GOWIN FPGA

Why do you say they are not available? Have you tried talking to a distributor?

I have two who have quoted me delivery on the -9 part in the QFP100 package. I'm pretty sure other packages are also very available.

Oh, I should mention that the documentation is still a work in progress. They have some real crap but at least they update it fairly often.

No major distributor has them: Mouser, DigiKey, Edge. They either have never heard of the chips (they only have an eval board) or label them as "Discontinued". I need 15 pieces of the 3.3V GW1N in QFN48 for a prototype run. Not even close to the production stage to consider a reel, too many to request samples.

What distributor should I talk to?

Best regards, Piotr

The Gowin web site lists Edge and Rutronik. Both were very happy to provide good prices ~3-4$. Edge has a part number search but it is not very effective. Best to contact someone and put in a sample order. They will sell you 15. Here is an email contact at rutronik, Neal.Enzenauer at rutronik.com. Just let him know how many you will buy the rest of the year. Tell him Rick, the guy working on the ventilator sent you.

Gowin wants $80 for their programming cable. It uses the FT2232HL which you can get in generic cables from FTDI (Digikey etc.) at half that price.

Have you test driven the development software yet? It's a bit rough around the edges still, but at least supports VHDL now.

Checked both, no luck with an on-line search. Maybe contacting someone there would help, you're right. Thanks!

No, I am still at the chip selection phase. The requirement is a small-non BGA package with about 30 IOs (which excludes TQFP), which narrows down the choice to QFN48. This basically leaves 3 viable options: Lattice MachOX2 640HC, ICE40UP5K and GW1N. About the same price in low volume, the 640 and the GOWIN are capable of running from a 3.3V directly, which is a big advantage. The FPGA can be SRAM-based; I am not going to use the built-in flash.

Lattice Diamond is a big PITA, too; I got used to the quality of Quartus. And it turns out that 13 years is not enough to implement decent support for VHDL-2008 in LSE, Synplify and Modelsim at the same time. The documentation sucks even more, so I don't know if the Gowin's can be any worse. I would rather use a Xilinx or Altera/Intel part, but they seem not to care about the low-end applications anymore, where the only fast part required is a differential Manchester multidrop LVDS link. All the rest is a boring CIC decimator, an I2C master and so on. It is basically MCU-like stuff, but the only FRAM-based MCUs are the MSP430s, which are way too slow for a 50Mbps low-latency link. This is the only reason for an FPGA here.

Best regards, Piotr

Mouser signed a disti agreement with Gowin in December:

-Jim

TQFP is excluded because of the physical dimension? Just too big?

What about the iCE40 Ultra and UltraPlus which come in the same 48 pin QFN?

Lattice has a very useful selection guide.

I don't know why people rag on the vendor tool. I don't spend much time with it. I do most of my work in the editor and simulator. Lattice used to provide Active HDL from Aldec, but now it's Modelsim which I understand is a bit more clumsy for some things. I was working with someone who said he couldn't view the state of variables until he was running the code for that unit. AHDL lets you throw the up on the waveform display as soon as you start the simulation.

With Gowin you won't care so much about Synplify I think. Gowin has it's own synthesis tool, but I suppose you can buy the Synopsis tool. For simulation you are on your own. I'm working in the AHDL with the Lattice tools.

It's funny sometimes. There's someone who doesn't like every tool. Lots of folks rag on Xilinx and even Altera tools.

They are, believe me. Trying to use their math blocks I finally had to reverse engineer the simulation code to see how it works. Rather complex too with all the generics for the variations. Eventually I was able to distill it down to something I could understand.

Yep, they go after the real money. X and A have no interest in becoming a Microchip type company selling a bazillion different small parts with low margin. I think there is a market for that. FPGAs are fast enough now that many apps can be implemented on the slowest chips without even thinking about it. The high end will always create demand, but that doesn't mean there isn't also significant demand at the low end. I've read a lot about Anlogic as well, but they don't seem to support English as well as Gowin does. Some of the Anlogic data sheets are only in Chinese.

Keep in touch. If you get to using the programming tools before we do I'd like to hear your results. We are skipping the cable thing and are putting the FTDI chip on our prototype boards. Then we don't have to worry with who has programming cables and who doesn't. The people on this project are scattered around the globe.

They were listed on Mouser for a bit, but Gowin was put on the US CCMC list (Communist Chinese Military Company) which creates restrictions in funding. They can become further restricted so that you can't ship your product to various locations, but not yet. I expect this was an issue for a global disti.

The GW1N-1-UV is roughly equivalent to the 1200HC, but for some unclear reason, there is no 1200 in QFN48. It is available in QFN32, but then there are only 21 IO lines, which are not sufficient in this application. This way, Lattice forces me to use the 640HC in QFN48, which is quite small and does not allow me to use the LatticeMicro8 soft CPU, as there are only 2 EBRs, for which I already have found a better application.

I cannot really compare the documentation quality, but that for the Lattice parts is appealing. For example, after two days of scratching my head thinking about using the I2C hard cores, I decided to roll my own in VHDL, which took me 6 hours. The PFU documentation also lacks details, e.g. I still don't know if I can access the intermediate carry chain or not. The carries in a ripple counter make perfect clock enables/triggers -- but is the tool wasting some resources to emulate the "true" carries for me? No idea. It makes a difference if you have got 256 or even 640 LUTs.

Best regards, Piotr

Yes, precisely. The BGA/CSP would make the PCB too expesive -- and I am BGA-phobic anyway...

These would certainly do, but I have found the QFN48 variant a bit too late. They require an external regulator for V_core, BTW. :(

Mouser has even better. ;)

I thought that the Lattice docs mark the bottom. Not sure if I am willing to pursue the Gowin direction then, time will tell.

Sure. For some time I will be playing with the MachXO2 series, as I have got an eval board.

Best regards, Piotr

When I have questions like that I take the section of code and synthesize it, then look at the result in the RTL schematic viewer. Even Gowin has one of those.

Once I was getting two carry chains when I expected one only to find out I had a very small discrepancy in my coding that made them distinct. Fixed that and reduced the LUT count by dozens from the many places I was using that method.

I used to be that way. Recently someone on eevblog convinced me to consider them as the cost difference is gone for the 1 mm and even 0.8 mm ball spacing parts. LCSC sells the Spartan 6 at incredible prices and the BGA256 may be too big for you, but it fits in the 100 pin QFP my board uses. I would probably only use half the IOs or less, maybe a third. Xilinx has the Spartan 7 in a 196 pin BGA with 1 mm ball pitch if I remember correctly, so smaller than the 256 a bit, still not 48 QFN size though. I've only found Mouser pricing which is not realistic if you build more than a couple thousand.

If I have to spin my current board because of the Lattice part price increasing, I can't use the Gowin part because of being on the CCMC list. So the Xilinx part looks very attractive. I'd need to find a way to get a pricing quote they will honor for years. One really good thing about Gowin is the Edge web prices are not much different from the quotes.

Is the Vcore really that big a deal? It's $0.25 part, no?

I've been down this road and other than the Chinese startups Lattice is the only vendor supporting the low end. So their selection guide has red boxes around the "good" parts to make them easy to find.

All the other small fish and even the big fish in the FPGA pool have been bought up. I wonder how much longer Lattice will swim free?

Real estate is. That would be another part with non-trivial cooling requirements. The QFN48 has a huge thermal pad, which would help me a lot.

I have absolutely no problem with Chinese parts, no national pride hurt here. ;-)

Best regards, Piotr

I've never designed an FPGA where I had to give thought to the power dissipation other than to estimate so I could say it was estimated. Are you really going to be cranking on it so it needs real current? Dropping 2 volts even at 100 mA is not much more than an 0603 resistor can handle. What's a reasonable dissipation for an SC-70 regulator? I've never considered how much heat an SC-70 can get rid of. You did say you could consider a 640 LUT device, right? I would point out there is a reason why they call them iCE40 devices. They really are low current.

The key word there was "start up". Gowin seems to be doing ok, at least until they got on the CCMC list. Don't know how that will impact them. The others seem to be focusing on markets where they are not so visible. Their docs are a lot less English too, not to mention support, tools, etc.

There's one that is a bit different that *is* with either Mouser or Digikey, can't remember which and I can't remember their name. I think they might have a larger QFN, but maybe not because every time I consider them they drop off my list. They do the is it logic or is it routing thing. So you don't get to use all of the LUTs. In theory it means lower price since you aren't paying for as much routing that you won't use. I remember years ago the FPGA FAE chant was, "We sell you the routing and throw in the logic for free". I don't see this in this company's prices, but they are not real prices as we all know. You have to get a quote to know what they will really sell their chips for.

Here is an example of the poor quality of the Gowin docs. In fact, this one is pretty extreme. UG166 lists the pin outs of different family members in each package. This document shows a consistent match between the designators TBLR as in IOT21A being bank 0, 1, 2 and 3 (not necessarily in that order). But in UG114 the they list IOT I/Os as being part of banks 0, 1 and 3! If you are not running all I/Os at the same voltage this would be a big issue!!!

We just caught this and I've sent an email to the FAE. He usually gets back within 24 hours.

OK, I am abandoning the Gowin direction. iCE40UP5K wins hands down and the volume is not likely to be big enough to pay for the NRE. If I am wrong here, Lattice will probably make me a special offer... ;)

Best regards, Piotr