New BrainChip IP Core Enables Generative AI on Edge
The Sequora core allows devices to run large language models while keeping memory requirements constant.
Updated on Oct. 21, 2026 in Semiconductors
Live Poll
Would you trust your personal edge devices to process generative AI locally instead of the cloud?
BrainChip has released its new Sequora processor IP core, designed to enable generative AI and large language models on edge devices. This licensable semiconductor intellectual property is now available for integration into third-party hardware designs.
Why it matters
The Sequora architecture targets the memory-intensive bottleneck of generative AI, aiming to maintain constant memory requirements on resource-constrained devices. It provides a specialized pathway for deploying state-space models and neural networks without the heavy compute overhead typical of data-center-class hardware.
Sequora employs a unified multiply-accumulate array that performs signal processing and neural-network computation in a single operation. The design is optimized for state-space models, utilizing an instruction-set architecture that allows for algorithmic updates via software.
The players
BrainChip
A developer of neuromorphic and neural processor intellectual property designed for edge computing.
The details
The Sequora processor IP core operates by consolidating arithmetic tasks that typically require separate signal processing and neural units. By using a unified multiply-accumulate array—a circuit component that multiplies two numbers and adds the result to an accumulator—the design streamlines data flow to handle generative AI workloads. This hardware-level consolidation is designed to keep memory usage constant, allowing resource-constrained edge devices to run large language models that would otherwise exceed their local storage capacity.
Timeline
October 21, 2026: Sequora processor IP core becomes available for licensing.
The Tech Race
Sequora extends the BrainChip product line by pairing directly with the existing Akida neural processor IP to broaden the company's edge AI capabilities. This release represents a move to capture the emerging market for hardware-accelerated state-space models on compact devices.
Hardware manufacturers can now license the Sequora core to integrate advanced AI capabilities into edge devices, bypassing the need for cloud connectivity. Developers will receive a full delivery package including verification IP, synthesis scripts, and a software development kit to begin integration.
The takeaway
The Sequora release highlights the industry push to shift generative AI from cloud-based systems to local edge devices. Future performance benchmarks will determine if this unified array architecture provides a measurable efficiency gain over competing NPU implementations.
Further reading
For more on the current landscape of edge computing hardware, see our recent analysis in Semiconductors.
Source note: This article includes information reported by EDN.
Live Poll
Would you trust your personal edge devices to process generative AI locally instead of the cloud?
