BrainChip releases Xilinx-based neural network acceleration board for object recognition, surveillance

September 12, 2017 ECD Staff

BrainChip has released the BrainChip Accelerator, an 8-lane PCIe expansion board that is capable of delivering a 6x increase in the speed and accuracy of object recognition functions in BrainChip Studio software. In tandem, the BrainChip accelerator supports 16 simultaneous video channels per card, and can be deployed in existing video surveillance systems without requiring updated power supplies or thermal management equipment.

The BrainChip Accelerator consists of six neuromorphic cores in a Xilinx Kintex Ultrascale FPGA. Each BrainChip Accelerator core performs user-defined image scaling, spike generation, and spiking neural network comparisons to identify objects.

The individual cores are capable of processing up to 100 fps while consuming 1 W of power, which signifies a 7x improvement in frames/second/watt over GPU-accelerated deep learning networks like GoogleNet and AlexNet.

“BrainChip’s spiking neural network technology is unique in its ability to provide outstanding performance while avoiding the math-intensive, power-hungry, and high-cost downsides of deep learning in convolutional neural networks,” says Christoph Fritsch, senior director for the industrial, scientific, and medical business at Xilinx.

BrainChip representatives claim that the BrainChip accelerator is the first commercial implementation of a hardware-accelerated spiking neural network system. 

The board is compatible with Windows or Linux platforms.

More information is available at www.brainchipinc.com/products.

Previous Article
HCC Embedded advanced middleware suite available for Xilinx Zynq-7000 SoCs

Next White Paper
A Low-risk, COTS Approach to Building Safety Certifiable Processing Subsystems
A Low-risk, COTS Approach to Building Safety Certifiable Processing Subsystems

How to affordably decrease safety critical processing subsystem development time and program risk using DO-...

×

Stay updated on processing and related topics with the Processing edition of our Embedded Daily newsletter

Subscribed! Look for 1st copy soon.
Error - something went wrong!