FPGA/Programmable Device Content: Seeking your feedback on what you are interested in

My choice was "problems to be solved in machine learning" although just Machine Learning Applications would have sufficed.  There are a ton of edge IoT or similar applications where a low power FPGA would be quite useful especially considering it can be be reprogrammed with new functionality without replacing it.  This is important when the device is in a location that is either not easily reached or even impossible to get to once the device has been deployed.  Microsoft is investing heavily in FPGA development in their Azure cloud offering especially with regards to edge devices. So, rather than deploy a GPU that is designed to handle a particular Machine Learning model, and FPGA based device could be deployed and if a new model is required, it can be reprogrammed for the new model.   This is rather cool from my perspective.

From my experience, FPGAs are a different animal to the typical microprocessor/microcontroller development and are typically assigned to a hardware team.  I've seen them used in prototyping or pre-fab testing for a device so the firmware folks can test code before an ASIC is spun.  I've also seen them used as the main processing device in a board as the costliest part on a particular board.  I think the current trend of seeing FPGAs used more and more in standard applications is due to the FPGAs becoming cheaper as well as requiring less power where historically they were very costly and required significant power.  With the FPGA SOCs such as Xilinx Zynq or Intel's Agilex where a multicore ARM processor is married with a FPGA in the same silicone and the two work in unison the application possibilities of an FPGA have expanded and companies are taking advantage of its use.  

As a trip to Mars becomes more and more of a possibility, the idea of having a device that can be used like a space age Swiss Army knife would have a big advantage over one that is designed for a particular use case.

Watch out for that power thing - there are low power FPGAs like Lattice's ICE40 series but the biggest of them offers only 8k LUTs,

Lattice do have some "AI" or machine learning support for them but it is a million miles from the GPU type of applications. It's probably the best place to start for self teaching or if you have a low budget.

The Lattice tools are cheap.free, way simpler than X or I/A and the devices are easy to understand (and with a nod of respect to John Beetem, there are open source tools as well).

(Admission of bias here - I've used mainly Lattice parts for more than 10 years so I have that comfortable feeling of familiarity with them. I've used X and I/A parts too, but it feels like wearing your best suit to go to a "do" rather than your favorite cords for walk with the dog.)

When Xilinx and Intel/Altera talk about low power (in an AI context)  they tend to mean that core current is only a few amps rather than a few 10s of amps !

MK

Watch out for that power thing - there are low power FPGAs like Lattice's ICE40 series but the biggest of them offers only 8k LUTs,

Lattice do have some "AI" or machine learning support for them but it is a million miles from the GPU type of applications. It's probably the best place to start for self teaching or if you have a low budget.

The Lattice tools are cheap.free, way simpler than X or I/A and the devices are easy to understand (and with a nod of respect to John Beetem, there are open source tools as well).

(Admission of bias here - I've used mainly Lattice parts for more than 10 years so I have that comfortable feeling of familiarity with them. I've used X and I/A parts too, but it feels like wearing your best suit to go to a "do" rather than your favorite cords for walk with the dog.)

When Xilinx and Intel/Altera talk about low power (in an AI context)  they tend to mean that core current is only a few amps rather than a few 10s of amps !

MK

I have one of the Mpression  Odyssey MAX 10 FPGA and BLE Sensor Kit which has an Altera/Intel MAX 10 FPGA and, yes, only 8k Logic Elements. However, it is coin cell powered and is reconfigurable via a smart phone over BLE which is a bit impressive.  

As far as Open Source FPGAs, I have spoke with Luke, who is local to me, from TinyFPGA regarding his Open Source FPGAs and the tools he uses.   He too uses Lattice FPGAs and his boards can be programmed with either the open source Project IceStorm tools or Lattice iCEcube2. He also hosts The Hobbyists Guide to FPGAs on Hackaday.io.  Luke has done a good job both promoting as well developing open source FPGA tools. 

The current trend in IoT, or as Dave Jones from the AMP Hour calls it "The Internet of Terror", is to push more of the processing to the edge so less data is sent back to the server farm where it is expensive to have to parse through a ton of unnecessary data.  NVIDIA is pushing their Jetson line, especially the Nano, for this purpose, but Microsoft is teaming with Intel to come up with an low power FPGA solution so it will be more flexible.  I'm sure there will be trade offs with regards to the size of the FPGA, but with two tech giants like these working on a solution, I am sure they will come up with a low power FPGA that still has a fair amount of LEs on it.