Need some advice and opinion about two FPGA dev boards

[bte]

Hi all,

I would like the opinions of esteemed forum members about "virtual career" advice in general and in particular, two FPGA dev boards available for sale here. I need to first provide a background to this plea so that it will not be treated like a generic "what is the best xyz" question. I apologize in advance for the lengthy text.

I am an EE with BS and MS degrees but I have never done EE work professionally. Towards the end of my masters studies, I found a job as a software engineer and there was no return back. During the covid time-frame, I was deemed as no longer useful for doing commercial software work (I was 45+ at the time) so my career has come to an end.

This is when I thought about going back to my roots. However, it didn't take for me to figure out that what I learnt in school 30 years ago regarding "how EE is done" is no longer there. In addition, I also noticed the fact that "these new ways of doing commercial EE" were also limited to youngsters or already experienced/working EEs, at least where I live. A friend of mine living in the US (who is also an EE yet did software work for his whole life) also told me that nowadays EE work is done by large groups, since so much is involved in coming up with a viable product and there's not much I can do by myself and make a living out of it. And due to some family events that happened around that time frame with consequences up until right now and in the foreseeable future, I also had to give up the idea of "being an EE". For the moment, let's just say that I do neither on a commercial basis.

Nevertheless, I can't just stop my brain thinking about the prospect. I feel as if all those years studying EE will be a waste if I don't do something about it and this drives me nuts. I still love learning new things. I don't mind spending the time to learn stuff, even though my available time is quite fragmented. When I started the university, I thought I wanted to be an analog engineer but there weren't enough encouragement or the courses to make it come to reality (not to mention that the primary "analog" professor went to US to work in industry during my junior year). The curriculum and the interests of the professors in the department were mostly slanted towards DSP. After pondering about and reading upon reading, I came to the conclusion that that sub-specialty that fits my brain and that I would like to do at this point in my life is digital design.

During my school time, FPGAs or related courses weren't a thing (we had of course a digital design course but it only involved discrete chips --- I checked the latest curriculum and see many updates and upgrades have been done, including digital design). So it has been a lot of reading and somewhat of experimenting but it is going good I think. I am hoping to get rid of the "EE study was useless for me" feeling at some point.

OK, it's been too long of an intro; let's come to the topic. I have been playing with a small FPGA board until now. For my purposes, I don't see the need to go with a bigger board of the same kind since I think "I got the ropes" one way or the other regarding the concept of FPGA. Now I am wondering about considering some variety: I have seen boards with chips that are FPGA+CPU combined, like Xilinx Zynq. Would it be an uphill battle for me to learn about this part of the technology (making CPU+FPGA work together)? Will it end with frustration? I don't mind reading datasheets and I don't mind trying to find solutions to problems but if proper knowledge in this field does kind of "require club membership" then it would be very discouraging in the short term and frustrating in the long term.

I also want to learn about things that would be relevant some time into the future. Yes, I did give up the idea of being a real EE but still, I would like to play along as if one day I would become one (it's all in the brain). Therefore I don't want to spend time on gaining knowledge that is considered obsolete already.

I saw two boards available in the local "second-hand site" (think ebay or craigs list maybe). One is a brand new Digilent ZedBoard for about $488 (the official distributors sell it for $785 including VAT; they also require that I fill a form to be submitted to Digilent for approval). The other one is a Digilent Genesys (legacy) board for about $215. I checked this one and it is the older version with Virtex-5 LX50T. According to wikipedia (https://en.wikipedia.org/wiki/Virtex_(FPGA)#SoC), this family of FPGAs come with IBM PowerPC cores.

Would any of these boards provide for what I want to do? Would the legacy board be useful as well or is it considered old technology to the extent that whatever knowledge I gain will be of no value going forward? Is what I have written so far incoherent drivel? Should I find some other (EE) thing to keep my brain working, my conscious subdued and my sanity intact?

If you did read so far, thank you for your time and attention. Any feedback is more than welcome.

PS: If this post would be more appropriate in another section, let me know and I will move it (or ask a mod to move it).

[Someone]

OK, it's been too long of an intro; let's come to the topic. I have been playing with a small FPGA board until now. For my purposes, I don't see the need to go with a bigger board of the same kind since I think "I got the ropes" one way or the other regarding the concept of FPGA. Now I am wondering about considering some variety: I have seen boards with chips that are FPGA+CPU combined, like Xilinx Zynq. Would it be an uphill battle for me to learn about this part of the technology (making CPU+FPGA work together)? Will it end with frustration? I don't mind reading datasheets and I don't mind trying to find solutions to problems but if proper knowledge in this field does kind of "require club membership" then it would be very discouraging in the short term and frustrating in the long term.

FPGA methodologies are pretty slow/consistent but very few standard libraries in comparison to software. If you already have a software background consider being the person who can bridge the connection between software and FPGA, which for many teams today means being a linux kernel developer. Then you can learn more FPGA on the job and start in something closer to your prior experience.

[gerbay]

If you already have a software background consider being the person who can bridge the connection between software and FPGA, which for many teams today means being a linux kernel developer.

I don't understand the connection between being a Linux kernel developer and FPGA (digital design). I consider them separate fields of expertise.

[zapta]

I would start with an idea of a project you want to accomplish and then pick whatever fits that goal. Having a clear goal and going through all the grind of making it to actually work is the best way to really learn IMO.

[Someone]

If you already have a software background consider being the person who can bridge the connection between software and FPGA, which for many teams today means being a linux kernel developer.

I don't understand the connection between being a Linux kernel developer and FPGA (digital design). I consider them separate fields of expertise.

They have been separate fields but there is significant overlap which needs people with understanding of both:

I have seen boards with chips that are FPGA+CPU combined, like Xilinx Zynq. Would it be an uphill battle for me to learn about this part of the technology (making CPU+FPGA work together)?

I’d estimate most of the hard cores in those FPGA categories are running Linux. If you have a software background that is one way to approach FPGAs (at the boundary with software) without having to understand/learn about the larger FPGA field first.

[pcprogrammer]

The combined CPU + FPGA fabric chips like the Xilinx zynq are not that different than having them separate. The basic difference is the connection between the two being internal in the chip.

For me it is also on the list of things to play with, but I have looked at the datasheets and reference manuals of the zynq parts, where they talk about domains. There is the CPU domain and the FPGA domain. For what I understand from it, it is possible to run the CPU part without the FPGA part, so you can gradually work your way up.

Take a look on Aliexpress, they sell cheap boards that do just as well as the expensive Digilent boards. https://www.aliexpress.com/item/1005005919123170.html I have a couple of these and they work just fine, for as far as I tested them.

[xvr]

You shouldn't start by choosing an FPGA, but by choosing the project you want to implement. The Zync series (like any CPU+FPGA combination) is designed for fairly complex tasks. If you're starting to learn FPGAs from scratch, it's wiser to choose a simple project—something that fits the Spartan series (Spartan 7 is recommended, but Spartan 6 is also suitable). Many FPGAs in this category are also available from other manufacturers (Sipeed's Tango series of boards based on Gowin's FPGA, and Latice's iCE40 FPGA).
Virtex 5 is strongly not recommended - it is a very outdated model.

[paulca]

If you already have a software background consider being the person who can bridge the connection between software and FPGA, which for many teams today means being a linux kernel developer.

I don't understand the connection between being a Linux kernel developer and FPGA (digital design). I consider them separate fields of expertise.

They have been separate fields but there is significant overlap which needs people with understanding of both:

I have seen boards with chips that are FPGA+CPU combined, like Xilinx Zynq. Would it be an uphill battle for me to learn about this part of the technology (making CPU+FPGA work together)?

I’d estimate most of the hard cores in those FPGA categories are running Linux. If you have a software background that is one way to approach FPGAs (at the boundary with software) without having to understand/learn about the larger FPGA field first.

I think Linus Torvalds is adamant he is NOT a hardware guy.  On LTT he showed an electronics project he did though.

However.  I bet he knows how to read and then trust a datasheet and write (or code review) a driver for some hardware.

[pcprogrammer]

You shouldn't start by choosing an FPGA, but by choosing the project you want to implement. The Zync series (like any CPU+FPGA combination) is designed for fairly complex tasks. If you're starting to learn FPGAs from scratch, it's wiser to choose a simple project—something that fits the Spartan series (Spartan 7 is recommended, but Spartan 6 is also suitable). Many FPGAs in this category are also available from other manufacturers (Sipeed's Tango series of boards based on Gowin's FPGA, and Latice's iCE40 FPGA).
Virtex 5 is strongly not recommended - it is a very outdated model.

If you had read more of the OP's story, you would have seen he is not a beginner.

I have been playing with a small FPGA board until now. For my purposes, I don't see the need to go with a bigger board of the same kind since I think "I got the ropes" one way or the other regarding the concept of FPGA.

He wants to experiment with something new and more advanced.

There are other FPGA's with smaller MCU's embedded though, and might be simpler to get to grips with. Gowin has some and with some searching one might even find something on a ready made dev board from Sipeed.

[Someone]

However.  I bet he knows how to read and then trust a datasheet and write (or code review) a driver for some hardware.

That is the problem, most FPGA “hardware” is some product/company specific design and doesnt have a datasheet or a mainline driver. But that hardware does need some sort of driver that can be deployed so that most of the programming can remain in userspace.

Everything remains internal to the company, and isnt reviewed by OS/kernel developers/providers.

[paulca]

FPGA'ing is just collapsing digital hardware into a mutable IC package.  Often as a "step" towards an ASIC fab of an actual die.

It's hard to do anything but LED chasers and remain engaged without a driving purpose.

The best way to write garbage Verilog is to not have a 100% clear picture of what it does FIRST.  Unlike software were we can abstract and decouple components easily, FPGAs tend to be extremely tightly coupled.  It's where abstraction hits the end of the road.

The shift has been moving "hardware" into "software" as the later is mutable.  However when this hits the hard metal of a PCB the architecture shifts to "what runs the software", "what hardware acts for it", "what hardware support it", "how do they all talk nicely?"

That "glue" or "instanteneous" logic (from the timeframe of the CPU) hardware for buses, MMUs, IOMMUs, hardware expansion signals, clocks, counters, timers, watchdogs, interrupt chains ... are all done in hardware, under software control.

The future looks, at the moment, to be diversifying on the CPU core level.  Moving from "basic sequential CPU" architectures to highly parallel compute architectures.  From "x86/arm" to CUDA and Tensor cores. 

While we may not like it, but "AI Agents" are a thing and yes, they want to make them smaller and smaller and they also want to offload some of the AI processing locally too.  Current MCUs can't do this, they can do the CPU stuff.  A new bread of hybrid board is emerging, MCUs with tensor and AI cores on them. 

If you are looking for a "novel" "bleeding edge" field where FPGAs will play a huge part in prototyping, I'd take a deep dive down that route.  The "back channel" is that some things which start in software, actually end up being moved back to hardware.  That is the strategy that ultimately led us to things like CUDA cores doing nothing buy parallel floating point pipelines.  Those in software were excruciatingly slow.

[bte]

A lot of insightful comments, thank you all very much. I really appreciate everyone providing guidance. Now that I had my midnight coffee, let me respond to the best of my abilities.

I would start with an idea of a project you want to accomplish and then pick whatever fits that goal. Having a clear goal and going through all the grind of making it to actually work is the best way to really learn IMO.

You shouldn't start by choosing an FPGA, but by choosing the project you want to implement. The Zync series (like any CPU+FPGA combination) is designed for fairly complex tasks. If you're starting to learn FPGAs from scratch, it's wiser to choose a simple project—something that fits the Spartan series (Spartan 7 is recommended, but Spartan 6 is also suitable). Many FPGAs in this category are also available from other manufacturers (Sipeed's Tango series of boards based on Gowin's FPGA, and Latice's iCE40 FPGA).
Virtex 5 is strongly not recommended - it is a very outdated model.

FPGA'ing is just collapsing digital hardware into a mutable IC package.  Often as a "step" towards an ASIC fab of an actual die.

It's hard to do anything but LED chasers and remain engaged without a driving purpose.

The best way to write garbage Verilog is to not have a 100% clear picture of what it does FIRST.  Unlike software were we can abstract and decouple components easily, FPGAs tend to be extremely tightly coupled.  It's where abstraction hits the end of the road.

I expected this answer and I myself think the same way if the idea was to learn FPGA and be proficient at it to some degree (to be able to get a job or similar). Now that I am into my 50s, that is not a realistic option (at least not where I live). What I am intending to do at this point is to "fill in my blanks", be it during my education or professional life. In a sense, this is a form of "let me not leave this world before learning/doing this, that etc". I still don't want to spend time, effort or money into obsolete tech or gear, hence the plea for guidance. The last item that I worked on my list was FPGA, so the thinking of "where I would/could go from here" is what prompted my question.

As to why I asked about two particular boards: Here Digilent boards above a certain feature level are expensive or not existent (this also applies to other electronic items but let's not digress). We are also required to submit an end user information document and have it approved in order to be able to buy. Therefore when it is possible to find them available for sale as secondhand, especially at discount, I do consider the purchase in case I can make use of it in the future.

Our international purchase limit is limited to 30 EUR including shipping so that limits the choices a lot.

Virtex 5 is strongly not recommended - it is a very outdated model.

Duly noted. I also did a further check on it and the exact chip on the board that I mentioned doesn't have any CPU cores so considering what I am doing/trying to do, it would not be much different than another FPGA-only board.

Many FPGAs in this category are also available from other manufacturers (Sipeed's Tango series of boards based on Gowin's FPGA, and Latice's iCE40 FPGA).

I do have a board with Artix-7 (XC7A35T) so I think I am good on FPGA front for the moment.

The suggestions from @pcprogrammer seems to fit the bill from both a financial and my initial intention standpoint so that direction is where I intend to venture for the moment.

[bte]

The combined CPU + FPGA fabric chips like the Xilinx zynq are not that different than having them separate. The basic difference is the connection between the two being internal in the chip.

For me it is also on the list of things to play with, but I have looked at the datasheets and reference manuals of the zynq parts, where they talk about domains. There is the CPU domain and the FPGA domain. For what I understand from it, it is possible to run the CPU part without the FPGA part, so you can gradually work your way up.

Take a look on Aliexpress, they sell cheap boards that do just as well as the expensive Digilent boards. https://www.aliexpress.com/item/1005005919123170.html I have a couple of these and they work just fine, for as far as I tested them.

Thank you so much for that link and the commentary! The price of 20 USD fits within our import limits and that card should provide enough to get started. I checked the links and had some of the text translated into English. That text says

The EBAZ4205 is the control board for the EBIT BTC mining rig, and it's the cheapest mining board available. Other boards include the EBAZ4203. The EBAZ4205 features a Xilinx ZYNQ7Z010 SOC, 256MB DDR3 memory, a 128MB NAND flash memory, and a 100M Ethernet PHY. The schematic and PCB were obtained from a secondhand dealer. By default, the board requires power from a 2.0mm PHD header at 5V-12V. Linux boots from the NAND flash memory and executes the mining control program. The board has an SD card slot; you can solder the SD card slot yourself and change the resistors to allow ZYNQ to boot from the SD card. The board is inexpensive and suitable for learning ZYNQ. However, it's regrettable that the ZYNQ's USB pins are not brought out.

I don't know how much of an inconvenience that last part would be but I am willing to give it a go.

There are other FPGA's with smaller MCU's embedded though, and might be simpler to get to grips with. Gowin has some and with some searching one might even find something on a ready made dev board from Sipeed.

Comparison on this page says that Tang Nano 4k has Cortex-M3 and camera interface. Upon finding this, I did some search and just found this board:

https://tr.aliexpress.com/item/1005004562311534.html

Looks interesting, definitely worth some attention.

[bte]

I have seen boards with chips that are FPGA+CPU combined, like Xilinx Zynq. Would it be an uphill battle for me to learn about this part of the technology (making CPU+FPGA work together)? Will it end with frustration? I don't mind reading datasheets and I don't mind trying to find solutions to problems but if proper knowledge in this field does kind of "require club membership" then it would be very discouraging in the short term and frustrating in the long term.

FPGA methodologies are pretty slow/consistent but very few standard libraries in comparison to software. If you already have a software background consider being the person who can bridge the connection between software and FPGA, which for many teams today means being a linux kernel developer. Then you can learn more FPGA on the job and start in something closer to your prior experience.

My work software experience is limited to user-mode. Before that, during middle and high school, it was 6510 assembly (C64) but it was only dabbling here and there, hacking games etc. type of work. During university timeframe, it was linux and C/C++ (1). At my first job (that lasted for more than 10 years), it was strictly Windows and C++/C# with regular app and user-mode service coding (2). I still did play with linux but it never turned into a development effort, user or kernel-mode.

In my initial post, I had mentioned noticing that what I learnt as "EE" during school time applied no more. Sadly, I also noticed during my "involuntary pondering time period" that it applies to the software discipline as well. Being 50+, I have no more hopes of finding a regular job here as EE or software person, much less as a kernel developer. But I do appreciate the guidance, since that is what I have been primarily asking for.


(1) My MS advisor was part of a EU project and his image processing team was part of this. His PhD student and MS students completed their work and left by the time I started to work on my thesis. As part of this project, one of our duties was to do the software coordination but there was nobody else to do the work. So my MS thesis became primarily a software development and management project that involves dealing with all the international teams. I assume this work had quite some part in landing my first job at a big corporation.

(2) That C++ was unfortunately in the form of "C with classes" since anything else was deemed as "unnecessary invention" by the department policy (i.e. most other developers couldn't grok such code). At times I felt like a government employee just doing a job but I couldn't quit (the job was in the USA and my H1B would expire if I quit the job). There was no personal development involved. When I mentioned this to my lead (and my desire to work with people more skilled than I am so that I could actually gain experience), he suggested that I could get a mentor within the company. Mentoring wasn't what I needed but continued working while keeping some hope that maybe I could find a chance to work with an architect or similar experienced person. After a couple years later, a merger with another group happened and the other group's manager became the manager for the combined group. He was hostile towards us and turned the group into hell. I just couldn't take it anymore and left after 10+ years, returning back home (it was interesting to learn later on that three other people from my original group quit at the same time).

[pcprogrammer]

About the cheap zynq board and not having the USB interface available being a problem depends entirely on what one wants to do with the board. Using it as a special Linux computer with a keyboard and a mouse won't be much fun as it will be to slow and graphics would be kind a crap. Would also need the conversion to a VGA connector on the "PMOD" connectors. They do sell additional boards for this though.

Using the FPGA part from the CPU part with Linux running might need kernel level drivers to be more real time, but I found that it is not that difficult, and a lot of information can be found on the net on how to do this.

Another option is to go bare metal and start from the reset vector. This is also the way to go with the Tang Nano 4K. But from what you wrote so far I get the feeling that you have no experience with embedded programming and microcontrollers. Maybe another option on the learning path you want to take.

Start with some Arduino project and work your way up to bare metal ARM-Cortex Mx programming. There are loads of interesting development boards for cheap to be found, like for instance the on a STM32F411 based blackpill. https://www.aliexpress.com/item/1005006364250911.html This is a beefier version of the bluepill which is based on the old STM32F103.

After gaining some experience with this, move up to the FPGA combined with CPU option.

[iMo]

As mentioned above, do not buy $$ board without having a clear idea what kind of a project you want to follow (as usually the board you have already got is not the best for a new project)..
Also the starting with fpgas is rather slowish process, different from usual programing, where you have to learn a lot of new stuff (best is the training on the job). If starting today do buy a board for those $20 which is suitable for an existing project, and start from there. Implement the project and start to enhance it with new functionality, that is the best method of learning, imho..

[zapta]

(Deleted. Posted to the wrong thread)

[Gleeson]

If it's any help I have two ARTIX-7 boards (ALINX and ARTY) that I will never use and would be willing to send them to you at no cost.

[bte]

About the cheap zynq board and not having the USB interface available being a problem depends entirely on what one wants to do with the board. Using it as a special Linux computer with a keyboard and a mouse won't be much fun as it will be to slow and graphics would be kind a crap. Would also need the conversion to a VGA connector on the "PMOD" connectors. They do sell additional boards for this though.

Using the FPGA part from the CPU part with Linux running might need kernel level drivers to be more real time, but I found that it is not that difficult, and a lot of information can be found on the net on how to do this.

Another option is to go bare metal and start from the reset vector. This is also the way to go with the Tang Nano 4K. But from what you wrote so far I get the feeling that you have no experience with embedded programming and microcontrollers. Maybe another option on the learning path you want to take.

Start with some Arduino project and work your way up to bare metal ARM-Cortex Mx programming. There are loads of interesting development boards for cheap to be found, like for instance the on a STM32F411 based blackpill. https://www.aliexpress.com/item/1005006364250911.html This is a beefier version of the bluepill which is based on the old STM32F103.

After gaining some experience with this, move up to the FPGA combined with CPU option.

Once again, thank you for your continued assistance. Your feeling is correct, I didn't work with micro-controllers in the bare-metal sense. I do have Arduino boards and many shields that I did projects with using the provided IDE. When you say "work your way up to bare metal ARM-Cortex Mx programming", do you mean using the Arduino or some other board? Also, does the "x" matter (i.e. should I choose the "x" carefully or would anything be good at this stage)?

I checked the STM32F411-based board but AliExpress said it was not available to me (I hate the new "unavailable" policy where the item is not shown at all rather than showing the item and saying it's unavailable in a given region). I was able to see the details through VPN connection. I think I can find that one or something similar here. I also just checked my stash and besides various Arduinos and shields, I found an Atmel SAM V71 evaluation kit and a Cypress CY8CKIT-049-42xx PSoC 4 prototyping kit. Would these be helpful or should I try to find the STM32F411-based board?

That ex-miner-controller Zynq 7010 board still looks attractive for future consideration. Aliexpress items tend to become unavailable to us shortly after being available so I am still thinking about ordering it. It's just $20 (+60% tax) so it won't break the bank. I just read its reviews and saw that some people got damaged or scrap-looking boards that have quite bad rework done. Was yours in OK condition?

[bte]

The shift has been moving "hardware" into "software" as the later is mutable.  However when this hits the hard metal of a PCB the architecture shifts to "what runs the software", "what hardware acts for it", "what hardware support it", "how do they all talk nicely?"

That "glue" or "instanteneous" logic (from the timeframe of the CPU) hardware for buses, MMUs, IOMMUs, hardware expansion signals, clocks, counters, timers, watchdogs, interrupt chains ... are all done in hardware, under software control.

The future looks, at the moment, to be diversifying on the CPU core level.  Moving from "basic sequential CPU" architectures to highly parallel compute architectures.  From "x86/arm" to CUDA and Tensor cores. 

While we may not like it, but "AI Agents" are a thing and yes, they want to make them smaller and smaller and they also want to offload some of the AI processing locally too.  Current MCUs can't do this, they can do the CPU stuff.  A new bread of hybrid board is emerging, MCUs with tensor and AI cores on them. 

If you are looking for a "novel" "bleeding edge" field where FPGAs will play a huge part in prototyping, I'd take a deep dive down that route.  The "back channel" is that some things which start in software, actually end up being moved back to hardware.  That is the strategy that ultimately led us to things like CUDA cores doing nothing buy parallel floating point pipelines.  Those in software were excruciatingly slow.

It looks like I didn't address your reply properly last night. I do apologize for that. I only have peace and quiet in the wee hours of the night and things sometimes fall through the cracks when working at such odd hours.

I am not sure about what you mean when you said "From "x86/arm" to CUDA and Tensor cores." Do you mean that x86/arm will no longer be there in the future and work will be done via CUDA and Tensor cores? I also googled "MCUs with tensor and AI cores" and found some information. I will include a few links here:

https://www.renesas.com/en/about/newsroom/renesas-sets-new-mcu-performance-bar-1-ghz-ra8p1-devices-ai-acceleration
https://www.st.com/content/st_com/en/campaigns/microcontrollers-for-edge-ai-z11.html
https://www.nxp.com/applications/technologies/ai-and-machine-learning:MACHINE-LEARNING

I am assuming your comment regarding "MCUs with tensor and AI cores" was that being proficient in such architectures is what will be valuable in the future (though I am not sure what role FPGA plays in such environment). If I misunderstood, I would appreciate some clarification and guidance.

[bte]

If it's any help I have two ARTIX-7 boards (ALINX and ARTY) that I will never use and would be willing to send them to you at no cost.

Thank you so much for your offer. I do have a small Artix-7 board and that is sufficient for the moment. In case there's someone else who could put them to good use, you could list the boards in the buy/sell section of the forum.

(Note: even if I needed them and you sent them at no cost, our customs don't have the concept of "no cost". They will look up the items and assign a value. If the item value plus shipping is more than 30 EUR, I have to become an importer and use the services of a professional broker in addition to paying customs tax and VAT. If I reject hiring a broker, the item will be confiscated and sold at an auction to companies who then will sell the items in the market. I assume it's obvious that the service charge of these brokers is not trivial so nobody actually buys anything anymore that is valued over 30 EUR including shipping --- that pretty much leaves us with AliExpress and similar sites).

[pcprogrammer]

Once again, thank you for your continued assistance. Your feeling is correct, I didn't work with micro-controllers in the bare-metal sense. I do have Arduino boards and many shields that I did projects with using the provided IDE. When you say "work your way up to bare metal ARM-Cortex Mx programming", do you mean using the Arduino or some other board? Also, does the "x" matter (i.e. should I choose the "x" carefully or would anything be good at this stage)?

Oh, with Arduino experience under your belt you are already somewhat proficient. 

On the ARM-Cortex Mx it does not matter which one. The STM32F411 is M4 while the STM32F103 is M3, and your Atmel SAM V71 is M7. The PSoC 4 is M0.

With these two you are well stocked to undertake some bare metal adventure.

I checked the STM32F411-based board but AliExpress said it was not available to me (I hate the new "unavailable" policy where the item is not shown at all rather than showing the item and saying it's unavailable in a given region). I was able to see the details through VPN connection. I think I can find that one or something similar here. I also just checked my stash and besides various Arduinos and shields, I found an Atmel SAM V71 evaluation kit and a Cypress CY8CKIT-049-42xx PSoC 4 prototyping kit. Would these be helpful or should I try to find the STM32F411-based board?

Yeah this Aliexpress dedicated country policy is not nice for sure, but what can one do about it. 

I just mentioned the STM32F411 as it is a nice MCU that I use myself at the moment. I did not find a lot of listings on Aliexpress, but if you search for "STM32F411 board" you might find some that can be delivered to you. Within the STM32 range of MCU's the used peripherals don't differ that much, which makes it easier to write code for another one out of the range once one already wrote some for another one.

For going bare metal both of your boards are suitable, but a thing with bare metal is that one likes to start at the reset vector which is a bit bother some when a boot loader is involved. Some will state that bare metal is just about using the bare registers of the MCU and the peripherals though, but for me is is about owning the complete MCU, and this not in the material sense.

To put things in perspective with the Arduino environment, my bare metal work begins with setting up the vector table, a reset handler and then the main function, where as the Arduino environment hides all of this from you and gives you the setup() and loop() functions to work with. I'm also void of using standard libraries, which some would mark as foolish, but makes debugging completely about your own code.

In my case, when I build, it results in a binary (or .elf file containing the binary) that needs to be written to the start of the FLASH in the MCU. There are a couple of options to get this done. What I use is a JTAG or SWD programmer to write the binary to the FLASH. I can also debug the code with this tool. An other option is to make use of the build in programming code of the MCU. For this one has to select another boot mode and reset the MCU to start this code. The binary can then be send to the MCU via some interface between the PC and the MCU. Options are USB, UART, I2C, CAN, etc. Whatever the chosen MCU supports.

With Arduino the code is loaded via the bootloader which means that it has to be build to run from another location in the memory, so the linker setup has to be different.

Another difference between Arduino and what I do, is that in my case the MCU directly runs my code after startup or a reset. In the (standard) Arduino case the bootloader starts first and checks if there is a host connection that wants to load new code, and if not then your code is started. So there is a small delay.

There are many possible work flows though and each will have it's own preference on how to do things.

For your boards you need to check the documentation as to what is possible.

That ex-miner-controller Zynq 7010 board still looks attractive for future consideration. Aliexpress items tend to become unavailable to us shortly after being available so I am still thinking about ordering it. It's just $20 (+60% tax) so it won't break the bank. I just read its reviews and saw that some people got damaged or scrap-looking boards that have quite bad rework done. Was yours in OK condition?

I bought a couple of them in 2020, but the links on Aliexpress result in a 404 page. I'm a bit of a hoarder on these kind of items. I see something like this and think "oh that is nice, can use that for some project" and buy a couple to have stock. There are still several listings of the same board though and searching for them might even result in a better deal for you. Searching for "zynq development board" results in a large list with all sorts of boards in different price ranges.

My boards arrived in good order, but when something arrives that is defective one can always start a refund dispute to get your money back. Provide photos of the defective item and good motivation on what is wrong. You might have to send it back, which probably is free of charge, and in the end get your money back. (Your experience might differ due to other local regulations though)

I have to say that for you in your country it seems quite strict and tax on these things to be super high. 60% feels like legalized theft. Here in France (I'm Dutch but live in France) it is "only" 20%, and already included in the prices I see on Aliexpress.

[laugensalm]

I think I can very much anticipate your motivation of "going back to the roots", as I've probably taken a similar path via SW development.
I used to purchase plenty of FPGA hardware on special occasions to learn about the (back then) advanced technology, as I had to bridge a gap between PLD development from the 90s and advanced LUT/FF mapping of an FPGA. Bottomline: Most of the hardware I wouldn't really use or deploy in depth.
In deployment scenarios, either a very new chip had to be somewhat reverse-engineered (debugged) or I would rely on older technology still able to achieve the goal. Second is probably your best option (old pots to learn cooking) *if* you insist on having something to touch.
The other option is going fully virtual and stick everything into a simulated model. This is IMHO the most productive way to move from an idea to a functional implementation, also helping to keep focus on the actual goal, without having to mess with maintenance of complex toolchains in the first place. When I read 'ZynQ' in the beginner's context, I hear "no no no no". You'll want to have something that you can simulate down to the core without the complexity of an embedded Linux or an AXI bus. For instance, you could grab a virtual STM32 setup as for instance from a default Renode (CPU simulator) configuration and develop a virtual peripheral for it using Verilator. Then you could dive into various existing yosys setups that demonstrate how your Verilog/VHDL is turned into hardware elements. That's a learning curve, but with IMHO most effect.

[paulca]

I am assuming your comment regarding "MCUs with tensor and AI cores" was that being proficient in such architectures is what will be valuable in the future (though I am not sure what role FPGA plays in such environment). If I misunderstood, I would appreciate some clarification and guidance.

That's the theory.  Architectures with AI (or AI accelerating) co-processors are "potentially" going to become more common.  "Edge AI" is the term.  Not necessarily LLM though.

The FPGA fits in when you want to use an MCU or rather an MPU and a co-processor and need to create the glue, pipelines, buffers in hardware.

Of course, it may also be that MCU and MPU manufacturers just start releasing dies with both types of core.

[bte]

Oh, with Arduino experience under your belt you are already somewhat proficient. 

On the ARM-Cortex Mx it does not matter which one. The STM32F411 is M4 while the STM32F103 is M3, and your Atmel SAM V71 is M7. The PSoC 4 is M0.

With these two you are well stocked to undertake some bare metal adventure.

Thank you for the confirmation. Now I can dust them off and get going...

Yeah this Aliexpress dedicated country policy is not nice for sure, but what can one do about it. 

The country-specific items policy is ugly but I meant the despicable new policy where now one even doesn't get to see what the clicked item was. It now shows just a whole blank page with a "not available in your area" message. AliExpress used to show the item and all the details but included a note saying "not available".

I just mentioned the STM32F411 as it is a nice MCU that I use myself at the moment. I did not find a lot of listings on Aliexpress, but if you search for "STM32F411 board" you might find some that can be delivered to you. Within the STM32 range of MCU's the used peripherals don't differ that much, which makes it easier to write code for another one out of the range once one already wrote some for another one.

I found that STM32F411 board locally for $8 so no problem there. I also found fancier boards with more included devices etc so in case a more complex board becomes necessary, I don't need to resort to AliExpress' less-than-stellar search facility. Google sometimes helps but not always. Here we have the official distributors for Farnell, Digikey and RS Delivers so common parts of these companies' inventories are available most of the time.

For going bare metal both of your boards are suitable, but a thing with bare metal is that one likes to start at the reset vector which is a bit bother some when a boot loader is involved. Some will state that bare metal is just about using the bare registers of the MCU and the peripherals though, but for me is is about owning the complete MCU, and this not in the material sense.

Yes, that one was the first thing I noticed when I did a cursory search for Arduino bare metal programming. I already dabbled with the toolset when I had to reprogram the Atmega 328 on my "transistor tester" with the new firmware and do some mods at the same time.

To put things in perspective with the Arduino environment, my bare metal work begins with setting up the vector table, a reset handler and then the main function, where as the Arduino environment hides all of this from you and gives you the setup() and loop() functions to work with. I'm also void of using standard libraries, which some would mark as foolish, but makes debugging completely about your own code.

In my case, when I build, it results in a binary (or .elf file containing the binary) that needs to be written to the start of the FLASH in the MCU. There are a couple of options to get this done. What I use is a JTAG or SWD programmer to write the binary to the FLASH. I can also debug the code with this tool. An other option is to make use of the build in programming code of the MCU. For this one has to select another boot mode and reset the MCU to start this code. The binary can then be send to the MCU via some interface between the PC and the MCU. Options are USB, UART, I2C, CAN, etc. Whatever the chosen MCU supports.

With Arduino the code is loaded via the bootloader which means that it has to be build to run from another location in the memory, so the linker setup has to be different.

Another difference between Arduino and what I do, is that in my case the MCU directly runs my code after startup or a reset. In the (standard) Arduino case the bootloader starts first and checks if there is a host connection that wants to load new code, and if not then your code is started. So there is a small delay.

There are many possible work flows though and each will have it's own preference on how to do things.

For your boards you need to check the documentation as to what is possible.

So that's where I will be headed to, then. Let me try to find some info about specifically the sort of bare metal you just mentioned. Vector tables, reset handlers don't scare me (at least conceptually) since I dealt with them during my C64 times (yes, long time ago, but I am assuming it's like riding a bike --- something you never forget once you learn).

I bought a couple of them in 2020, but the links on Aliexpress result in a 404 page. I'm a bit of a hoarder on these kind of items. I see something like this and think "oh that is nice, can use that for some project" and buy a couple to have stock. There are still several listings of the same board though and searching for them might even result in a better deal for you. Searching for "zynq development board" results in a large list with all sorts of boards in different price ranges.

My boards arrived in good order, but when something arrives that is defective one can always start a refund dispute to get your money back. Provide photos of the defective item and good motivation on what is wrong. You might have to send it back, which probably is free of charge, and in the end get your money back. (Your experience might differ due to other local regulations though)

This link works for me:

https://www.aliexpress.com/item/1005005919123170.html

The English wiki for that board (https://github.com/xjtuecho/EBAZ4205/wiki) suggests buying a few since these were stored in less than ideal conditions and one doesn't know the condition beforehand. Sending back from here is a hassle so for the moment I decided not to buy this card. As a specimen of FPGA with CPU cores, I placed an order for the Tang 4k board (I couldn't find this locally).

https://www.aliexpress.com/item/1005004562311534.html (I chose the "with camera" version).

I have to say that for you in your country it seems quite strict and tax on these things to be super high. 60% feels like legalized theft. Here in France (I'm Dutch but live in France) it is "only" 20%, and already included in the prices I see on Aliexpress.

Ha ha. 60% would not be a problem for most people if it weren't for the 30 EUR personal limit (recently, they made a change so any item that comes with "free shipping" gets added 3 EUR to its customs value because well, shipping can't be free, right?). We have 40% or 50% "special consumption tax" on top of 20% VAT for cell phones (1). For cars, that "special consumption tax" is one of 80%, 150% or 220% (plus of course the 20% VAT) depending on price, engine size etc.

The whole process can be summarized as the picture in this link:

https://www.amazon.com/Dont-Need-Government-Fucks-Every/dp/B093VRGVYF

According to the story I read at a local site (many years ago, I must add), this was uttered by a Croatian cameraman who got fired afterwards.

Anyways, digressed too much. Thank you once again for all the help and guidance. I will make another post if I stop making any progress.


(1) no, you can't buy it from outside; it will be confiscated --- bringing the phone with you doesn't work as well since the whole system is IMEI-controlled. There's only one provision from buying outside: If you go out of the country, you can bring one phone every three years. That phone needs to be registered to the system for a paltry sum of USD$1000. For the year 2026, the registration price has been increased to $1250.