I wrote up some of my thoughts about various microcontroller families. My intent is to mostly to help newbies and students decide what type of microcontrollers to start with.
formatting link
Feel free to correct any possible errors, but I have a right to keep my opinions unless you can change them by presenting new information.
There are many families I have never tried so I did not try to comment on them.
I want to follow this up with another document that points out some good developer boards, ISP programmers, and JTAG debuggers for the various families.
Err ?!? - Do you have any opcode examples of this feature ?
bit
It also has atomic boolean opcodes, and the Bit Addressable space is split 50:50 to SFR space, and RAM data Space. With a language that supports named Boolean variables, this makes for very compact, and readable control code.
Incorrect. There are many C compilers that prove this to be false. Perhaps you meant to say the core was designed before C was ever comtemplated on uC, so more effort is needed to create a usable compiler for the 80C51. There are also other languages, like Basic, Pascal and Modula-2 for the C51 core. PL/M was an early language, that showed what could be done, and a BASIC52 was also very early.
of
Err?!? There is no paging in C51's - you are thinking of PICs (or 8048?) ? The 80C51 has variable length opcodes, and these give tight code. eg Appx speaking, these can access ACC or Registers in 1 byte, direct memory in 2 bytes and XDATA in 3 bytes.
Not correct. There is Move Dir,Dir, and DJNZ Dir, (& others) and the Boolean opcodes mean you CAN write workable interrupt routines that do not use any registers. The 80C51 has MUCH better Memory-Memory opcode access, than an AVR, or PIC. The only device ahead, on Memory-Memory opcodes, would be the Intel
196, and the Zilog Z8 or Infineon XC166 both have register-frame control, that emulates memory-memory, but with the wider opcodes of reg-reg access.
Correct. It is the most sourced core on the planet. [Microchip volumes are climbing, but that is spread over MANY cores - PIC is a brand, _not_ a Core]
peripherals such
Nearly, SiLabs have not released Ethernet C51's yet, but others have.
don?t play
running full
correct. SiLabs offer 12/16/24 bit ADCs, and if tou want HighEnd ADCs, you will likely buy a C51 from Silabs, ADi, BurrBrown (TI), TDK etc
use in high
Although some
the norm.
Other cores:
how NOT
Err, no, I'd call the RS08 well designed, and a good example of "throw everything overboard, until the compiler stops working" It is a minimalist core. ( Like the 80C51, it has variabe length opcodes for good code density.)
toothbrushes.
meager, this
family. This is
choices even
production
Err ?. The RS08 has some unique and pioneering new features, besides the core : It has a wide-supply range, thanks to a clever sub-uA OnChip regulator and a VERY LOW current (< 1uA) timer mode, as well as Analog Comparitor, and Low Voltage Detector, and an On Chip OSC. If you think there are better Smart 555 class choices, I'd like to know what they are!
On-Chip regulators are an emerging technology. (The newest SiLabs and Infineon etc device do this too). It provides a good way to make a truly wide-usage device, that has a Shrink-process core for low cost, and low EMC footprint.
The RS08 is stocked by many catalogue companies, and prices are easily under 50c in smallish volumes. So that's NOT a 'high volume only' device.
MSP430:
this
Yes, but the MSP430 lacks a Low power Cal On-Chip OSC. You have to use the 32KHz Crystal option, which gives HIGH precision, but has cost and vibration impacts.
allows them
of power in
before completing
Other devices with OnChip OSC have fast wakeups, the newest Silabs devices wake up in well under 1uS! The Newest Silabs devices have uA comparitors, and Low icc precison on-Chip trimable oscillators - which opens up lower power RUN than MSP430, if you do not need crystal accuracy. (If you do, then the SiLabs devices also have 32Khz Xtal osc ability, and performance are more comparable.)
Coldfire: Freescale have just released Coldfire V1 core variants, and have (IIRC) a Free compiler up to 128K in Size. The Code fire V1 targets easy Migration from 8 Bit cores.
For learning : SiLabs have an expanding range of impressive, Low cost Eval Stick level products. These all have USB debug access.
Other recent cores of interest: On the fringes, Korean Coreriver have just released a new 4 bit '80c51 subset' variant GC89C510A0. ASM only. Precise on chip Osc, no timers, or INTs, and targeting toys & remote controls. FLASH models under US30c
I could swear that the 8051 had a barrel shifter, but maybe it was one of the variants that had that feature and I just assumed it was part of the original 8051.
The 8051 memory paging I was referring to should have been named differently to avoid confusion. I had in mind the bottleneck of using DPTR to reference memory and the fact that most instructions do not have the ability to reference "external" memory. That's not really paging...it's worse in my mind.
There are msp430 devices with 2% accuracy in the internal osc.
The RS08 can not be cost effective for low volume applications because it can't be used in a lot of different applications. In most real devices you'll need more horsepower. Their are a number of consumer applications where it will shine, but only if the volume is high enough to justify retooling for that chip. An ATTINY or a small PIC has a lot more flexibility and the tooling cost can be amortized out over many applications. The software tooling cost of an ATTINY is also cheap due to gcc. Only very specialized compilers can generate efficient C code for the RS08, and I can't imagine writing assembler code for it (I know some of you can do it well and my hat is definitely off to you).
I really like the Coldfire architecture and I hope it catches on in the GP market. Freescale has had a lot of marketing problems and I hope they get over them. But unfortunately, they sat on their butt and watched Arm eat away at their market share. By the time they woke up they ended up far behind. This new V1 thing seems like a good idea, but is probably a few years late in coming.
I'll make some changes to the document to fix the errors with 8051 info. and maybe to clarify some other points. If anyone else sees issues, please let me know.
8051 These are a Harvard family that pre-dated the usage of the word Harvard.
I thought that the term Harvard Architecture came from Harvard Mark I developed during the second world war 1944 or so.
8051 I don't know any versions of the 8051 that has a barrel shifter for multiple bit shifts
8051 Has a complete instruction set, many C compilers have been successfully written for it.
PIC 8-bit (12F, 16F, 18F) Very successfully used for the applications that it was intended. The Stamp serial memory example shows its flexibility to execute an interpreted language.
" extremely limited hardware stack that makes them especially unsuitable for use with C compilers."
Stack size does not prevent C from being implemented on the architecture there are several good C compilers for the Microchip PIC families of parts including ours.
"RS08 ? the weakest 8-bit line in existence, and a terrifically good example of how NOT to implement a Harvard architecture."
The RS08 is a von Neumann architecture part not Harvard.
The RS08 is a very well implemented minimalist processor with a diverse application possibilities. The RS08 is sub $1.00 part with background mode debugging BDM on chip and a surprisingly powerful instruction set.
Benchmarks on our C6808 compiler tools show that code generated for the RS08 is between 10 and 15% larger than the same applications on 68hc08 and S08 parts. Most of the difference is lack of hardware multiply and divide.
General comment about C compilers. All of the architectures reviewed have good C compilers available. You may be doing yourself a disservice by limiting yourself to no cost tools. The typical commercial C compiler supporting this class of processors costs between 2 and 3 man days. If it saves that much development costs it should be considered.
This has not been my experience with the RS08. This part has a surprising amount of processing power.
I wrote the code generator for a RS08 C compiler. The instruction set is orthogonal and complete. There were no specifically difficult parts in the RS08 compiler implementation At the assembly level it is very similar to many of freescales offerings including others in your review.
What about the PIC32? You seem to have missed it off :o)
--
Regards,
Richard.
+ http://www.FreeRTOS.org
14 official architecture ports, 1000 downloads per week.
+ http://www.SafeRTOS.com
Certified by TÜV as meeting the requirements for safety related systems.
Nice job, your always going to get complains when you hit a nerve of someone's "love child"
anyway, newbies typical uses small processors to do floating point, so here is a modified list I posted a while back to give newbies a ballpark estimate what they can expect
MPS430, 32 bit floats, imagecraft complier, typical cycles add 158 sub 184 mul 332 div 620
ARM, keil complier 32 bit floats, typical cycles add 53 sub 53 mul 48 div 224 sqrt 439 log 435
ARM, GNU complier 3.3.1 32 bit floats, typical cycles add 472 sub 478 mul 439 div 652 sqrt 2387 log 13,523
ARM, GNU complier 3.4.3 32 bit floats add 73 sub 74 mul 49 div 142
8051, keil complier, 32 bit floats, typical cycles add 199 sub 201 mul 219 div 895 sqrt 1117 log 2006
AVR, IAR complier, 32 bit floats add 173 sub 176 mul 175 div 694 sqrt 2586 log 3255
You've skipped the entire R8C/M16C/M32C family from Renesas, which is 16 bit, ranging from 20 pin to 144 pin packages, 4K to 1M flash, lots of peripherals, both free and non-free dev tools, and easy to program in-circuit.
formatting link
They're stack-oriented CISC (i386 developers will find them "comfortable" as the addressing modes and register sets are similarly designed). The larger chips have externally-accessible data/address busses for expansion; I have one board on my desk with 4Mb of external RAM and a CP2200 ethernet port. Speed range is 16-32 MHz; most chips run on 3.3 or 5v (the bigger ones can be split - half the pins run on one voltage, the rest on the other).
The only drawback from a "beginner" point of view is that they're not available in common 0.1" DIP packages, although you can get plans for adapters for the small ones.
That's an opcode reach threshold issue, and every device have those. In the 80C51 you have 128+128 bytes of DATA memory, which can be accessed & also INC/DEC/DJNZ/AND/OR etc _without_ consuming a register pointer. In this memory, the 80C51 is _more_ code efficent than a core like the AVR, where ALL Data memory goes via some pointer, or offset pointer.
A better idea on a uC register ('RISC') core, is a register frame pointer a la C166, and z8. The AVR started life as a RISC IP block, so that was not on their radar.
XDATA is Microcontroller optional, and so has less access opcodes. Many newer C51's have multiple DPTR sets, which eases the bottleneck.
It is not the precision, it is the absence of Low Power on Chip OSC. We have an app right now, where we cannot afford a 32Khz Xtal, and do not need high precision. That's actually not easy to resolve. eg AVR P versions OnChip LF OSC, consumes a large 100-220uA.
Best are newest Silabs devices at 5.5uA and some WDOG Osc, at 1-2uA, but most WDOG osc are very poor in Precison AND trim. There is a definite Icc/Precision trade off. Ideal would be a RS08 with more pins, but that's not around yet...
That does not quite make sense - did you mean tiny (hobby) volumes, where the chip price is immaterial. There you certainly should overshoot the core, as it saves time. Low volume can be up to 10,000 in many chip vendors terms.
There are some with an additional ALU SFR mapped, first seen in the Siemens 80C517. That allowed 32 bit operand, and IIRC included a barrel shift. There is no native opcode for barrel shift.
On Sun, 04 Nov 2007 21:56:00 -0800, I said, "Pick a card, any card" and Eric instead replied:
With all the controversy already embedded (pun intended) in this thread, a Wiki page might be a good start towards a meaningful gathering of thoughts and real information..
[STM32 is way too big for this app, but..] Interesting. No mention of trim, but quite good numbers for 2:1 range. Even their HSI is pretty good at 80uA for 'precision' 8MHz.
The Zilog eZ8-XP series have a similar uA region Wdog osc, and a full 32 bit Wdog divider, so trim is possible.
Being lazy, we'd like a uA osc with precision too :)
But being realistic we can tolerate a uA osc that can be infrequently recalibrated against a burst run of a higher precision CalOsc.
Most chips have 32KHz watch Xtal osc, but on chip micropower ones are 'emerging technology'.
As ST, and Freescale show, it IS possible to make uA region LF RC oscs, all we need, is that peripheral on the right uC !
[Notice again, how the core is not part of this selection process...]
I am not familiar with the RS08, but comparing it with older chips such as the 8080 and z80, and assuming that you are not talking about floating point operations, 16 bit multiply and divide subroutines should take in the order of 200 bytes of code. Somehow I doubt that this explains your 10 to 15% size expansion.
--
Chuck F (cbfalconer at maineline dot net)
Try the download section.
The 10 to 15% the range is from a benchmark of ~20 applications run through both the 6808 and RS08 compilers. The largest single difference was the availability of mul and divide hardware, there are other differences each accounting for a few bytes one way or another.
Our RS08 16 bit mul and divide total 81 bytes 16x16 mult is 38 bytes and 16 by 16 divide is 43 bytes
The 8x8 mult is 18 bytes the accum824 is about 100 bytes. All of the libraries were written in C.
Thanks to all for the helpful comments. I agree with several posters that a Wiki would be a good way to let various people contribute. I don't have any experience setting them up and I'd appreciate it if someone could help set one up.
When I get some time I'll update my document the best I can. It won't please everyone but I want to at least be accurate in the things I say.
I didn't know about extended 8051's with multiple DPTRs. That sounds interesting indeed. I don't think Silabs has that, though. I really haven't looked much at 8051's and I mostly included info on them because they are perhaps the best known MCU family. I'll try to paint them in a more accurate light. One odd thing is that many 8051 variants still don't support internal debugging (breakpoints, single- step, etc). Silabs has their C2 debugger and I guess something like that is integrated on their USB stick. I guess the Intel 80251 died in the marketplace. What ever happened to the Philips XA - is that still around and still being updated?
I'm also very excited about the STM32 and I will add more info about them. They now have a number of DEV kits, including that extremely cool little game with the color LCD and MEMS. At the price they're charging for it, it's almost free and it has an internal JTAG USB solution.
Regarding the R8/M16, I have to say that I'm not the right one to comment about them. I have noticied them, and their recent attempts to actively market them. I'm sure they are fine devices and they have a lot of variants available. Again, this is why a Wiki would be good. I'd love to contribute what I can, but there's a lot of good stuff out there that I don't know about.
I still see a need for 2 separate topics: one on the general features of the families, and another that talks about hardware and software tools. Or maybe the tools should be broken out so we cover hardware and software separately?
Walter - I always appreciate your posts because you are a brilliant man and you're one of the best propeller heads I know (in the most positive sense). In all honesty, you can make anything sound simple, but those of us less fortunate in the gray matter department can't keep up with you. With my limited skillset, I can say that it would be almost impossible for me to write a code generator for the RS08. That's why you're worth every penny you make. I stand by my claims that the RS08 is not really suitable for low volume work, though. If a company already has the tools then it might indeed make sense to use it in low volume applications. But the cost of tooling for something that limited in reach is hard to mitigate.
I'm not against commercial software tools. My paper is aimed at students and newbies and they need low-cost tools. Eval versions of compilers make sense in some cases, especially if they don't expire and their limits are reasonable. Two kinds of limited toolsets have recently surfaced that make a lot of sense: one is tethered to a specific dev board, usually through some kind of serial number on the board. The reasoning here is that I can't use that compiler anywhere except with that one board, so it can be cheap. The other kind of limitation is something like a student license. It can display a notice every time it starts up saying that it's only for use in a non- commerical setting. You might think there'd be a lot of cheating with that, but I disagree. Companies are very sensitive to being sued and they often have consultants and outsiders visit them. I think very few companies would cheat by using tools designated as non-commercial.
For the record, I use IAR with the msp430, and Keil with the Arm. But I use gcc with the AVR and 9s12, and sometimes with the Arm and msp430.
All 80C561's also allow PDATA or MOVX @ Ri access, so for an extra
256 bytes of DATA in XDATA space can be added 'almost for free', without using DPTR.
That's true, it is the most widely sourced core on the planet.
On new ones, that's pretty universal :
Atmel's AT89C51RE2 has OCD, as do the new AT89LPxx series
Winbonds' W89E series also have OCD
Infineion XC8xx also has OCD, and a USB Key evaluation with THREE family members on it.
Neither are candidates for new designs. Both rolled out in OTP, (and small OTP, at that ) right when the market started moving to FLASH, and neither hit critical mass. These also give a lesson in the importance of Binary Compatible, and how a vendors assurance of "just recompile" goes down like a lead-balloon in the real world.
That one is a good example, of the importance of Low Cost Eval/Debug access.
The PIC32 may yet be similar, as Microchip are talking about screen PICtails plug in cards.
Cost of tooling ?! - is ~$40 hard to mitigate :)
formatting link
For this you get a Free C Compiler, USB debug, and RS08 and HC08 support (If you get serious, you can Buy Walters Compiler..:) )
There are thousands of RS08 on the shelf in catalogue companies, in DIP8.
To me that all adds up to VERY student accessible, and a simple core is not a bad place to start learning how these things actually work.
Would you really want to throw over 10 Megabytes of manuals, at a novice student ?
At the Student end of the scale, the focus has moved partly from Chips, to Modules & Debug Access - see the TI and Silabs latest low-cost solutions of a 'dock-able' USB debug, and Freescale's, Infineon's and ST's USB sticks (etc).
Also all the small devices on Student's radar, usually have FREE Tools.
Freescale IIRC are up to 128K free on the ColdFire C1, and all the 908 and RS08 tools are free.
The new PIC32 goes to 64K, and most cores have some form of open source compiler, that may not be the tightest code, but does work.
Zilog have free, and unlmited tools (they bought a compiler company) Zilog also have a opto isolated USB debug, good for power development work. Zilog also have good simulators.
ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here.
All logos and trade names are the property of their respective owners.