Deep Learning Network on a Xilinx Cora Z7-10

Hello,

is it possible to apply a SeriesNetwork, which i created using the Deep Learning Toolbox on a Digilent Cora Z7-10 board. Maybe getting the VHDL code or the bitstream.

Best Regards, Martin

Answer 1

[Tom Richter]

Hi Martin,

You can use Deep Learning HDL Toolbox to deploy a network on FPGAs/SoSs from AMD or Intel. We don't support your board directly. However, you can generate the IP of the Deep Learning Processor and implement it for your board manually. We also have an example to register a new FPGA board. It can be found here.

Best regards,

Tom

Answer 2

[佳禹]

Hi, Martin

I am trying to deploy Neural Network (NN) on some custom board, I have the Zybo-z7-10 and something else.

You use the Cora Z7-10. If I was right, the core of it is zynq-7010. As the information provided by the mathwork support team, they think the resource of 7010 is too limited.

I am using the DL HDL tool-box. The needed size of the DL processor for deployment can be adjusted. But, as I tried it, the lightest size (only turn on the FC module) is

DSP RAM tales LUTs

DL_Processor 49 48 31291

It seems that the usage of the LUTs is out the scope of the 7010.

In my imagination, if the NN to be deployed is small, e.g., single FC layer and single neuron, the necessary resource must be relative small.

And intuitively, the 7010 must can realize such a light NN.

But, it seems that the structure of the DL processor is not like what I imaged.

So, now, on one hand, I am trying to further reduce the size of the DL processor; on the other hand, I am trying to use another board with core of kintex ultrascale FPGA.

I hope the information helps.

Regards,

Jiayu