[AVR] the new 8 bit micro to rule them all

[Simon]

Microchip have come out with a new µC. Looks like they are attempting an ARM style series and on paper these look great:

https://www.microchip.com/wwwproducts/en/ATMEGA4809

[westfw]

Meh.  Adding an extra 16k of flash and 4k of RAM is not enough, especially since the new "single address space" scheme appears to make that an architectural limit rather than just a chip limit. No speed bump.  No high-speed crystal oscillator.  Wastes most of the 32 bytes of IN/OUT accessible IO registers.Also suffering from the usual Atmel: "oh, we shipped out the first set of chips.  More will be available ... in october"
Still, I bought one of the associated Xplained boards.  This chip is supposed to be used on the next "Arduino Uno Wifi v2" board.
What was wrong with the XMega chips?

[D-Jack]

Same old dying architecture. They were great before ARM took over and they didn't keep up. ARM processors have surpassed them in virtually every aspect, even cost!

Today I sincerely can't think of a situation where it would be better to use an ATMega instead of a low cost ARM cortex-M0

[Dubbie]

About 1/2 the price of a STM32f0

But less than 1/2 the performance as well....

[ataradov]

About 1/2 the price of a STM32f0

What specific devices you are comparing here? As far as I can see it is more expensive than comparable (memory + pins) STM32F0 and even SAM D21 devices in 1K quantity on DigiKey.

5V operation is nice in some applications, I guess.

I like the data section as well. It would be better if any code could just write any flash. This was one of the most annoying things about classical AVRs.

But otherwise, I personally don't see myself going back to AVRs.

I'm not sure if it has been this case for other devices, but this datasheet describes UPDI in great detail. This is a nice touch.

[westfw]

Between architectural limitations, and questionable compiler optimizations, I've been pretty unimpressed with the CM0 recently.   Some of the code gets really ugly, doesn't actually seem faster or more compact than an 8bit cpu EXCEPT for 32bit math, and you pay for that 32bit address space by needing to store a bunch of 32bit constants.  ("lds r1,globalvariable" on AVR: 2 cycles, 3 bytes.  equivalent on M0:8 bytes, somewhat indeterminate cycles.  Which would probably be fine if it was one of those 72MHz chips with 128k of flash, but vendors are selling 25MHz chips with 16k of flash...))(Though I guess the 16k/20MHz chips are supposed to replace the 4k/10MHz 8bit chips, and might, price-wise.  But you need to be careful.)
Fortunately, vendors seem to be jumping on CM4, and it's nicer...
And yeah, these aren't good excuses for limiting yourself to 48kb, architecturally...
(Hmm.  Does the 4809 match the pinout of the smaller Xmega chips?  (Nope, not even close - 48 vs 44 pins.)  How about the smaller SAM chips?  (Nope again.).  Sigh.

[ataradov]

doesn't actually seem faster or more compact than an 8bit cpu EXCEPT for 32bit math

Well, with CM0+ you get 48 MHz for free (or at least 32 MHz). And the core performance is not all that important if you can offload a lot of work to more advanced peripherals, which 8-bitters don't seem to get even with this new device.

Wait for the wave of Cortex-M23 / Cortex-M33 devices (SAM L10/L11 were just announced, I'm sure other vendors will follow soon). Not a bad followup to M0+.

[coppice]

Which would probably be fine if it was one of those 72MHz chips with 128k of flash, but vendors are selling 25MHz chips with 16k of flash...

The vast majority of those 25MHz chips are only running at a few MHz, because that's all most applications need. If you've ever had involvement with MCU applications support, and seen a large number of real world applications, you'll know that processor speed is really important to a certain group of users, but totally uninteresting to most.

[T3sl4co1l]

About 1/2 the price of a STM32f0

But less than 1/2 the performance as well....

I see figures of >1.7 DMIPS/MHz for most ARMs, maybe call it 1-1.5 for the simplest M0s?  I see figures more like 0.3 for AVR, so that'd be more like 1/4th the performance.  Which makes sense given the wider bus and richer instruction set.  And then add on proportionally higher clock rates, and the bus width advantage for higher bandwidth applications...

Tim

[Dubbie]

About 1/2 the price of a STM32f0

What specific devices you are comparing here? As far as I can see it is more expensive than comparable (memory + pins) STM32F0 and even SAM D21 devices in 1K quantity on DigiKey.

You're right, I messed up the part number. A low end comparable pins/memory f0, is about $1 same for the AVR.

So same price for 1/4 the performance...

Not really a great value proposition.

[NorthGuy]

I see figures of >1.7 DMIPS/MHz for most ARMs, maybe call it 1-1.5 for the simplest M0s?  I see figures more like 0.3 for AVR, so that'd be more like 1/4th the performance.  Which makes sense given the wider bus and richer instruction set.  And then add on proportionally higher clock rates, and the bus width advantage for higher bandwidth applications...

Dhrystone operates on "int", which will be 32-bit on ARM and 16-bit on AVR. So, the direct DMIPS comparison is not possible.

Also, the composition of the Dhrystone test is completely different from what you would typically do on 8-bit MCUs.

IMHO, the real-world performance of 8-bit MCUs depends mostly on the periphery selection and organization. Typically, CPU is used mostly for control, not for number crunching. You wouldn't use 8-bit MCU if you wanted number crunching, would you?

[Dubbie]

No but for low power, sometimes it's better to wake up, do your business as fast as possible then go back to sleep.

Having a slower less power hungry clock doesn't always make up for the longer time you spend awake.

[Simon]

The documentation has gone ARM style with things like DAC listed an unavailable. I'd not be surprised if they put an ARM core in this or develop the architecture further to an ATmega-1 and then put an ARM in that.

[ataradov]

I'd not be surprised if they put an ARM core in this or develop the architecture further to an ATmega-1 and then put an ARM in that.

What would be the point of that and how it will actually be different from existing ARM chips that already have better versions of the peripherals present in this device?

The style of documentation has changed because of transition to a common document preparation flow, it has no alternative or hidden meaning.

[Simon]

commonality of architecture and an easier transition.

[Kleinstein]

That new chip is just another 8 bit AVR. Nothing really new. The AVRs are can not really compete with the ARMs when it comes to CPU performance in most cases. Most of them will not run at a high speed - more like 1 MHz from the internal RC clock is usually enough for the typical applications for a 8 bit µC. The limited memory is not a big deal in small application - if the need some 1-4 K maybe 6 K it is sufficient to know there are upgrades to 16 K / 32 K / 48 K too. The large memory µCs are not that attractive for price reasons anyway - here the smaller feature size in the ARM based µCs can make them even cheaper.  The 8 Bit µCs are more for those 80-90% of the chips where 8 K and 4 MIPS are sufficient. 

The predictable run time is a plus in some applications - race conditions with the cache sometimes kicking are difficult to find errors. Another point is that the AVRs are rather easy for the beginners - it starts with the length of the erratas.

With the actual availability of the chips, the Xmega series was under different condition: here Atmel was still separate and AFAIK had trouble with external FABs.

[NivagSwerdna]

Microchip have come out with a new µC. Looks like they are attempting an ARM style series and on paper these look great:

https://www.microchip.com/wwwproducts/en/ATMEGA4809

TBH I am getting very bored with the choice of uCs nowadays... so many and so little to differentiate.   I like PIC24 and PIC32MZ because at least they have interesting DMA features but otherwise it is a bit of a yawn.
IMHO Microchip are missing a trick... they have CLC (Configurable Logic Cell) which looks really fun but in reality it is only very small scale/ just a few gates.  If they were to have CLC on steroids in a PIC24 or above then at least it would be interesting.
Yesterday I was sorting the collection of dev boards I have but have never got around to trying e.g. TI launchpad Value Sensing line and.... meh.
STM is on my list to explore deeper and I have a few PIC projects already on the drawing board but having configurable logic (and either soft-core or SoC) i.e. FPGA is far more interesting.

[tggzzz]

Microchip have come out with a new µC. Looks like they are attempting an ARM style series and on paper these look great:

https://www.microchip.com/wwwproducts/en/ATMEGA4809

TBH I am getting very bored with the choice of uCs nowadays... so many and so little to differentiate.

Very true.

Now have a look at the XMOS xCORE processors and xC.

They are halfway to giving FPGA performance and predictability.

[firewalker]

One aspect AVR are better than ARM is the ease of use. The hardware setup of an AVR is more straight foeward quick and easier to debug. Of course this is a consequence of the limited hardware in comparison with an ARM.

Alexander.

[tggzzz]

One aspect AVR are better than ARM is the ease of use. The hardware setup of an AVR is more straight foeward quick and easier to debug. Of course this is a consequence of the limited hardware in comparison with an ARM.

Alexander.

The hardware setup of the xCORE intercore comms, i/o ports and timers is trivial. This is because the hardware and software was designed by people with a clue and without being unnecessarily constrained by past bandaids stuck on castles built of sand.

[Simon]

One aspect AVR are better than ARM is the ease of use. The hardware setup of an AVR is more straight foeward quick and easier to debug. Of course this is a consequence of the limited hardware in comparison with an ARM.

Alexander.

And the way they have designed this one looks a bit like the ARM style. The ADC feels like it was pinched from a SAM device, it runs 10x faster and has some autonomy. I'm sure there is a market for a device with ARM performance and AVR simplicity.

[NivagSwerdna]

I see this device does have "CCL – Configurable Custom Logic"... i'm sure that used to be called CLC.  Anyway... it's a good start... but it's too small... Go LARGE Microchip!

[nctnico]

One aspect AVR are better than ARM is the ease of use. The hardware setup of an AVR is more straight foeward quick and easier to debug. Of course this is a consequence of the limited hardware in comparison with an ARM.

That depends entirely on what ARM MCU you are looking at. NXP's documentation for their LPC ARM series is exellent and their peripherals are easy to use.

[ataradov]

That depends entirely on what ARM MCU you are looking at. NXP's documentation for their LPC ARM series is exellent and their peripherals are easy to use.

I have not seen a Cortex-M0+ MCU that was particularly hard to use. They all are about the same.

The common pattern I see is people assume that they are hard because of something they read on the internet and resort to manufacturers libraries, which are bad in all cases. And this further reassures that Cortex-M0+ MCUs are hard.

[SiliconWizard]

I wouldn't touch those unless I absolutely had to deal with AVR legacy stuff, as they don't seem to quite compete with even Microchip's own 8-bitters, at least a good chunk of them. They only guarantee operation at > 10 MHz at 5V and power consumption is ok but not that low either. But hey, they do have a 2-cycle hardware multiplier and a 150ksps ADC. Whoo-hoo! For one minute, I thought they might be very cheap and that would warrant a nice market share, but they are not even that cheap compared to a lot of other 8-bitters.

Anyway, just my opinion of course. But I'm pretty sure they mostly target design teams that have a lot of AVR-based designs and knowledge (being playful we could say Atmel's fan base  ), rather than trying to gain new markets. I'd be interested in hearing opinions that differ though.