Will FPGAs be the Successors of the Microcontrollers?
Christof Eb.
Posts: 1,157
The availability of Tang Nano 20K brings a FPGA module with closely coupled 8 megabytes of SDRAM into a price region, that had been occupied by microcontrollers. It's 20K LUTs are well sufficient for all the "classic softcores". Special peripheral pin functions -complex or simple- can freely be implemented as needed by the project.
After playing with such board for some days I ask myself, if this is a critical step into a rather changed world of controllers? There also still seems to be room for educational stuff in this area?
Christof
Comments
Not really, the price point is still very different.
Small MCU have dropped to a 20~30c ballpark, and as with flash drives, you keep getting more for that few cents.
If you want to compare module prices, the Pi PICO is a useful MCU module price point.
Maybe for Parallax parts you could compare P1 FLIP with Tang Nano ?
Well, yes, I was thinking about modules for makers more than about chips. I own PICOs and ESP32s and P1 FLIP or rather it's predecessor. I am impressed, that my humble first try port of a 32bit Forth processor to the Tang running at 27MHz is practically executing Forth in a simple benchmark as fast as one core of P2 @200MHz. I also ported some clever VHDL, which can handle the protocol to init a SDcard without software!!! Seems to be much beyond "Smart Pins". But funnily the 8MB of RAM might make the difference.
Perhaps the future will bring a combination of a fixed processor with FPGA area to flexibly implement special functions?
This has been around for some years now, for example like tis. Not super cheap but still $89 MiniZed seems like a reasonable offer for what one gets.
Yes, ARM cores partnered with FPGA fabric have been at the high end for quite a lot of years. Not cheap. The AI craze has prompted even more power hungry parts to be pushed recently. Hence Intel buying Altera then AMD buying Xilinx.
PS: I do wonder how those companies pairings occurred. Was Altera already utilising Intel foundries before Intel gobbled them. And was Xilinx a regular fixture at TSMC like AMD is.
I had been looking for low cost FPGA boards for some time. During Covid they had vanished somehow. It seems, that during this period of difficult sourcing, the western firms have focused even more on the high end, where the profit per part will be bigger. And these "pairings" are in line with focus on the high end. It is interesting, that Chinese firms seem to think differently and invest to attack on the lower end, but just powerful "enough".
The combination with SDRAM inside the package is a fine step. When I came to limits with existing processors, it has nearly always been because of lack of memory, not because of power of the processor.
(Of course there has been the parallel universe of way more powerful Linux boards for some years, enabled but also hindered by the complexity of Linux.)
All in all I have the feeling, that these Tang Boards have now reached a critical mass as combination of availability, price, amount of memory and amount of luts, direct usability with USB and breadboard compatibility. It is there NOW. I assume, that this will have some heavy impact, much more than Pico. In my opinion, it will speed up a transition in thinking how controllers work and also about education about controllers, which makes this a topic for the education @ Parallax. ( @"Ken Gracey" ?)
Yes, FPGAs are very useful for lots of things.
The affordable but capable ones even more so but...most of them lack analog peripherals like comparators, op amps, ADC and DAC converters. To have these capabilities one is most often forced to use external components, thus increasing circuit complexity, board space and cost so some of the benefits are lost.
In short, I doubt the FPGAs will replace micro controllers entirely any time soon. They both evolve and both have their area of dominance. That's how I see it at present moment.
OK, should have done some Googling and estimations before...
A MC68000 has a transistor count of 68.000 and can be done in an FPGA with about 2500 LUT4s. The given number of transistors per LUT do vary very much. I have found numbers between 200 and 3500. If we take 500, the relation of needed chip sizes 2500*500/68.000 would be about 20. So I see, for larger production numbers purely-Soft-CPUs will remain no option.