$/gate for MCU vs FPGA

culating the approximately dollar cost of a typical microcontroller of toda y (say in the $5 to $10 range) in terms of $ per logic gate as compare that to the cost of a typical FPGA chip of today in a similar price range?

ious and I thought it would be neat to see where the trend is moving.

t by how much?

It doesn't strike me as a useful question. MCU's are - first and foremost - easier to program than FPGA's, and you use them if you don't need multi-th reading, or very fine control of process timing. If you want completely ind ependent - but interacting - control of several parallel threads/processes, then an FPGA (or some other form programmable logic) becomes necessary, th ough multi-processor MCU's may provide an alternative.

MCU's seem to be produced in larger volumes, so they are very likely to be cheaper, even before you figure in the extra silicon area used up by the de vices that effect the programmed-in choices.

They are also more likely to use the very latest design rules and manufactu ring technology to get more devices into a smaller area of silicon, which a gain equates to more bang for your buck - if it's the kind of bang you can use.

EnigmaPaul schrieb:

Hello,

a typical microcontroller has some RAM and some EEPROM, but how do you calculate the equivalent gate count?

Bye

An MCU can be synthesized in an FPGA, but an MCU cannot be used as DSP or computation at nearly the same speed. And as others have said, the quantities aren't very similar either. So the comparison isn't very fair.

A better comparison might be, FPGA to digital ASIC (sea-of-gates perhaps). But getting an ASIC isn't cheap, either, so you'd have to further compare in quantity, say 100k or 1M pieces. In quantities like that, I think it's supposed to be a ~5x difference in both area and cost. The ASIC is also faster.

Tim

-- Deep Friar: a very philosophical monk. Website:

You'll probably ask 'why would one want to do that?'. Well, I'm just curious and I thought it would be neat to see where the trend is moving.

I'm presuming that the mass produced gates in an MCU would be cheaper, but by how much?

pw

I appreciate the feedback guys but I was hoping to throw out a challenge to compute the gate cost of each. Throwing out reasons why one should not look at at comparison does not move us towards an answer to the questions. There are scenarios where such a comparison makes sense but justifying those reasons is well beyond the scope of what I had envisioned for this thread.

pw

I will never agree that MCUs are easier to program than FPGAs in any real sense. Most people are accustomed to programming MCUs and so FPGAs seem harder, but in reality HDL is quite easy compared to the numerous issues faced when programming any sort of MCU because of the difficulty making a single CPU do multiple tasks. Regardless...

I think you list of reasons why FPGAs are uses is a bit short. It is not at all uncommon to use and FPGA for high speed or complex interfaces which simply can't be done in software with MCU I/O pins. FPGAs also have a lot more flexibilty in I/O protocols offering a wide range of voltages and standards, much more than what you will find on nearly any MCU.

I can't say FPGAs are produced in larger volumes than MCUs because I have seen no data. But what is important is how many of a particular die are sold and I would bet that is very similar. It is not so important how many chips the industry sells, but how much the volume can reduce the price and that is determined by the volume of a given design.

What is much more likely to impact the "per gate" cost is the additional circuitry on the die. In an MCU the entire chip is designed to minimize complexity and reduce the size. The programmability comes from the memory on the chip which is a highly optimized structure with a minimum of wasted space. In an FPGA the emphasis is on flexibility of the logic so the logic is very space inefficient. Mostly this is because of the routing resources which interconnect the actual gates. In fact there is far more area used for the routing than for the logic on any FPGA.

I think this is exactly backwards. MCUs often use older, more mature processes because in many cases they aren't pushing the logic capacity of the die, rather the die is I/O limited. FPGAs are such regular structures that they are often one of the first products to be produced in a given process technology behind the massive Intel CPUs. I know Xilinx loves to tout this as why they will replace ASICs someday, the cost of the mask sets becomes prohibitive, pushing the minimum production run up and up. But they ignore the fact that an ASIC design can often be done on a process two or even three generations behind what is used in an FPGA because of the huge overhead for the programmability of the FPGA continually making them a *lot* less expensive to bring to market than the FPGAs themselves. So it still comes down to die size for most designs and FPGAs take a big cost hit from the inefficiencies of programmability while ASICs and devices like MCUs have a lot less overhead.

Ok, what numbers have you come up with so far? I can get you pricing for various products. How do you plan to turn that into cost per gate?

I've seen some references to certain MCUs quoting a gate count so I'm presuming someone might have a reference to tabulate some common MCUs with a gate count equivalent.

pw

I think you need to use a better figure of merit.

such as (gates*Hz)/(mm^2)

See:

?-)