Neuromorphic Processor

Overview

BrainChip’s first-to-market neuromorphic processor IP, Akida™, mimics the human brain to analyze only essential sensor inputs at the point of acquisition, processing data with unparalleled efficiency, precision, and economy of energy. Keeping AI/ML local to the chip and independent of the cloud dramatically reduces latency while improving privacy and data security

Key Features

  • Neural Processing Unit (NPU) at memory compute architecture implementing Integrate and fire neuron.
  • Emulate multiple neurons with configurable Synapses.
  • Compute only when events occur.
  • Up-to 4 bits for weights and activation.
  • 4 NPU interconnect to make up 1 node.
  • Multiple Node interconnect via packet switched mesh network.
  • Same hardware supports standard CNN after conversion to SNN, and native SNN.

Benefits

  • Ultra-low latency
  • Runs multiple networks in real-time
  • Performs one-shot learning
  • Remarkably power efficient
  • Cloud independent
  • Flexible and quick to deploy
  • Privacy and security protected

Block Diagram

Neuromorphic Processor Block Diagram

Technical Specifications

Availability
now
×
Semiconductor IP