columns of the ReRAM crossbar
- Each write cycle takes approximately 10 µs per page, enabling fast updates while maintaining energy efficiency
Source: Neuromorphic X1 – bmsemi.io