What's Stopping You Building Your Next Project with an FPGA? (Please, Don't Blame the Cost!)

rscasny I just posted a comment for this in the vivado discussion. To answer this question in short, I find very few projects worthy of an FPGA. An IoT application with FPGAs sounds fascinating though it would be a terrible waste of hardware capabilities to employ it to send temperature over the network.

I would love to flex my Verilog muscles once again if a suitable challenge presents itself.

Hi Randall, I have been thinking about getting started with FPGAs this year. I've got myself a "TinyFPGA". I've gone for the $12 A version which is based on the Lattice Mach XO2-256. That's it's about as simple as they get.

TinyFPGA A-Series User Guide

The main reasons stopping me getting going.

1) The need for a change of mindshift from sequential programming to logical building blocks.

2) Learning a whole new toolset to get these things programmed.

3) Killer app. Given that I can do most things I need using a microcontoller I've no need to swap to FPGAs.

Most of the FPGA dev boards I've seen are of the sledge hammer style. They've got masses of processing power and are suited to things like processing video signals or SDR rather than the kind of things I normally do such as flashing a few LEDs and moving some servos.

I have been looking at the Arty and Digilent tools and there do look to be some good examples and tutorials. I'll maybe be applying if I can thing of a suitable task to roadtest the board.

I agree. For me the reasons are the same as Andy's 1/2/3.

To sum up: "Learning Curve"

ps, Andy - what is the startup time for these things?

I have a logic application (brake/signal light converter), but it needs to pretty much be instant on (so it can be powered by the brake/signal lights themselves).

-Nico

That's a good question. I should be able to tell using the PicoScope test rig I built. I'll let you know.

oh hehe, I don't need anything scope accurate - just more along the lines of "if I press the brakes, will the guy behind me have enough time to stop?", so a third of a second or faster would do the trick.

FPGA normally read their configuration from an external memory (its non-volatile), so they need some startup time for re-configuration. There are FPGA such as the Spartan-3AN that come with internal Flash memory, but even they must do a re-configuration on startup. Actel has the e.g. the ProAsic3 family for which they claim instant-on functionality, but I did not read through all data sheets to verify that.

Greetings,

While FPGAs can be big and frighting, there cousin devices, CPLDs are more often an easier and friendlier fit to people working on embedded projects.  I have designed a lot of products where the majority of the work is being done on the microcontroller and a small amount of work is done with a CPLD.  Some examples of these designs are high speed sampling systems and high speed (non-standard) communications, where the data is in small bursts.  In these instances the CPLD can take measurements and/or collect data, while the microcontroller is doing its normal processing.  An interrupt or polling can be used to determine when data is ready. The microcontroller and CPLD can be configured to exchange data using simple I2C, serial or parallel buses.

Typically I find FPGA parts packaged in very high pin count packages with huge gate counts, while CPLDs come in smaller pin count packages with a much smaller gate count.

Gene

For me, there are different factors to consider whether a FPGA is a valid solution or not.

1. not everything needs an FPGA: many times simple logic or a CPLD is enough, sometimes you can use the flexibility of a MCU (and can use all the nicer peripherals)
2. packages and power requirements: many (all?) FPGAs some in hard-to-use packages such as BGA, and their power requirements are not always simple (multiple different voltages, high currents, decoupling requirements). You can solve that by using existing breakout boards (e.g. the TinyFPGA mentioned above), but it adds to the cost and needs room.
3. learning curve: procedural programming in C (e.g. the Arduino) is quite accessible, easy to grasp and there is a lot of documentation out there. Learning Verilog or VHDL with its very hardware-driven concepts takes more time, and its more difficult to debug (no step-by-step debugger...). This can be solved by taking the needed time, starting small and learning systematically.
4. cost of tooling: when the free versions of the IDEs are not enough, it can get expensive (esp. when you need these IP modules, e.g. to implement a CPU core in an FPGA, or for special peripherals). Fortunately this seems to be solved by time, more and more stuff seems to be available on the free tool versions.

That said, an FPGA is the best solution when you have to do with high-speed data and signals, especially when its massively in parallel. A good example are logic analyzers, or anything to do with image handling (think of 4k digital movie cameras).

The Xilinx XC9572XL is the go-to device for the hobbyist. Powerful enough to solve real problems (for which conventional logic gets too large) and fast enough at the same time (faster than most MCUs). It can handle 5V on its inputs, so its compatible to many peripherals. Its quite cheap and there are many breakout boards available. Xilinx ISE is free, and programmers are cheap.

The data sheet says instant on for the Lattice MachXO2 it has inbuilt flash.

The Spartan 7 in the Arty board seems to use external flash so Roadtesting boot time would definately be worth while