Akida is an event-based spiking neural network processing core (NPC) that includes multiple training modes and both inference and unsupervised learning. It comes with a development environment to aid integration within ASICs.
The company launched its Akida NSoC as a discrete device back in September 2018 (see BrainChip launches spiking neural network SoC ) with enough neuron fabric to contain 1.2 million neurons and 10 billion synapse connections. The Akida NSoC includes an M-class processor. Comparisons to leading CNN accelerator chips showed gains of an order of magnitude in terms of images/second/watt, the company claims.
The Akida neuron fabric is now licensable for use in edge SoCs for tasks such as intelligent environmental controls, driver safety monitoring, vehicle safety and preventative maintenance, medical image classification, wearable health monitors, retail item recognition, access control, the industrial Internet-of-Things (IoT) and acoustic analysis.
Validation of the Akida NPC is available with the Akida Development Environment (ADE) which fully simulates the Akida NPC size, performance and power and can be verified in an FPGA. Akida can also be supplied as a chip within a USB dongle for use in development environments and cybersecurity applications, or a reference board.
Suppliers of smart sensors represent significant opportunities for early adopters to integrate the Akida NPC.
"The small size, high performance, low power and reconfigurable nature of the Akida IP provides ASIC integrators with a unique solution for incorporation into their products over multiple generations and serves to improve their ASIC performance," said Peter van der Made, BrainChip founder and CTO, in a statement. "I believe that we have the right solution at the right time. The market is ready for artificial intelligence in edge devices."
Related links and articles: