BrainChip strengthens AI capabilities with new patent for Akida neuromorphic processor

Go to Imelda Cotton author's page
By Imelda Cotton - 
BrainChip ASX BRN new neural network patent

Artificial intelligence (AI) specialist BrainChip Holdings (ASX: BRN) has strengthened its intellectual property portfolio with the issuance of a new patent by IP Australia.

Patent AU2022287647 (An Improved Spiking Neural Network) was granted for the company’s neuromorphic processor Akida and machine learning framework MetaTF, designed to seamlessly transform contemporary neural networks into event-based or spiking networks.

Akida mimics the human brain by analysing only essential sensor inputs at the point of acquisition and is reported to be capable of processing data with superior performance, precision and reduced power consumption.

MetaTF is used for the creation, training and testing of neural networks and supports the development of systems for edge AI on the Akida processor.

The technologies synergise with the converted spiking networks, enabling the streamlined deployment of edge learning algorithms and unlocking use cases which conventional AI tools or solutions are unable to attain.

Low-shot learning

The new patent was granted earlier this month and is considered a valuable asset to facilitate low-shot learning, meaning learned higher-end features can be utilised when learning a new class while retaining other neurons or classes.

Learning is performed by adding neurons to the final layer of a previously-trained network to represent a new class, with the neural network weights of the added neuron being trained with only a few samples while the remainder of the network remains unchanged.

Example applications include biometric facial recognition, speech recognition and anomaly detection in industrial systems.

Advancing AI work

BrainChip chief executive officer Sean Hehir said the new patent would help advance the company’s AI work.

“It confirms our leadership in the field of low-shot learning as an alternative to traditional machine learning processes, which are, by contrast, extremely inefficient, expensive, and energy-intensive,” he said.

“This patent illustrates why our neuromorphic technology is perfect for edge AI applications where low-shot learning, ultra-low power consumption and efficiency are key differentiators for customers seeking a competitive advantage.”

Patent portfolio

BrainChip now has 20 issued patents in its portfolio, comprising 13 in the US, five in Australia and two in Canada.

It has an additional 23 pending patent applications across the US, Europe, Australia, Canada, Japan, Korea, India and Israel.