UPGRADE YOUR BROWSER

We have detected your current browser version is not the latest one. Xilinx.com uses the latest web technologies to bring you the best online experience possible. Please upgrade to a Xilinx.com supported browser:Chrome, Firefox, Internet Explorer 11, Safari. Thank you!

cancel
Showing results for 
Search instead for 
Did you mean: 

nn-X: A Zynq-based low-power mobile coprocessor for accelerating deep neural networks

Xilinx Employee
Xilinx Employee
0 0 43.4K

Deep neural networks that achieve state-of-the-art perception in both vision and auditory systems make sense of raw input data and parse them into symbols. A typical deep neural network consists of multiple convolution layers followed by a classification module. Deep neural network models are computationally very expensive, requiring up to billions of operations per second, usually requiring high-performance processors like server CPUs and GPUs to process large deep networks in real-time. Such processors consume a lot of power. A recent invited paper presented by a team from Purdue University at the 2014 IEEE Conference on Computer Vision and Pattern Recognition Workshops held in Columbus, Ohio demonstrated a 240Gops mobile (low-power) processor for deep neural networks based on the Xilinx Zynq SoC. (“A 240 G-ops/s Mobile Coprocessor for Deep Neural Networks.”)

 

The block diagram below shows the Zynq-based low-power mobile coprocessor for accelerating deep neural networks. It’s called “nn-X”for “neural network next.”

 

 

nnX Neural Network Coprocessor.png

 

nn-X: A Zynq-based low-power mobile coprocessor for accelerating deep neural networks

 

 

The coprocessor efficiently implements pipelined operators with large parallelism, delivering very high performance per unit power consumed. The nn-X coprocessor advances the state-of-the-art in multiple domains:

 

  • In data-streaming architectures
  • In efficient processing of deep neural networks
  • In providing high performance in real applications
  • In efficient use of system power

 

Work was done on a Xilinx ZC706 Eval Kit. For a full description of the design, please take a look at the paper referenced and linked above. For a quick look at what the coprocessor can do, here’s a face-recognition result on a couple of familiar faces taken from the paper:

 

 

Kirk and Spock.png

 

How about that? It even works on Vulcans.

Tags (2)