BrainChip Holdings Ltd recently unveiled its AKD1500, a neuromorphic Edge AI accelerator co-processor, at the Embedded World North America exhibition. Designed to meet the rigorous demands of battery-powered and heat-constrained environments, this chip represents a significant leap in edge computing efficiency.
The AKD1500 is built on the Akida™ neuromorphic processing engine, utilizing a purely digital, event-based architecture that mimics the human brain by processing only essential sensor inputs. This breakthrough approach allows the chip to achieve an impressive 800 giga operations per second (GOPS) while consuming less than 300 milliwatts of power. This efficiency—roughly 1mW per GOPS—sets a new benchmark for Edge AI, enabling sophisticated AI capabilities in wearables, smart sensors, and industrial IoT devices without the need for cloud dependency.
A standout feature of the AKD1500 is its built-in on-chip learning. Unlike conventional accelerators that require cloud-based retraining, the AKD1500 can learn and adapt directly on the device, ensuring secure application personalization. It integrates seamlessly with x86, ARM, and RISC-V platforms via PCIe or SPI interfaces, and developers can leverage the MetaTF™ software development environment to deploy models using standard TensorFlow/Keras and PyTorch APIs. By moving AI closer to the sensor, BrainChip is paving the way for ubiquitous intelligence in smart homes, healthcare, and defense industries.
AKD1500 Technical Specifications
| Feature | Specification |
| Architecture | Akida™ Neuromorphic Neuron Fabric; On-Chip Conversion Complex |
| Performance | Up to 800 Effective GOPS |
| Power Efficiency | 1mW/GOPS; Operating under 300mW |
| Process Technology | 22 nm FD-SOI CMOS (GlobalFoundries 22FDX®) |
| Memory | 1MB On-Chip Local memory; SPI D/Q/O expansion interface |
| Interfaces | PCIe Gen2 Endpoint; SPI S/D/Q/O Peripheral Interface |
| Clock Frequency | 5 MHz – 400 MHz |
| Package | 7×7 mm MFCTFBGA169, 0.5 mm pitch |
| Software Support | MetaTF™; TensorFlow/Keras, PyTorch, and ONNX APIs |
Distribution and Pricing
- Availability: AKD1500 samples are currently available for select partners and customers.
- Mass Production: Volume production is scheduled to begin in Q3 2026.
- Pricing: Official unit pricing has not been publicly disclosed; however, the chip is marketed as a cost-effective solution for large-scale AIoT deployments.
- Early Adopters: The product has already been delivered to partners, including Parsons, Bascom Hunter, and Onsor Technologies for medical and defense applications
