Re:Microchip to Acquire Atmel for $3.56B
by jwatte on Wednesday January 20, 2016 @03:44PM
(#10941195)
While Atmel MCUs are nice and robust (especially avr-gcc and avr-libc are great advantages) they seen like quite high cost. Cheaper ARMs probably are eating their lunch.
Meanwhile, I've never been friends with the microchip Dev tools. Clunky, expensive, proprietary. Here's hoping we'll get AVR ergonomics and PIC economics...
But it might be too late to matter. Cortex-M0 eats the low end, M4 the mid end, and the phone SoCs the high end. (Even Intel may be threatened at some point... Different business though)
Re:Microchip to Acquire Atmel for $3.56B
by buserror on Wednesday January 20, 2016 @04:27PM
(#10941501)
Well there still is a case for the AVR -- you get zero pipelining effects, no flash latency etc; it's like a tiny, slow, scalar CPLD for all intent and purpose.
Cortex can't beat that, you can tight loop an AVR to cycle precision and get pretty cool return for your running frequency; the Cortex are nowhere as linear. [0]
Altho, I agree that the bigger parts are useless... xmegas and such have absolutely no reason to exist IMO.
On the other hand the CM3 and especially M4's are awesome beast; and the parts like the STM32's are just peerless in terms of features -- they are just borderline scary.
On the other hand, I was worried a little while back that there was nothing between these trusted 8 bits cores and the ARMs -- but nowadays there is; Espressif esp8266 is pretty cool, but the next one in the pipeline is awesome [1]; 2 (well 3 but) cores, and more peripherals than you can shake a stick at. Thats the first SoC that for me can give the STM32 (and NXP equivalent) a good run for their money (and win, in my opinion).
Dislaimer:
[0] I know the AVR very well, I'm the author of simavr: http://github.com/buserror/simavr -- [1] I'm also one of the beta tester for the ESP31/32 new chipset.
Re:Microchip to Acquire Atmel for $3.56B
by revelation on Wednesday January 20, 2016 @05:31PM
(#10941891)
People care about time, not cycles, and chances are with whatever crystal is running your AVR, you will still not be comfortable despite the lack of post-1990 CPU features in ATmegas.
And finally, the idea of tight loops engineered to cycle precision is dying a slow death in general. You just smother the hard timing requirements with excess frequency and performance.
Curious thing about the ESP8266: it doesn't use ARM or any of the other popular kids on the block but a Tensilica Xtensa architecture.
Re:Microchip to Acquire Atmel for $3.56B
by VLM on Wednesday January 20, 2016 @04:20PM
(#10941447)
The low end is something like the PIC 10F family, when compared to the Cortex-M0 its about a tenth the cost and a fiftieth the power consumed. Lighting an LED takes 100x the power of running a 10F series at full speed.
I can buy some 10F series PICs cheaper than a 555 timer chip, which is pretty amazing.
Re:Microchip to Acquire Atmel for $3.56B
by zokier on Wednesday January 20, 2016 @04:44PM
(#10941603)
> The low end is something like the PIC 10F family, when compared to the Cortex-M0 its about a tenth the cost and a fiftieth the power consumed
I don't really believe either of those claims. At 1ku quantity on Digikey (which admittedly is not the cheapest vendor) cheapest Cortex-M0 chips cost $0.378 and low-power chips (NXP KL03) start from $0.504. That aforementioned KL03 is advertised to run at 50 µA/MHz.
Meanwhile cheapest PIC costs $0.35 and is rated 175µA at 4 MHz, which equals to 43.75 µA/MHz. Of course the power consumption figures are not directly comparable, and something like µA/MIPS might be better metric anyways. But based on this there is no way PIC is anywhere near fifty times more power efficient or ten times cheaper.
Re:Microchip to Acquire Atmel for $3.56B
by emeraldd on Wednesday January 20, 2016 @05:40PM
(#10941933)
Several of the PIC models have a low power mode where they run off an internal RC circuit somewhere in the KHz range. Just enough to wake up on an interrupt or some external stimuli and really pushes their power usage way down.
Re:Microchip to Acquire Atmel for $3.56B
by makomk on Wednesday January 20, 2016 @07:39PM
(#10942567)
For comparison, TI's MSP432 series is specified at 90 uA/MHz running, 850nA sleep with RTC enabled - and that's a Cortex M4 part with hardware FPU, not even an M0. I think Atmel's newest low-power M0 parts are even more impressive. Plus the Internet of Things means companies are wanting to run increasingly demanding code on low-power embedded devices. 8-bit chips are getting squeezed hard by ARM.
Re:Microchip to Acquire Atmel for $3.56B
by IshKebab on Thursday January 21, 2016 @03:14AM
(#10944028)
> But it might be too late to matter. Cortex-M0 eats the low end, M4 the mid end, and the phone SoCs the high end.
You know Atmel makes Cortex M chips too right? They're very good in my experience - e.g. they have some of the very few Cortex chips that support USB High Speed (480 Mb/s).
Re:Microchip to Acquire Atmel for $3.56B
by chillingeffect on Wednesday January 20, 2016 @04:19PM
(#10941443)
Atmel was a sleepy, ran-with company until they had a hit with the tiny and mega series. They were a fresh architecture, ample peripherals, great documentation and open. Who could have anticipated the Arduino's success? Their transition past 8-bit was lackluster due to poor peripherals, poor documentation and unimpressive performance.
They were on the ropes when the cortex cores delivered the coup de grace. I'm not sure they were worth even this much. So long, Atmel. It's been a slice.
Re:Microchip to Acquire Atmel for $3.56B
by Gibbon1 on Wednesday January 20, 2016 @09:53PM
(#10943048)
Would have been nice if instead of the xmega parts with the old AVR core + brand new complex and somewhat buggier peripherals they'd instead swapped out the AVR core with an ARM Cortex M0 core.
Re:Microchip to Acquire Atmel for $3.56B
by buserror on Thursday January 21, 2016 @02:24AM
(#10943869)
Not really. for all intent and purpose, the AVR peripherals are very limited. Yes they are easy to use etc, but they are VERY limited -- no FIFO on any of the serial interfaces for example...
There are much, much better IP to do these kind of things if you decide to upgrade the core...
Re:Microchip to Acquire Atmel for $3.56B
by phyllostachys on Wednesday January 20, 2016 @05:18PM
(#10941813)
RIP Atmel Studios...
I can only imagine how this is going to affect choosing a microcontroller in the future. When I was in school, PIC vs AVR was often debated. I also really liked the selection of ARM Cortex-M parts that Atmel offered. I can only hope that the good parts of Atmel remain into the future.
Re:Microchip to Acquire Atmel for $3.56B
by jason_s on Wednesday January 20, 2016 @05:47PM
(#10941975)
??? Microchip's IDE has always been free; the XC compilers all have free versions, the only restriction being that you can't use the higher level optimizations (-O2 and -O3 for XC16 are premium; not sure about XC8 and XC32)
Re:Microchip to Acquire Atmel for $3.56B
by vvanders on Wednesday January 20, 2016 @05:49PM
(#10941991)
It's been a few years but I also remember a program size restriction on the free compiler.
Honestly, it's a difference in approach to tooling, Atmel took the approach of better, freely distributed tools. Now that they won't be around I don't know that Microchip will have a competitor in that space to keep them honest.
People in this thread keep talking like this is the "death" of Atmel. I don't see it changing much other than being guided by Microchip management to meet higher profit margins. If you think that Atmel's compilers and IDE's are going to be dropped overnight in favor of MPLAB X and Microchip compilers, you are almost surely mistaken; there's a big gap between architectures so it would be a lot of work to get Microchip tools to support Atmel parts.
-- Having a number of I/O pins that is less than the instruction
word size; the SX Series 12-bit instruction word is clearly
smaller than its number of I/O pins.
Re:Microchip to Acquire Atmel for $3.56B
by dmschulman on Wednesday January 20, 2016 @04:02PM
(#10941317)
Atmel is so visible though, no wonder they're getting gobbled up. From a hobbyist point of view, I see so many projects involving Atmel chips is almost makes me think they have a monopoly on that, growing, segment of the market. This probably has as much to do with cheaper cost as it does with (relative) ease of use.
Re:Microchip to Acquire Atmel for $3.56B
by gherkin0 on Wednesday January 20, 2016 @06:59PM
(#10942395)
Isn't it more that Arduino boards dominate the hobbyist microcontroller space, and they have historically used AVR? IIRC, most new Arduino models are now ARM based.
Re:Microchip to Acquire Atmel for $3.56B
by smcl on Wednesday January 20, 2016 @05:23PM
(#10941849)
Not sure if we're able to give specifics, but maybe if the overall offers weren't too different in overall value Microchip offered more cash over stock? The only previous story I read about this is here, which isn't too informative http://www.theregister.co.uk/2016/01/15/microchip_atmel/
Meanwhile, I've never been friends with the microchip Dev tools. Clunky, expensive, proprietary. Here's hoping we'll get AVR ergonomics and PIC economics...
But it might be too late to matter. Cortex-M0 eats the low end, M4 the mid end, and the phone SoCs the high end. (Even Intel may be threatened at some point... Different business though)
Cortex can't beat that, you can tight loop an AVR to cycle precision and get pretty cool return for your running frequency; the Cortex are nowhere as linear. [0]
Altho, I agree that the bigger parts are useless... xmegas and such have absolutely no reason to exist IMO.
On the other hand the CM3 and especially M4's are awesome beast; and the parts like the STM32's are just peerless in terms of features -- they are just borderline scary.
On the other hand, I was worried a little while back that there was nothing between these trusted 8 bits cores and the ARMs -- but nowadays there is; Espressif esp8266 is pretty cool, but the next one in the pipeline is awesome [1]; 2 (well 3 but) cores, and more peripherals than you can shake a stick at. Thats the first SoC that for me can give the STM32 (and NXP equivalent) a good run for their money (and win, in my opinion).
Dislaimer: [0] I know the AVR very well, I'm the author of simavr: http://github.com/buserror/simavr -- [1] I'm also one of the beta tester for the ESP31/32 new chipset.
And finally, the idea of tight loops engineered to cycle precision is dying a slow death in general. You just smother the hard timing requirements with excess frequency and performance.
Curious thing about the ESP8266: it doesn't use ARM or any of the other popular kids on the block but a Tensilica Xtensa architecture.
I can buy some 10F series PICs cheaper than a 555 timer chip, which is pretty amazing.
I don't really believe either of those claims. At 1ku quantity on Digikey (which admittedly is not the cheapest vendor) cheapest Cortex-M0 chips cost $0.378 and low-power chips (NXP KL03) start from $0.504. That aforementioned KL03 is advertised to run at 50 µA/MHz.
Meanwhile cheapest PIC costs $0.35 and is rated 175µA at 4 MHz, which equals to 43.75 µA/MHz. Of course the power consumption figures are not directly comparable, and something like µA/MIPS might be better metric anyways. But based on this there is no way PIC is anywhere near fifty times more power efficient or ten times cheaper.
For a PIC10KF320/322 you're looking at:
Sleep: 20nA, 1.8V Operating: 25uA @ 1Mhz, 1.8V Watchdog Timer: 500nA, 1.8V
Of course that's dependent on what' you're doing with the io pins etc.
See:http://ww1.microchip.com/downloads/en/DeviceDoc/40001585D.pd...
edit: fix typo
You know Atmel makes Cortex M chips too right? They're very good in my experience - e.g. they have some of the very few Cortex chips that support USB High Speed (480 Mb/s).
I have no complaints with them.
Atmel AVR will redesign things out from under you. We could not get a new JTAG interface to work with an old chip.
They were on the ropes when the cortex cores delivered the coup de grace. I'm not sure they were worth even this much. So long, Atmel. It's been a slice.
There are much, much better IP to do these kind of things if you decide to upgrade the core...
Really loved Ateml's IDE. Microchip always charged for their IDE/compiler and I never felt like it was quite as nice to use.
I can only imagine how this is going to affect choosing a microcontroller in the future. When I was in school, PIC vs AVR was often debated. I also really liked the selection of ARM Cortex-M parts that Atmel offered. I can only hope that the good parts of Atmel remain into the future.
Honestly, it's a difference in approach to tooling, Atmel took the approach of better, freely distributed tools. Now that they won't be around I don't know that Microchip will have a competitor in that space to keep them honest.
nope, just optimization.
>Now that they won't be around
People in this thread keep talking like this is the "death" of Atmel. I don't see it changing much other than being guided by Microchip management to meet higher profit margins. If you think that Atmel's compilers and IDE's are going to be dropped overnight in favor of MPLAB X and Microchip compilers, you are almost surely mistaken; there's a big gap between architectures so it would be a lot of work to get Microchip tools to support Atmel parts.
http://www.thefreelibrary.com/Court+Rules+for+Scenix+Semicon...
Featuring such fabulous claims as:
-- Having a number of I/O pins that is less than the instruction word size; the SX Series 12-bit instruction word is clearly smaller than its number of I/O pins.
uno currently retails for $25, the arm-based zero is going for $50