Behold The BrainChip PCIe Card, World's First Accelerator For Neuromorphic Computing

BrainChip has just announced a new add-in card that is specifically designed to accelerate neuromorphic computing. Neuromorphic hardware is tasked with mimicking the pathways of the human brain by using neural technology to process large amounts of data parallel in bursts or "spikes" over short span of time. In this case, the BrainChip Accelerator will be used to greatly enhance the speed of object and facial recognition.

The BrainChip Accelerator is described as an 8-lane PCIe add-in card, which when used in conjunction with BrainChip Studio software can provide a 6x uplift in the speed and accuracy of object recognition. In addition, the accelerator is capable of processing 16 simultaneous video streams while operating under a low-power profile.

The accelerators are being targeted at the video surveillance market (for live or archived video streams), and can be used as a drop-in replacement for existing solutions, while at the same time not requiring a PSU upgrade to take advantage of the additional processing power. Training the system can be done using 20x20 pixel images, while low-light and high noise situations aren't enough to stymie the accelerator. 

Processing duties are handle by a six-core Xilinx Kintex Ultrascale field-programmable gate array (FPGA), with each core capable of processing 100fps at 1-watt per core. BrainChip makes it known that this fps/watt metric puts it at a 7x advantage over GPU-based solutions like GoogleNet and AlexNet. Total board power is listed at 15 watts, which should result in a lower total cost of ownership over the life of the device compared to use a third-party server to handle computations.

BrainChip Studio

"There is an estimated four exabytes of video data stored in video surveillance systems," said Robert Beachler, BrainChip's Senior Vice President of marketing and business development. "The ability of BrainChip Accelerator to process video frames six times faster, while improving the accuracy of object recognition, is a significant force multiplier. It is also a further demonstration of the valuable role that artificial intelligence can now play in these applications."

We have no doubt that law enforcement agencies (and other organizations) will be taking a long and hard look at solutions like this to aid in crime fighting and general surveillance operations.