Products
Applications
Solutions
The BrainChip TENNs algorithm introduces a new paradigm in edge computing by enabling low-power LLM on the edge – a capability previously restricted to cloud data centres or high-performance computing environments. Unveiled at Embedded World 2025, this innovation marks a turning point for AI applications in consumer electronics, robotics, assistive care, and embedded intelligence.
TENNs stands for Temporal Event-based Neural Networks, a neural architecture developed by BrainChip to deliver efficient recurrent computation with high adaptability and low power requirements. Unlike transformers or RNNs that typically require intensive compute resources and memory, the BrainChip TENNs algorithm is designed to operate on-device, in real time, and on ultra-low-power hardware.
The demo showcased at the event ran a large language model entirely on an FPGA, completely offline. Users could ask questions and receive intelligent responses—no cloud inference, no API latency, and no privacy concerns. This proof of concept demonstrates how TENNs enables intuitive machine interaction on devices constrained by power, size, or connectivity.
The versatility of the BrainChip TENNs algorithm spans beyond LLMs. It supports workloads such as audio denoising, keyword spotting, and automatic speech recognition (ASR). Developers can train models using standard GPU-based frameworks and then deploy them in a compact recurrent form for ultra-efficient edge inference on BrainChip-compatible IP.
What sets TENNs apart is its ability to maintain conversational and analytical intelligence without compromising power budgets. The algorithm architecture is a perfect match for edge AI use cases in social robotics, healthcare monitoring, automotive HMIs, and smart appliances. Tony Dawe, BrainChip CTO, described it as “an artificial son” – capable of explaining, interpreting, and assisting – without ever needing a cloud connection.
The implications are profound. In environments like eldercare or smart homes, the BrainChip TENNs algorithm could support always-on assistants that respect user privacy and deliver real-time feedback with zero latency. In industrial or defence applications, it enables autonomous operation in disconnected environments.
With TENNs, BrainChip isn’t just improving edge AI – they’re redefining what edge AI can do. For design teams seeking local intelligence, context awareness, and ultra-low-power operation, TENNs represents the next generation of neural network deployment.