Implemented a 6-core processing unit, see this FPGA neuromorphic circuit board!

BrainChip Holdings has just released a PCIe server accelerator, the BrainChip Accelerator, which can simultaneously process 16 channels of video in multiple video formats using a pulsed neural network instead of a convolutional neural network (CNN). The BrainChip accelerator card uses the Xilinx Kintex UltraScale FPGA to implement the BrainChip Spiking Neural Network (SNN) processor for the 6-core processing unit.

This is a photo of the BrainChip accelerator card:

BrainChip Accelerator card with six SNNs instantiated in a Kintex UltraScale FPGA

Each BrainChip core performs fast user-defined image scaling, pulse generation and SNN comparison to identify targets. The SNN can be trained with low resolution images as low as 20x20 pixels. According to BrainChip, SNNs used in the BrainChip accelerator core excel at identifying objects in low-light, low-resolution, and noisy environments.

The BrainChip accelerator card can handle 16 channels simultaneously, with an effective throughput of over 600 frames per second, while the entire card consumes only 15W. According to BrainChip, this speed is increased by 7 times/second/watt compared to CNN-based deep learning neural networks such as GPUNetNet and AlexNet. This is a chart from BrainChip that illustrates this statement:

SNN mimics human brain function (synaptic connections, neuronal thresholds) closer than CNN and relies on models based on peak time and intensity. This is a graphic of BrainChip comparing CNN model and impulse neural network model: