LSTM network implemention on FPGA

Hi! I want to implement an LSTM network on an FPGA for predictive detection of defects. I saw that Deep Learning HDL Toolbox not supported lstmlayer and sequenceInputLayer. So my question is : There is a way to implement an LSTM network on FPGA and/or generate VHDL/Verilog code from it and that from Matlab ?

Thanks in advance!

Answer 1

[Donald Baltus]

Hello Sami,

As you've observed, the Deep Learning HDL Toolbox does not currently support LSTM layers. Adding that support is on our roadmap but it does not exist today.

Furthermore, the Deep Learning HDL Toolbox does not currently support any layers including state. Without that capability, I don't believe it's possible to implement an LSTM network on an FPGA using the Deep Learning HDL Toolbox.

Please feel free to follow up if you have additional questions.

Regards,

[Sami Foery]

Thanks for the reply!

There is some examples to do time series forcasting without LSTM/RNN ? But with CNN for example ?

[Donald Baltus]

Hi Sami,

Its difficult to to time series forcasting without state .... but perhaps possible if you load an entire sequence as a single input.

I haven't found any suitable examples but I will keep looking.

[Sami Foery]

Thanks for you're return, I'll probably try another way for that.

[상인 이]

Still no update for LSTM or GRU?

[Donald Baltus]

LSTM will be supported in 2022B, GRU in 2023A.