ItInsightsWorld Breakdown: BrainChip Sequora and Akida Edge IP for Local Generative AI

BrainChip pairs Sequora processor IP with Akida neural processor IP to bring generative AI and large language model (LLM) workloads onto battery-powered edge hardware, including earbuds and sensor nodes. Each core targets a different class of task: Sequora focuses on generative and sequential AI, while Akida is used for streaming sensor data.

Task Split Across the Two IP Cores

IP core	Primary role	Workload examples
Akida	Processes streaming sensor inputs	Vibration, audio, radar, raw RF IQ data, and biological signals
Sequora	Runs generative and sequential AI	On-device LLM inference, voice-to-voice interaction, and AI agents

State-Space Model Support

Sequora is optimized for state-space models (SSMs). Because these models use a fixed-size state to represent input history, memory requirements remain constant regardless of sequence length. That characteristic supports generative AI deployment on resource-constrained edge devices.

Unified Compute Array and Software Updates

Sequora uses a unified multiply-accumulate (MAC) array for signal processing and neural-network computation, removing the need for a separate DSP preprocessing stage. The instruction-set architecture (ISA) is also presented as enabling new algorithms to be delivered through software updates.

IP Availability and Deliverables

Sequora is currently available as licensable semiconductor IP from Brainchip Holdings. Delivery is described as process-portable, fully synthesizable RTL, accompanied by verification IP, synthesis scripts, and an SDK containing a compiler, C++ runtime library, and model zoo.