Comment by jauntywundrkind 19 hours ago
That folks are still making new Cortex-A7 (2011) designs is wild. A-35 doesn't seem to be very popular or better.
Cortex-M33 (2016) derives–as you allude to–from ARMv8 (2015). But yeah it barely seems only barely popular, even now.
Having witnessed some of the 9p's & aughts computing, I never in a million years would have guessed microcontrollers & power efficient small chips would see so little change across a decade of time!!
Some chips that have come out in the past 3 years with Cortex A7:
Microchip SAMA7D65 and SAMA7G54. Allwinner V853 and T113-S3.
It's not like a massive stream of A7's. But even pretty big players don't really seem to have any competitive options to try. The A-35 has some adoption. There is an A34 and A32 that I don't see much of, don't know what they'd bring above the A7. All over a decade old now and barely seen.
To be fair, just this year ARM announced Cortex-A320 which I don't know much about, but might perhaps be a viable new low power chip.
It’s a serious problem for their lawyers, primarily: https://olimex.wordpress.com/2015/11/09/chinese-clones-attac...
cheap Chinese clones of US IP is a broader problem that affects more than just STM
There are no similarities between Cortex-M7 and Cortex-A7 from the POV of obsolescence.
Cortex-M7 belongs to the biggest-size class of ARM-based microcontrollers. There is one newer replacement for it, Cortex-M85, but for now Cortex-M7 is not completely obsolete, because it is available in various configurations from much more vendors and at lower prices than Cortex-M85.
Cortex-M7 and its successor Cortex-M85 have similar die sizes and instructions-per-clock performance with the Cortex-R8x and Cortex-A5xx cores (Cortex-M5x, Cortex-R5x and Cortex-A3x are smaller and slower cores), but while the Cortex-M8x and Cortex-R8x cores have short instruction pipelines, suitable for maximum clock frequencies around 1 GHz, the Cortex-A5xx cores have longer instruction pipelines, suitable for maximum clock frequencies around 2 GHz (allowing greater throughput, but also greater worst-case latency).
Unlike Cortex-M7, Cortex-A7 is really completely obsolete. It has been succeeded by Cortex-A53, then by Cortex-A55, then by Cortex-A510, then by Cortex-A520.
For now, Cortex-A55 is the most frequently used among this class of cores and both Cortex-A7 and Cortex-A53 are truly completely obsolete.
Even Cortex-A55 should have been obsolete by now, but the inertia in embedded computers is great, so it will remain for some time the choice for cheap embedded computers where the price of the complete computer must be well under $50 (above that price Cortex-A7x or Intel Atom cores become preferable).
In embedded old Cortex-A53 (and A72) are the most "new" cores used compared to A9 (and A15). E.g. TI AMxxxx [1] and Xilinx UltraScale+ vs Zynq.
[1] https://www.ti.com/microcontrollers-mcus-processors/arm-base...
For cores included in FPGAs, sadly there are none better than Cortex-A53 and Cortex-A72, because there have been no significant upgrades to the families of bigger FPGAs for many years. However in that case you buy the chip mainly for the FPGA and you have to be content with whatever CPU core happens to be included.
On the other hand, for the CPUs intended for cheap embedded computers there are a very large number of companies that offer products with Cortex-A55, or with Cortex-A76 or Cortex-A78, so there is no reason to accept anything older than that.
Texas Instruments is not really representative for embedded microcontrollers or computers, because everything that it offers is based on exceedingly obsolete cores.
Even if we ignore the Chinese companies, which usually have more up-to-date products, there are other companies, like Renesas and NXP, or for smaller microcontrollers Infineon and ST, all of which offer much less ancient chips than TI.
Unfortunately, the US-based companies that are active in the embedded ARM-based computer segment have clearly the most obsolete lines of products, with the exception of NVIDIA and Qualcomm, which however target only the higher end of the automotive and embedded markets, by having expensive products. If you want something made or at least designed in USA, embedded computers with Intel Atom CPUs are likely to be a better choice than something with an archaic ARM core.
For the Intel Atom cores, Gracemont has similar performance to Cortex-A78, Tremont to Cortex-A76 and Goldmont Plus to Cortex-A75; moreover, unlike the CPUs based on Cortex-A78, which are either scarce or expensive (like Qualcomm QCM6490 or NVIDIA Orin), the CPUs based on Gracemont, i.e. Amston Lake (Atom x7000 series) or Twin Lake (N?50 series), are both cheap and ubiquitous.
The latest Cortex-A7xx cores that implement the Armv9-A ISA are better than any Intel Atom core, but for now they are available only in smartphones from 2022 or more recent or in some servers, not in embedded computers (with the exception of a product with Cortex-A720 offered by some obscure Chinese company).
In general Renesas offers more modern microcontrollers than all the other vendors of MCUs, which have decreased a lot the rate of new product launches during recent years, but unfortunately they are also among the most expensive.
I have also not seen Cortex-M85 except from Renesas. Cortex-M55 is seldom an alternative to Cortex-M7, because Cortex-M55 is smaller and slower than Cortex-M7 (but faster than the old Cortex-M4 or than the newer Cortex-M33).
Cortex-M55 implements the Helium vector instruction set, so for an application that can use Helium it may match or exceed the speed of a Cortex-M7, in which case it could replace it. Cortex-M55 may also be used to upgrade an old product with Cortex-M7, if the M7 was used only because Cortex-M4 would have been too slow, but the full speed of Cortex-M7 was not really needed.
> I never in a million years would have guessed microcontrollers & power efficient small chips would see so little change across a decade of time
It's because the software ecosystem around them is so incredibly lousy and painful.
Once you get something embedded to work, you never want to touch it again if you can avoid it.
I was really, really, really hoping that the RISC-V folks were going to do better. Alas, the RISC-V ecosystem seems doomed to be repeating the same levels of idiocy.
Switching microcontrollers means you have a lot of work to do to redo the HW design, re-run all of your pre-production testing, update mfg/QA with new processes and tests, possibly rewrite some of your application.. and you need to price in a new part to your BOM, figure out a secure supply for some number of years... And that just assumes you don't want to do even more work to take advantage of the new chip's capabilities by rewriting even more of your code. All while your original CPU probably still does fine because this is embedded we're talking about and your product already does what it needs to do.
RISC-V is even worse: The Cortex-M series have standardized interrupt handling and are built so you can avoid writing any assembly for the startup code.
Meanwhile the RISC-V spec only defines very basic interrupt functionality, with most MCU vendors adding different external interrupt controllers or changing their cores to more closely follow the faster Cortex-M style, where the core itself handles stashing/unstashing registers, exit of interrupt handler on ret, vectoring for external interrupts, ... .
The low knowledge/priority of embedded of RISC-V can be seen in how long it took to specify an extension tha only includes multiplication, not division.
Especially for smaller MCUs the debug situation is unfortunate: In ARM-World you can use any CMSIS-DAP debug probe to debug different MCUs over SWD. RISC-V MCUs either have JTAG or a custom pin-reduced variant (as 4 pins for debugging is quite a lot) which is usually only supported by very few debug probes.
RISC-V just standardizes a whole lot less (and not sensibly for small embedded) than ARM.
Being customizable is one of RISC-V’s strengths. Multiplication can be easily done in software by doing bit shifts and addition in a loop. If an embedded application does not make heavy use of multiplication, you can omit multiplication from the silicon for cost savings.
That said, ARM’s SWD is certainly nice. It appears to be possible to debug the Hazard3 cores in the RP2350 in the same way as the ARM cores:
https://gigazine.net/gsc_news/en/20241004-raspberry-pi-pico-...
> If an embedded application does not make heavy use of multiplication, you can omit multiplication from the silicon for cost savings.
The problem was that the initial extension that included multiplication also included division[1]. A lot of small microcontrollers have multiplication hardware but not division hardware.
Thus it would make sense to have a multiplication-only extension.
IIRC the argument was that the CPU should just trap the division instructions and emulate them, but in the embedded world you'll want to know your performance envelopes so better to explicitly know if hardware division is available or not.
[1]: https://docs.openhwgroup.org/projects/cva6-user-manual/01_cv...
> It's because the software ecosystem around them is so incredibly lousy and painful.
This is reaching breaking point entirely because of how powerful modern MCUs are too. You simply cannot develop and maintain software of scale and complexity to exploit those machines using the mainstream practices of the embedded industry.
Isn’t there some dynamic at play where STM will put one of these on a board, that board becomes a “standard” and then it’s cloned by other manufacturers, lowering cost? (legality aside)
STM32H5 in 2023 (M33): https://newsroom.st.com/media-center/press-item.html/p4519.h...
GD32F5 in 2024: https://www.gigadevice.com/about/news-and-event/news/gigadev...
STM32N6 in 2025 (M55): https://blog.st.com/stm32n6/
i.e. it takes some time for new chips to hit cost targets, and most applications don’t need the latest chips?
I don’t know a lot about the market, though, and interested to learn more