BrainChip confirms validation of its Akida neural processor

1 min read

BrainChip, a provider of ultra-low power AI technology, has announced that the validation of its Akida Neuromorphic System-on-Chip (NSoC) design with functional silicon.

The Akida NSoC is an integrated circuit that includes multiple interfaces, Data-to-Event Converters, a CPU complex, on-chip memory and a neuron fabric to implement a complete neural network with no need for any external components.

The device was produced at a TSMC foundry and uses a standard logic 28nm process. The Multi-Project Wafer (MPW) used by BrainChip earlier this year in order to accelerate the delivery of the integrated circuit engineering samples, provided packaged and tested devices for internal validation and testing in August 2020. These devices will support the demands of customers participating in the company's Early Access Program (EAP).

The Akida NSoC neuron fabric is comprised of cores that are organised in groups of four to create nodes, which are mesh networked. The cores can be implemented for either convolutional layers or fully-connected layers. This flexibility allows users to develop networks with ultra-low power Event-Based Convolution as well as Incremental Learning. The nodes also can be used to implement multiple networks on a single device.

The NSoC will be available in an M.2 PCIe format card and a USB alternative. Device drivers and SDK with example neural networks will be provided for both Linux and Windows operating systems.

Commenting Louis DiNardo, BrainChip CEO said, “We have validated the functionality of the Akida silicon and performed significant testing. The devices received from the MPW will provide engineering samples and evaluation boards for our early access customers. Now our focus is on continued development of software and firmware drivers.

"We expect to produce a production mask set in the fourth quarter of 2020 and then move forward with volume production wafer fabrication, assembly and test operations.”