Switching to 32-bit MCU from 8-bit "by default"

[cyberfish]

I realized that in most of my projects that don't have an FPGA, I need some sort of a microcontroller.

In most of those situations, the MCU used doesn't really matter. It would just need to do some timing, and toggle some pins. Anything will work.

I have always been using AVRs and PICs for those. I have built perhaps 10-20 AVR/PIC-based projects, as that's what I go to "by default", if there are not really any requirements.

But more and more, I'm realizing that I'm just giving myself a hard time, compared to, for example, just using a Cortex-M3 based MCU.

My biggest complaint for 8-bit micros is the lack of (good) tools. PIC doesn't have good free development tools - MPLAB has restrictions for the free version (and doesn't run on Linux), and SDCC is just too buggy. AVR is a little better thanks to avr-gcc, but debugging requires very expensive tools, and only has primitive hardware support (correct me if I'm wrong, as I haven't been able to afford JTAG ICE). 8051 is in similar situation.

On the other hand, Cortex-M3 has many high quality open source tools - GCC, OpenOCD, GDB, Eclipse. Cheap JTAG programmer/debugger ($50 - just slightly more expensive than PICKit or AVRISP), and is available from quite a few manufacturers, so a lot more choices of compatible chips with different emphasis (speed, power consumption, size, etc).

Sure, for applications where you need 8-pin $0.50 micros, 8-bit ones are probably a better idea. But most of my projects are a LITTLE bigger than that. My by far most used AVR/PICs are in the $4-$5 range, and there are many Cortex-M3s in that range also.

So now I don't really see any reason why I shouldn't switch to 32-bit. Sure, I'll need to learn some new stuff, but it's new stuff I'll probably be using for a long time, so I don't mind that.

I know some people stick with 8-bit micros for 5V and DIP. That's the reason I chose to start with them as well (beside the fact that my school taught them first). Nowadays, though, all my projects are 3.3V, and all PICs/AVRs I use are TQFP anyways, so those things aren't really important any more. I make a lot of PCBs and I'm pretty good at soldering TQFP now. I can probably do them faster than DIPs (no huge pins to heat up).

The extra speed/functionality is just nice bonus. It makes my life a lot easier to stick with one architecture from the $2 range to $20 range.

What do you think?

Edit:
Using a 32-bit architecture also makes porting PC code easier, because many people assume int's are at least 32-bits when they program for PC. That means, things like compression or XML libraries can be easily ported over.

There's also no weird memory layouts to worry about. Everything is flat.

[johnmx]

PIC doesn't have good free development tools - MPLAB has restrictions for the free version (and doesn't run on Linux), and SDCC is just too buggy.

MPLAB is completely free, no restrictions at all.

In your specific case you should move to Cortex-M3.

[slburris]

I'm 100% behind this.  I used to use various PICs and AVrs myself, but unless
I've got an application that really calls for a $0.50 - $2.00 micro, more and more
I'm using some flavor of TI (was Luminary Micro) Cortex M3. 

The cost difference is minimal now, and it's just silly to run out of flash or RAM
and have to quibble about a few bytes.  It'd nice to have some big boy features
like hardware division as well.

I know of people who go to the other extreme, like using a PC104 based PC to toggle
a signal line, but that's not a case of price parity, just not wanting to use unfamiliar stuff.

I've said it before in other parts of the forum, but for $12 you can put an ARM Cortex M3
and a Spartan 3A FPGA on a board, and with that combo, you can solve an awful lot of
problems without stressing out about it.  But that's just me.

Scott

[EEVblog]

You just gave yourself every reason to use them, no one will call you a fool for doing so, all your points are valid for you. They may not however be valid for someone else.
These choices depend entirely on your own circumstances.

Dave.

[cyberfish]

I guess what I'm really curious about, is why 8-bit MCUs are still so popular?

I totally agree that there's a market for it - the very low cost segment ($0.50-$1.50) and hobbyists, but they are very popular beyond those markets, too.

For example, this very popular PIC (18F4550) - http://search.digikey.com/scripts/DkSearch/dksus.dll?Detail&name=PIC18F4550-I/PT-ND

Digikey has almost 30000 of those. It's TQFP, so probably not for hobbyists. Besides, there is no way hobbyists can generate that kind of volume.

That means it must be used in some real products by engineers.

But we can easily find better, cheaper, faster 32-bit MCUs. Eg. http://search.digikey.com/scripts/DkSearch/dksus.dll?Detail&name=497-10499-ND

So why are the expensive PICs still on the market? What can they do that ARMs can't? Well, except running at 5V. But how important is that? Especially in non-hobbyist markets?

[Trigger]

When it's for your own use then there's no reason not to.  The Cortex-M3 has a lot of options and Energy Micro has placed them in the low power usage realm that was normally reserved for 8 bit micros.

[Rufus]

So why are the expensive PICs still on the market? What can they do that ARMs can't?

Do you think products get re-designed every time a new more capable cheaper processor comes out?   

It is likely I will soon be re-designing the internals of a product which has been in production since 1995. It is only being redesigned because some of the components are becoming impossible to buy.

For me that is a potential issue with the rash of dirt cheap ARM based processors. There is no money in making them without huge volume and too many manufacturers for them all to have huge volumes so I suspect a lot of parts (and possibly companies) will go obsolete pretty quickly.

Can anyone tell me which ARM cored processors I will still be able to buy in 15 years time?

[cyberfish]

Ah so they are only used to support old designs, and should not be used for new designs?

Can anyone tell me which ARM cored processors I will still be able to buy in 15 years time?

Isn't it a bigger problem when only 1 company makes your chip?

If your specific ARM chip disappears, you only need to change the PCB a bit and re-route a few signals to use a new compatible chip.

If the PIC series disappears, you'll have a lot of code to rewrite.

[johnmx]

If the PIC series disappears, you'll have a lot of code to rewrite.

That's why you can still buy the first microcontroller developed by Microchip Inc. (correct me if I’m wrong but I think it is the PIC16C54).

[EEVblog]

I guess what I'm really curious about, is why 8-bit MCUs are still so popular?

I totally agree that there's a market for it - the very low cost segment ($0.50-$1.50) and hobbyists, but they are very popular beyond those markets, too.

That means it must be used in some real products by engineers.

Of course they are, Microchip didn't become #1 in the microcontroller business at one point (dropped from the top now) by hobby sales. They sell billions of microcontrollers into everything you can possibly imagine.

But we can easily find better, cheaper, faster 32-bit MCUs. Eg. http://search.digikey.com/scripts/DkSearch/dksus.dll?Detail&name=497-10499-ND

So why are the expensive PICs still on the market? What can they do that ARMs can't? Well, except running at 5V. But how important is that? Especially in non-hobbyist markets?

PIC's and 8bit micros are still cheaper than 32bit micros. You can't just take one Digikey example and compare them.
When you manufacture in the millions you don't buy through Digikey, and you even get special deals from the manufacturers, and when you talk serious volume, silicon area matters, it matters a lot. All things considered, you simply can't make a 32bit die cheaper than a smaller 8bit die.
Not to mention all other things like existing tool/family/vendor familiarity. Companies that use these parts can't just change willy-nilly based on what chip they think is slightly cheaper this month.

Dave.

[EEVblog]

For me that is a potential issue with the rash of dirt cheap ARM based processors. There is no money in making them without huge volume and too many manufacturers for them all to have huge volumes so I suspect a lot of parts (and possibly companies) will go obsolete pretty quickly.

Actually, this is one of the reasons why the Microchip CEO himself told me they did not chose to go with ARM for their 32bit micros.
ARM has become a commodity, and as such there is no money in it, and hence actually more likelihood of your particular ARM micro being discontinued than Microchip (or another major player) folding and not making your "single source" part any more.
ARM micros are single source just like any other micro, only thing is hardly anyone is making money selling them except except for ARM themselves who sit back and collect the royalties and don't care if your chip gets discontinued.

Dave.

[johnmx]

I guess what I'm really curious about, is why 8-bit MCUs are still so popular?

Do you prefer to start learn some matter with a simple case or with a complex and advanced case? The best way is start with the simplest case. And as we learn we move to more advanced levels.

Most people start working with microcontrollers in a young age and don’t have money to spend in this hobby. In my case I had to work at night to pay the tuition fees, so there was no money left for hobbies. Thanks to Microchip Inc. and their free samples program I started to learn by myself without spending a nickel. Also the application notes they have are an excellent source of help. 3 years later, I knew more about 8-bit microcontrollers than any professor in my University, because they only knew about 32-bit microprocessors.

Nowadays I just use 8-bit micros for very simple projects. But if I need to make one, I would definitely use a PICmicro.
Microchip Inc. has predicted this and that’s why they grow up so much in the last couple of decades.

[EEVblog]

Ah so they are only used to support old designs, and should not be used for new designs?

Can anyone tell me which ARM cored processors I will still be able to buy in 15 years time?

Isn't it a bigger problem when only 1 company makes your chip?

No, it's actually safer, as has been proven over several decades with the major players.

If your specific ARM chip disappears, you only need to change the PCB a bit and re-route a few signals to use a new compatible chip.

Sure, but that's actually a big deal too, and actually more likely to happen with ARM chips than say a Microchip or Renesas part.

If the PIC series disappears, you'll have a lot of code to rewrite.

Microchip won't disappear, that sort of things does not happen in this industry to the major players. They are actually "too big to fail" because the industry itself demands they stay afloat.

Dave.

[Trigger]

Renesas decided to stay out of the ARM arena as well instead developing their own RX architecture to compete with ARM despite already having 2 types of 32bit micros

When the two largest (and most successful) micro suppliers in the industry shy away from ARM that's something to pay attention to because they'll still be around long after some of the smaller companies are gone.

[alm]

Microchip won't disappear, that sort of things does not happen in this industry to the major players. They are actually "too big to fail" because the industry itself demands they stay afloat.

Microchip may be 'too big to fail', but does the same apply to the PIC32? The PIC32 doesn't appear essential to the electronics industry. I wouldn't be surprised if the PIC32 was discontinued in favor of some other architecture like ARM, if it turned out they failed to achieve the market penetration they expected.

The 8051 still seems to be fairly popular in the industry, even though the situation is fairly similar to ARM (open architecture, lots of vendors).

I agree that the number of ARM 'partners' (or whatever they call it) is likely to be reduced, but ARM going away seems unlikely for now. What will they use in all the gadgets like smart phones?

Renesas decided to stay out of the ARM arena as well instead developing their own RX architecture to compete with ARM despite already having 2 types of 32bit micros

Of course, vendor lock-in is actually a good thing for the vendor. Doesn't mean the same arguments apply for the consumer.

When the two largest (and most successful) micro suppliers in the industry shy away from ARM that's something to pay attention to because they'll still be around long after some of the smaller companies are gone.

I wouldn't exactly call companies like TI and Samsung small.

[EEVblog]

Microchip may be 'too big to fail', but does the same apply to the PIC32? The PIC32 doesn't appear essential to the electronics industry. I wouldn't be surprised if the PIC32 was discontinued in favor of some other architecture like ARM, if it turned out they failed to achieve the market penetration they expected.

They don't expect market penetration, they wanted market differentiation (the CEO's words) from ARM. Renesas did it for the same reason.
I don't think Microchip have ever discontinued an entire family of micros, and market forced would likely not let them do that, it would be too damaging.
You don't know how many big influential customer might be using PIC32.
And there is actually little benefit to a company to discontinue an entire family like that, it would be massively damaging to confidence. No one (not just Microchip) would be dumb enough to do it.

The 8051 still seems to be fairly popular in the industry, even though the situation is fairly similar to ARM (open architecture, lots of vendors).

I agree that the number of ARM 'partners' (or whatever they call it) is likely to be reduced, but ARM going away seems unlikely for now. What will they use in all the gadgets like smart phones?

ARM will never go away, that's not what I meant.
Your comment about the number of partners being reduced is spot on, and that's the problem, it's almost inevitable. Margins on ARM chips are razor thin, and the license fees are high, so that does not equal security. That's why two big player didn't touch ARM.
If anyone thinks they are safer (from having to re-spin a design, even at just the PCB/BOM level) using a generic ARM chip over say a PIC or Renesas part then I think odds are you may ultimately end up being wrong.

When the two largest (and most successful) micro suppliers in the industry shy away from ARM that's something to pay attention to because they'll still be around long after some of the smaller companies are gone.

I wouldn't exactly call companies like TI and Samsung small.
[/quote]

Not at all. But Renesas are the biggest in the industry, and Microchip are what , #3, #4 at the moment?
Samsung are massive, but the problem is they are not just a semiconductor maker, so they have less rep to lose if they decide to discontinue a chip. They are not as focused a microcontroller manufacturer as say Microchip is, and that can mean a lot.

Dave.

[cyberfish]

Ah I see. I didn't realize how big of an issue part longetivity is. I have never worked in the electronics industry.

It would be very nice if ARM becomes something like 8051 in the future, when the patents expire, though. No royalty, and many implementations from many companies.

I don't think familiarity should be too big of a reason. It's not like we need to educate consumers about anything. Only need to educate engineers, and engineers are supposed to be easy to educate.

I think the ARM-making companies (eg. TI, ST, or NXP) face the same pressure to keep their ARM CPUs running, too. If they discontinue a popular series, like STM32, or LPC1000 series, they will be in similar situation.

PIC was #1 before, but that's before the 32-bit and 16-bit stuff became cheap enough to compete (high end 8-bit vs low end 32-bit). With the competition that only started in the last few years, they dropped to #3 or #4. And many people only use them because of familiarity and existing designs. ARM chips are now starting to compete in the mid to low end market as well (except for very lowest end). How much longer can they stay popular for? Even if it never completely goes away, they will probably shift most of their resources to other products. Wouldn't that make designing with them even harder?

[EEVblog]

PIC was #1 before, but that's before the 32-bit and 16-bit stuff became cheap enough to compete (high end 8-bit vs low end 32-bit). With the competition that only started in the last few years, they dropped to #3 or #4. And many people only use them because of familiarity and existing designs. ARM chips are now starting to compete in the mid to low end market as well (except for very lowest end). How much longer can they stay popular for? Even if it never completely goes away, they will probably shift most of their resources to other products. Wouldn't that make designing with them even harder?

People have been predicting the demise of 8bit micros for a long time, but it's really never going to happen. There are just too many apps that need ultra low cost in ultra tiny form factor.

Dave.

[mikeselectricstuff]

In terms of compatibility, peripherals and pinouts  are way more important than the core, both  in terms of both hardware functionality and programmer familiarity.
Changing a design from a NXP ARM to an ST ARM is likely to be harder than changing from PIC24 to PIC33 - Microchip is fairly unique in having very similar (and backwards compatible) peripherals on all their parts from 8 to 32 bit.

In terms of free devtools, although the free version of Hitech C for the 8-bit pics sucks due to code bloat (the paid version is very good and not very expensive), the free versions of Microchip's MPLAB-C for the 16 and 32 bit parts are a lot more useable as the architecture of the chips is more compiler-oriented.

If your projects are reasonably small, I'd highly recommend looking at IAR's free kickstart editions of embedded workbench, which are code-size limited (AVR 4K, ARM 32K) but otherwise fully functional professional quality compilers and debuggers, and available for many different MCUs (AVR,ARM,MSP430), so you can use the same tools whicheve part you find is most suitable for your application.

As with most things, you get what you pay for. If you're doing commercial development, skimping on tools is a false economy.
A good commercial devtool can easily pay for itself in saved time on a single project.

[EEVblog]

Changing a design from a NXP ARM to an ST ARM is likely to be harder than changing from PIC24 to PIC33 - Microchip is fairly unique in having very similar (and backwards compatible) peripherals on all their parts from 8 to 32 bit.

And pinout compatibility too. Often it's possible to move though the different families without having to change your board at all.

As far as cores go, it would have been really nice if Renesas chose the MIPS core like Microchip, but I guess they had their own agenda.

Dave.

[westfw]

For example, this very popular PIC (18F4550) ...
That means it must be used in some real products by engineers.

Ah yes.  One of the first microcontrollers to include a USB device peripheral, so a lot of people jumped on that one for a lot of applications.

But we can easily find better, cheaper, faster 32-bit MCUs. Eg. http://search.digikey.com/scripts/DkSearch/dksus.dll?Detail&name=497-10499-ND

That particular part doesn't have USB, so it's a poor comparison.

My personal experience is that 8bit microcontrollers are HARDWARILLY simple and robust, while the equivalently priced 32bit chips are sort of a pain; multiple voltages, bunch of pins reserved for JTAG, lousy pin drive capability and internal design rules that make them more suceptable to damage, etc.  The 18F4550 chip you mention is also available in a DIP package for easy experimentation on a protoboard, for example.  Software is pretty much a wash; most things have inexpensive debug tools (ICD2 or ICD3 for PIC, DebugWire for AVR, Whatever TI calls theirs.   JTAG (usually per-vendor specific?) for 32bit.  And most have reasonably decent reasonably price compilers (though I haven't been tempted to use SDCC.)  It not even like the 8bit chips are short on program or ram memory space anymore.

[cyberfish]

That particular part doesn't have USB, so it's a poor comparison.

http://search.digikey.com/scripts/DkSearch/dksus.dll?Detail&name=568-4944-ND

Many more expensive (~$10) ones nowadays have USB 2.0 HS OTG.

Almost if not all Cortex-M3s are single voltage (3.3V). Pins are not reserved for JTAG (they can be used for GPIO, just like ISP pins), STM32 has 20mA drive capability I believe.

DebugWire requires JTAGICE ($300). ICD is also $200+.

ARM JTAG adapters, just like the CPUs, are available from multiple vendors, from as low as 29 Euro (http://www.amontec.com/jtagkey-tiny.shtml), or you can easily build your own with a FT2232D and a few passive components. Or integrate it into your own ARM board. They all work with all ARM processors (ARM has a standard JTAG pin-out). As an added bonus, I think they also program Lattice FPGAs.

There are also state of art open source compiler and debugger, with huge online support communities. And everything has full Linux support (MPLAB and AVR Studio only run on Windows).

[mikeselectricstuff]

It not even like the 8bit chips are short on program or ram memory space anymore.

Actually lack of RAM is often the primary reason I need to go from 8 to 32 bit - once you need more than about 4K there are usually cheaper 32 bit options.

[alm]

DebugWire requires JTAGICE ($300).

debugWire also works with the $50 AVR Dragon. But your point about being proprietary protocol still stands.

ARM JTAG adapters, just like the CPUs, are available from multiple vendors, from as low as 29 Euro (http://www.amontec.com/jtagkey-tiny.shtml), or you can easily build your own with a FT2232D and a few passive components. Or integrate it into your own ARM board. They all work with all ARM processors (ARM has a standard JTAG pin-out). As an added bonus, I think they also program Lattice FPGAs.

Don't most of the small Cortex parts use a different protocol like SWD? Is this protocol open and are there cheap programmers available? Last time I checked, SWD in OpenOCD was still work in progress. Not sure about the STM32 equivalent.

There are also state of art open source compiler and debugger, with huge online support communities. And everything has full Linux support (MPLAB and AVR Studio only run on Windows).

Open source AVR support appears to be quite close to open source ARM support. An open source development environment is fairly easy to setup in my opinion (included in many Linux distributions), ARM seemed a bit more involved last time I checked.

People have been predicting the demise of 8bit micros for a long time, but it's really never going to happen. There are just too many apps that need ultra low cost in ultra tiny form factor.

I wouldn't expect the 8-bit micros to go away anytime soon, it's hard to compete with the bottom end, but I would expect the high-end 8/16-bit parts to get some stiff competition from the low-end 32-bit parts (eg. Cortex M0). Performance, RAM and ROM tends to be cheaper in 32-bit parts due to more advanced (and lower voltage) processes. Like Mike writes, more than say 4k of RAM tends to get expensive and rare in 8-bit micros. For example, the cheapest Atmel part with >4k of RAM is the $7.5 ATmega1284 with 40/44 pins.

[cyberfish]

Ah! I didn't know about AVR Dragon.

I don't know much about SWD, but I believe it is a standard from ARM also. I think STM32 also uses SWD, but I've never seen an ARM chip too small to have JTAG.

AVR open source toolchain is quite good. That's the main reason for my preference of AVR over PIC. I don't know how hard the ARM toolchain is to set up yet. I just ordered my JTAG adapter, and designing my first STM32 board right now.