BrainCard, putting neurons at work for the Intelligence of Things

BrainCard is a pattern recognition module featuring a NeuroMem CM1K chip with 1024 trainable neurons ready to learn objects and events on the board and recognize new objects and events immediately, reporting novelty if applicable.

  • It is a perfect shield for Arduino and Raspberry PI boards to develop smart appliances and near-sensor low-power edge intelligence.
  • It is a perfect extension for the Intel Arduino/Genuino101 to upgrage from 128 neurons (128 bytes per neurons) to 1024 neurons (256 bytes per neurons)

Neurons are ready for immediate learning and recognition of patterns extracted from signal, images or other data and broadcasted to the neurons via a simple SPI protocol.

Sensor data must be assembled into feature vectors of up to 256 bytes. Their learning can be triggered by external user inputs sending category values to the neurons. Once trained, the neurons can continuously monitor vectors extracted from new sensor data and report known patterns or events, or on the contrary novelties or anomalies. The knowledge built by the neurons can be saved for backup and transfer to other BrainCard.

FPGA programmers can configure the Xilinx Spartan6 to acquire data directly from the on-board audio MEMs, A/D converter, GPIO lines and compatible RaspiCam module. Outputs can be directed to SD card, mini HDMI or else.