BrainChip Holdings has announced the AKD1500 co-processor capable of delivering up to 800 giga operations per second (GOPS) while operating below 300 milliwatts, providing a power-efficient solution for edge AI tasks. This makes it suitable for use in battery-powered or thermally constrained systems such as wearable electronics, smart sensors, and compact embedded devices.
The co-processor connects to x86, ARM, and RISC-V platforms via PCIe or serial interfaces, supporting flexible integration into existing architectures. Its design allows system developers to add neuromorphic processing capabilities to existing SoCs or embedded microcontrollers without requiring full hardware redesign.
Applications include AI-enabled sensing for medical, defense, and industrial systems, as well as use in healthcare devices, smart infrastructure, and consumer products. Early implementations of the device are in development with several partners for real-time pattern recognition and adaptive signal analysis.
The AKD1500 uses BrainChip’s Akida neuromorphic architecture, which supports event-based data handling and on-chip learning. This architecture processes information as spikes or events rather than continuous data streams, reducing power consumption and latency.
Machine learning engineers can develop models using the company’s MetaTF software environment, which supports standard TensorFlow/Keras formats and allows conversion, quantization, compilation, and deployment for Akida-based hardware. This workflow aims to reduce model development time and simplify the process of implementing neuromorphic AI at the edge.
AKD1500 samples are available, with production volumes expected in the third quarter of 2026.
Leave a Reply