Create Your Own CPU on an FPGA | Hacker News Reader

Create Your Own CPU on an FPGA

embeddedmicro.com

118 pointskardashev20 comments

Some guy started a series some weeks ago on designing a CPU for an FPGA: http://labs.domipheus.com/blog/designing-a-cpu-in-vhdl-part-...

For those interested in a more complex processor, RISC-V seems to be a processor on par with ARM in terms of features: http://riscv.org/

Both of these are excellent projects to check out if you're interested in such things. And, bonus, the RISC-V folks have produced an open source, modern manufacturing process, 1GHz core that I hope is a game changer in open source hardware. It's not quite on par with ARM (lacking the billions in investment), but it's in the ballpark and it's the real deal. Definitely worth keeping up with.

Does anyone have knowledge about the comparison of different FPGA boards?

I've seen the Mojo v3 from Embedded Micro, the Papilio line of dev boards, and the miniSpartan6+ from Scarab Hardware.

It looks like all these dev boards are pretty similar since they use almost identical xilinx chips. Ideally, I'd like to find a board for which there is a robust community, good support, lots of examples, and has enough capacity/performance to take on more complicated projects with increased learning/skill.

I'm leaning towards the Mojo v3 board, since it seems very easy and has shields to expand capability, but does anyone have any experience, recommendations, or "I wish I had known..." stories for FPGA development?

Basically, go with the market leaders, either altera or xilinx.

Altera DE brand is used in universities all over the world and they have their alteraforum (which is really old fashioned but does the job).

I would go with a DE2 or something like that. (Which has a Cyclone IV)

almost 10 years ago i wrote vhdl code on spartan series cards, and that time i noticed that Xilinx IDE was really far from Visual Studio. hope things have changed since then. So my suggestion is to check development tools as well, if you dont need the "best" board just, but also the ease of development

The Spartan series and Xilinx ISE are dead now. The new future is Vivado and 7 series FPGAs. I guess Vivado is better in some ways but it's integration with version control software is pathetic.

I like the Avnet MicroZed which is built around a Xilinx Zynq. The Zynq is a latest generation FPGA and dual core ARM on a chip. There's a decent community and all the development tools are free.

... using the "Lucid" HDL, which at first glance seems like a de-crufted Verilog.

That's a good thing, right?

I'll just leave this here: https://github.com/zeusk/CS242

I did this during my sophomore year, didn't care much about implementing a proper memory implementation as that was way beyond the scope of our course.

Every Computer Engineering undergrad is rejoicing somebody just did their project for them.

This is not written in Verilog/VHDL. I doubt they will be able to pass this off as their own work.

Does anyone have any tips on determining if a given CPU core will fit onto a given FPGA? Most of the boards in the hobbyist price range seem kinda ... underwhelming.

You can just download the CAD free version (Quartus II or ISE) and compile it for a target FPGA board.

No discussion of CPU design is complete without trying to run your design as fast as possible. That's how people discovered that pipelining is good, and a flat implementation (such as the one in that code) is slow.

There is a trade-off between performance and area. Pipelined designs are bigger (not least because of all that pipeline registers).

Very good point. This is even more critical in a FPGA as you are working with a fairly constrained resource.

There were a number of people in my first hardware design course that had a great idea right up until they found it didn't fit on the xilinks chips we had in the lab.

From the proposed flat design, adding pipelining would probably add 10-20% more area, and double or triple the speed. Definitely worth exploring (and would indeed make for a great follow-up post).

Only if it's a 3-stage pipeline without a hazard detection. Otherwise the area would at least double. But, yes, I'd also like to see a pipelined core in this new HDL.

Don't forget simplicity - their goal is learning by creating a "super basic CPU" for which you can write "some assembly". Including pipelining in this design would distract from the main purpose.

It would make a nice topic for a follow-up tutorial though.

Reply on news.ycombinator.com