5 ms·

Atmel saw huge success with the AVR at the end of the 90s, and they had the ARM microcontroller market nearly to themselves with their SAM7 and SAM9 series of c

Atmel saw huge success with the AVR at the end of the 90s, and they had the ARM microcontroller market nearly to themselves with their SAM7 and SAM9 series of controllers. Unfortunately they dropped the ball when the Cortex cores came (and ARM really sprung into popularity) which ST gladly picked up and ran away with (around 2008, when the STM32F1 was introduced). What happened to Atmel is that they chose to invest their resources in the AVR32 series (riding on the success of the 8-bit AVR and the strong political position of the AVR group inside Atmel; IIRC the ARM group was in France while the AVR group was based in Norway), but they didn't read the market correctly. Everyone was going away from proprietary architectures and into the Cortex ecosystem (not always rationally) - customers bought into the pseudo-standardization ARM was selling them. This standardization was only perceived, as once you get into the 32-bit domain, the code is mostly C that can easily compile between cores - its the peripheral drivers that are non-standard and where most of the porting work exists and going from an ST Cortex to an Atmel Cortex is a big headache. Anyway, this shift in the market caught Atmel (and others, like Microchip and Renesas) off guard and pushed ST into the Cortex dominance it has today. Around that time, Atmel changed management and refocused the company (selling off their memory lines, other fringe products and their fabs) to be a primarily microcontroller company, but it was a bit too late and they found themselves behind. It's been a while since I worked with 8 bit micros, but the word is that their AVR series is not really innovative anymore. Seems like Atmel is back on its feet though with great new Cortex-F7 processors (I think they have the industry speed record at 300mhz and some other benefits). I haven't looked at the graphs, but $4.6bn is kind of a valuation low point considering their last 10 years or so. They probably understood that they're too small to survive by themselves, especially as how complicated microcontrollers are becoming. I don't think microcontrollers are such a fantastic business - talking to my friends at component distribution like Avnet and Arrow, they make a lot more money on some passive components (like power) than on micros, and for a long time some companies have been allegedly selling microcontrollers at a loss, just to get the designs.

aceperry 11y ago
Nice summary of Atmel. I had no idea that Atmel was not doing well. Atmel has the basic arduino business, although a lot of other architectures, especially newer and more powerful ones, are using the arduino framework for their CPUs. Even the newest official arduino platform is ARM based. I'm constantly reading about how the 8-bit market is still doing well, despite the emergence of ARM. Even though the AVR hasn't really innovated much, it has picked up a lot of peripherals and low power modes which has kept the AVRs from being strictly in the bargain basement MCU market. I don't think there has been much innovation in the 8-bit market, but the AVRs have a really good programming model, especially compared to the 8-bit PICs. I don't know if that's enough to give AVRs a shot at new designs, but they're one of the best 8-bit options today, IMHO.
- flyinglizard 11y ago
  Arduino has been huge for Atmel in terms of brand recognition and popularity of AVRs, but overall, its not a huge revenue source (even if 1m Arduino boards are sold yearly, that's at best $3m-$4m in sales). Of course that there are indirect benefits. Most of the designs I know, migrate away from 8 bits and into 32 bit. The Cortex M0 is great. While designing with 8 bit is somewhat of an upgrade dead-end, if you design a product with an ST Cortex-M0, you can pretty easily go up their Cortex family ladder while retaining a lot of compatibility on both the board and firmware levels (ST is particularly good at that, which is why its my default recommendation when designing a new product - I can explain why if anyone is interested). The denser 8 bit parts are really expensive anyway - I suspect its because manufacturers know its mostly "locked in" customers that need more flash or I/O for an existing design and need a more capable, near drop-in replacement. I also vastly prefer AVRs to PICs, but AVRs were kinda where I started my embedded programming career so I guess I just reflect on them fondly.
  - paulmd 11y ago
    Please do. I can easily see the benefits of being able to upgrade a chip while retaining software/hardware compatibility but I'd love to hear any other selling points. Are they reasonably competitive with equivalent AVRs in terms of cost and power? Do they have a PicoPower equivalent?
    - flyinglizard 11y ago
      PicoPower - I can't tell you the specifics, as comparing power consumption is a bit difficult (you need to figure out which peripherals you need running, whats the system clock, input voltage, sleep/wakeup regime etc and calculate from there), but they do have STM32L0 series of ultra-low power based on Cortex-M0+. As to why do I like ST: 1. They are very easy to design boards for. Each IO pin on the Cortex micros maps to up to 16 different internal functions. That gives amazing flexibility from the board design perspective. 2. They have fantastic long term migration paths. You can directly go from Cortex-M3 to Cortex-M4 to Cortex-M7, while, in many cases, retaining package compatibility. 3. They have very good design validation. Their errata sheets are quite shorter than what I remember working with other manufacturers. 4. They have a huge amount of useful IO, such as timers (I have one design using over 15 timer channels) and a quite flexible and powerful ADC architecture. 5. They probably have the widest selection of Cortex microcontrollers in the industry. There are downsides, too; I generally find their peripherals less advanced than Atmel's (regular CAN instead of CAN-FD, USB inferior to Atmel's, I2C that's really sucky and prone to errors, slower clock and no dual precision FPU on the new F7 family). There was also some story with ST's 2MB flash parts, being not really usable - but I'm not sure what was that about. Overall you can't go wrong with ST. The long term migration ability and board design flexibility are well worth the downsides. Coming from an AVR, you're immediately going to get way more memory, much higher clock frequencies, ARM CMSIS libraries (fun if you have any math/DSP going on) and more industry standard tools.
Gibbon1 11y ago
You're exactly right about the peripherals being the important part for micro controllers. Atmel when they developed the xmega and avr32 families radically changed the peripherals. At that point switching to an ARM Cortex was just as hard. Given that switching instruction sets is a nothing burger if most of your code is in C/C++.
robert_foss 11y ago
Cortex-F7 (or just F7) is the ST brand name of their Coretex-M7 based products. ARM only sells A/M/R series of processors.
- flyinglizard 11y ago
  Yes yes, sorry, that was a typo ;)
coryrc 11y ago
You've totally ignored the 8051 market, which is most competitive and gets the newest features well before AVR-core processors. Not to mention their dozens of other product lines. I think the Arduino has led people to believing the AVR are more popular in industry than they actually are -- it isn't hobbyists buying tens of millions of microcontrollers per year.
- dfox 11y ago
  I would think that in mass manufactured products AVRs are actually more common than Atmel made '51s, and for the whole '51 market Atmel seems to be mostly irrelevant as most '51s you see today are various highly integrated special purpose SoC's from mostly asian manufacturers. I've not seen anything that can be called "consumer electronics" that contained Atmel made '51, while I've seen some such things with AVR's (probably most often with mega169). In various low-volume industrial stuff there are heaps of Atmel made '51s, but only reason for that is familiarity (and to lesser extent AVR-compatible ISP). Another thing is that Atmel is prone to supply chain problems. In our case it meant that we had redesigned the same device three times, first design was based on mega8, which in late 2010 we found was completely unavailable in any non-prototype quantities, thus we found that our distributor had cheap inventory of AT89C2051 and went with that (with all the development pains that are inherent in non-ISP '51) and on last major revision we went with MSP430 (G2253 IIRC) and didn't have to change anything due to supply problems for 3 years. In any case I believe that for many uses of Atmel microcontrollers the familiarity and avaiability (like mega8 being fifth of the price of PIC16F84 in unit quantities in ~2007) were the major selling points.
sliverstorm 11y ago
I was designing a many-sensor system back in very early 2011 and we wanted to go with AVR, but couldn't. They didn't have enough memory, enough flash, enough IO, enough anything, even on their $20 chips. Then we found the STM32F lines and got chips that were dramatically, objectively superior to the AVR32 chips in literally every way, for about $3/piece. I still feel like I must have missed something for there to have been such a gap, but the project went great. So, I think there may have been more to it than marketing pitch about portability.
petra 11y ago
>> and for a long time some companies have been allegedly selling microcontrollers at a loss, So why aren't companies rushing to build 40nm commodity mcu's like infineon did ? it should cost almost nothing to make an mcu that way.
- flyinglizard 11y ago
  I don't know - I believe the barrier is in the embedded flash technology used (NOR flashes that are fast/reliable/big enough).

aceperry 11y ago

Nice summary of Atmel. I had no idea that Atmel was not doing well. Atmel has the basic arduino business, although a lot of other architectures, especially newer and more powerful ones, are using the arduino framework for their CPUs. Even the newest official arduino platform is ARM based. I'm constantly reading about how the 8-bit market is still doing well, despite the emergence of ARM. Even though the AVR hasn't really innovated much, it has picked up a lot of peripherals and low power modes which has kept the AVRs from being strictly in the bargain basement MCU market. I don't think there has been much innovation in the 8-bit market, but the AVRs have a really good programming model, especially compared to the 8-bit PICs. I don't know if that's enough to give AVRs a shot at new designs, but they're one of the best 8-bit options today, IMHO.

flyinglizard 11y ago
Arduino has been huge for Atmel in terms of brand recognition and popularity of AVRs, but overall, its not a huge revenue source (even if 1m Arduino boards are sold yearly, that's at best $3m-$4m in sales). Of course that there are indirect benefits. Most of the designs I know, migrate away from 8 bits and into 32 bit. The Cortex M0 is great. While designing with 8 bit is somewhat of an upgrade dead-end, if you design a product with an ST Cortex-M0, you can pretty easily go up their Cortex family ladder while retaining a lot of compatibility on both the board and firmware levels (ST is particularly good at that, which is why its my default recommendation when designing a new product - I can explain why if anyone is interested). The denser 8 bit parts are really expensive anyway - I suspect its because manufacturers know its mostly "locked in" customers that need more flash or I/O for an existing design and need a more capable, near drop-in replacement. I also vastly prefer AVRs to PICs, but AVRs were kinda where I started my embedded programming career so I guess I just reflect on them fondly.
- paulmd 11y ago
  Please do. I can easily see the benefits of being able to upgrade a chip while retaining software/hardware compatibility but I'd love to hear any other selling points. Are they reasonably competitive with equivalent AVRs in terms of cost and power? Do they have a PicoPower equivalent?
  - flyinglizard 11y ago
    PicoPower - I can't tell you the specifics, as comparing power consumption is a bit difficult (you need to figure out which peripherals you need running, whats the system clock, input voltage, sleep/wakeup regime etc and calculate from there), but they do have STM32L0 series of ultra-low power based on Cortex-M0+. As to why do I like ST: 1. They are very easy to design boards for. Each IO pin on the Cortex micros maps to up to 16 different internal functions. That gives amazing flexibility from the board design perspective. 2. They have fantastic long term migration paths. You can directly go from Cortex-M3 to Cortex-M4 to Cortex-M7, while, in many cases, retaining package compatibility. 3. They have very good design validation. Their errata sheets are quite shorter than what I remember working with other manufacturers. 4. They have a huge amount of useful IO, such as timers (I have one design using over 15 timer channels) and a quite flexible and powerful ADC architecture. 5. They probably have the widest selection of Cortex microcontrollers in the industry. There are downsides, too; I generally find their peripherals less advanced than Atmel's (regular CAN instead of CAN-FD, USB inferior to Atmel's, I2C that's really sucky and prone to errors, slower clock and no dual precision FPU on the new F7 family). There was also some story with ST's 2MB flash parts, being not really usable - but I'm not sure what was that about. Overall you can't go wrong with ST. The long term migration ability and board design flexibility are well worth the downsides. Coming from an AVR, you're immediately going to get way more memory, much higher clock frequencies, ARM CMSIS libraries (fun if you have any math/DSP going on) and more industry standard tools.

Gibbon1 11y ago

You're exactly right about the peripherals being the important part for micro controllers. Atmel when they developed the xmega and avr32 families radically changed the peripherals. At that point switching to an ARM Cortex was just as hard. Given that switching instruction sets is a nothing burger if most of your code is in C/C++.

robert_foss 11y ago

Cortex-F7 (or just F7) is the ST brand name of their Coretex-M7 based products. ARM only sells A/M/R series of processors.

flyinglizard 11y ago
Yes yes, sorry, that was a typo ;)

coryrc 11y ago

You've totally ignored the 8051 market, which is most competitive and gets the newest features well before AVR-core processors. Not to mention their dozens of other product lines. I think the Arduino has led people to believing the AVR are more popular in industry than they actually are -- it isn't hobbyists buying tens of millions of microcontrollers per year.

dfox 11y ago
I would think that in mass manufactured products AVRs are actually more common than Atmel made '51s, and for the whole '51 market Atmel seems to be mostly irrelevant as most '51s you see today are various highly integrated special purpose SoC's from mostly asian manufacturers. I've not seen anything that can be called "consumer electronics" that contained Atmel made '51, while I've seen some such things with AVR's (probably most often with mega169). In various low-volume industrial stuff there are heaps of Atmel made '51s, but only reason for that is familiarity (and to lesser extent AVR-compatible ISP). Another thing is that Atmel is prone to supply chain problems. In our case it meant that we had redesigned the same device three times, first design was based on mega8, which in late 2010 we found was completely unavailable in any non-prototype quantities, thus we found that our distributor had cheap inventory of AT89C2051 and went with that (with all the development pains that are inherent in non-ISP '51) and on last major revision we went with MSP430 (G2253 IIRC) and didn't have to change anything due to supply problems for 3 years. In any case I believe that for many uses of Atmel microcontrollers the familiarity and avaiability (like mega8 being fifth of the price of PIC16F84 in unit quantities in ~2007) were the major selling points.

sliverstorm 11y ago

I was designing a many-sensor system back in very early 2011 and we wanted to go with AVR, but couldn't. They didn't have enough memory, enough flash, enough IO, enough anything, even on their $20 chips. Then we found the STM32F lines and got chips that were dramatically, objectively superior to the AVR32 chips in literally every way, for about $3/piece. I still feel like I must have missed something for there to have been such a gap, but the project went great. So, I think there may have been more to it than marketing pitch about portability.

petra 11y ago

>> and for a long time some companies have been allegedly selling microcontrollers at a loss, So why aren't companies rushing to build 40nm commodity mcu's like infineon did ? it should cost almost nothing to make an mcu that way.

flyinglizard 11y ago
I don't know - I believe the barrier is in the embedded flash technology used (NOR flashes that are fast/reliable/big enough).