Working with FPGA

[jeroen74]

I think its main use is ASIC prototyping. Or very high end low volume stuff like MRI scanners and the like.

[alm]

I think Dave mentioned using some of these expensive FPGAs in military products: not enough volume for ASICs, but too complex for smaller FPGAs. Paying $50k per chip might still be cheaper than developing a custom ASIC if you will only sell 10-100 of them.

[Stiege]

I think Dave mentioned using some of these expensive FPGAs in military products: not enough volume for ASICs, but too complex for smaller FPGAs. Paying $50k per chip might still be cheaper than developing a custom ASIC if you will only sell 10-100 of them.

Mmmk. The PhD students here sometimes design their own ASICs and have 10 or so manufactured. Apparently the cost is around the $NZ10k mark. Think I'm talking about 45nm technology (but could be wrong, don't know a whole lot about it).

For $50 000, $83 per pin, I would really expect some sexy results from this thing.

[jeroen74]

How sexy the results are, only depends on the designer using the chip 530K LEs is enough to place ~150 NIOSII CPU on one chip.

These are monstrous chips with dies the size of big stamps. Low production, low yield... high price Probably made to order too.

[ChipArchitect]

Mmmk. The PhD students here sometimes design their own ASICs and have 10 or so manufactured. Apparently the cost is around the $NZ10k mark. Think I'm talking about 45nm technology (but could be wrong, don't know a whole lot about it).

 No semiconductor company will manufacture only 10 chips for a commercial customer - the cost figure you're mentioning is only valid for subsidized "shared mask" chip manufacturing programs for education. Pretty much every industrialized country either runs, or has access to such subsidized semiconductor prototyping programs (MOSIS in the USA, Europractice in the EU, etc.) that allows students and other university personnel to get their chips manufactured at low prices.

 Not too many small companies have access to these programs, and even if they did; the performance improvement they would get by getting an ASIC manufactured vs. simply implementing an FPGA is often not worth the complexity/cost of designing the ASIC; so an FPGA is usually the most viable solution for low-volume applications.

[amyk]

to get their chips manufactured at low prices.

"low", relatively speaking.

I've looked up prices for chip fab before and thought about whether ways to decrease the initial mask costs, processes, etc. could lead to much cheaper services much like what iTead/Seeed does for PCBs today. It doesn't have to be the latest process with a dozen metal layers, but making rapid chip prototyping more accessible to e.g. hobbyists (and then also small companies) may lower ASIC costs considerably.

[jeroen74]

Why would you want an ASIC? FPGAs are perfect for this if it's only digital. If you need analogue stuff, that's something for a specialist chip designer. Things are done completely different on a chip then you would with regular components.

You have no resistors and capacitors, only in extremely small values, so things are replaced with current sources and mirrors. That's why you some ICs have a 'reference resistor' pin. From that single current a whole bunch of other currents are derived using current mirrors and different transistor geometries. There is an ebook on the internet on analogue chip design, IIRC written by the 555 designer. Very interesting read.

IC fabrication is so expensive compared to PCBs that prices probably never get down that becomes feasible from low volume stuff. Where low is likely millions Except for specialist applications where there's is no alternative to an ASIC; like cochlear implants, there is a reason the implant costs like €25K and the external behind-the-ear piece about €8K.

[free_electron]

So just in my quest to learn more about FPGAs I decided to check out what was on the market. Found the most expensive FPGA on Mouser:

Mmm, you can get those cheaper directly from Altera.
But then again , thes parts are a selected few that perform extremely well from a whole batch.
 We got a few boards laying around from the Dini group. If you think this fpga is expensive.. Try a board that has six of em on it...



You do not wanna know the cost of those things.

These super fpga's are used to prototype asics. Nobody really uses these in production.

[Stiege]

Mmmk. The PhD students here sometimes design their own ASICs and have 10 or so manufactured. Apparently the cost is around the $NZ10k mark. Think I'm talking about 45nm technology (but could be wrong, don't know a whole lot about it).

 No semiconductor company will manufacture only 10 chips for a commercial customer - the cost figure you're mentioning is only valid for subsidized "shared mask" chip manufacturing programs for education. Pretty much every industrialized country either runs, or has access to such subsidized semiconductor prototyping programs (MOSIS in the USA, Europractice in the EU, etc.) that allows students and other university personnel to get their chips manufactured at low prices.

 Not too many small companies have access to these programs, and even if they did; the performance improvement they would get by getting an ASIC manufactured vs. simply implementing an FPGA is often not worth the complexity/cost of designing the ASIC; so an FPGA is usually the most viable solution for low-volume applications.

Thanks, I was aware it was shared mask, but wasn't aware that not everyone had access to that.

[free_electron]

Basically anyone has access to MOSIS. the problem is getting the toolkit and the layout tools...