BrainChip launches Sequora core for generative AI on battery powered devices

BrainChip Holdings Ltd announces in a press release Sequora Core IP, a processor core for running generative AI directly on battery powered devices. It supports language models and other sequential AI workloads with up to 3 billion parameters.

Sequora uses a state space architecture designed to keep memory use fixed as a data stream or conversation grows. It can work alongside BrainChip's Akida platform, with Akida processing sensor data and Sequora handling tasks such as language generation, voice interaction and AI agents.

The core is available now as licensable semiconductor intellectual property. Customers can integrate it into their own system on chip designs and manufacture it using their selected foundry and process node.

BrainChip supplies Sequora as synthesizable register transfer level code. The package includes verification tools, synthesis scripts, timing constraints, a compiler, a C++ runtime library and a collection of models.