Processor IP Brings Generative AI to Edge Devices

BrainChip recently announced a major advancement in edge computing with the launch of its Sequora processor IP core. This new technology enables battery powered edge devices to run generative AI and large language models directly on the hardware.

Historically, generative AI required massive cloud servers due to extreme memory and power demands. Sequora changes this paradigm by utilizing a state space model architecture. This design maintains a fixed size state to represent input history. Consequently, memory requirements remain constant regardless of sequence length. This breakthrough allows devices like earbuds and sensor nodes to process complex sequential AI tasks without sending data to external servers.

The technology addresses critical industry challenges. It eliminates the memory bandwidth bottlenecks that typically stall generative models. It also ensures predictable runtime branching, which is essential for automotive and medical certification. By keeping data local, the processor guarantees privacy and minimizes latency for real time applications.

Sequora is designed to pair seamlessly with the existing Akida neural processor IP from BrainChip. While Sequora handles generative tasks and voice to voice interactions, Akida processes streaming sensor data. This includes vibration, audio, radar, raw radio frequency data, and biological signals.

This dual approach allows a single system to manage both conventional inference and generative inference. Customers in aerospace, defense, and industrial sectors can now integrate both capabilities into one custom silicon design. This integration significantly reduces the size, weight, and power footprint of mission critical edge systems.

The Sequora core is available immediately as licensable semiconductor intellectual property. Developers receive process portable and fully synthesizable register transfer level code. The package also includes verification tools, synthesis scripts, and a software development kit with a static graph compiler.

This release marks a pivotal moment for edge artificial intelligence. Devices will no longer be limited to simple signal recognition. They can now understand context and respond intelligently while operating on minimal battery power. This evolution ensures that artificial intelligence becomes a practical and private utility in everyday infrastructure.