All those RP2040 hacks are impressive (HDMI, etc) but also suggest that the RP2040 is just slightly underpowered for many practical applications. I hope some cheap RISC-V MCU unbound by ARM licensing constraints (with Atomic ops+SIMD, unlike Cortex-M0) comes soon.
> currently the cheapest alternative on Mouser is ~$5 (STM32F730V8T6TR). And that will have less flash!
Yes and no: with the STM32 you don't necessarily need an external flash chip (= significantly less PCB space consumed) and you get a lot more integrated high speed peripherals than the RP2040 has, including 50MHz SDMMC, 50Mbit/s SPIs, 27Mbit/s USARTs, and proper 480Mbit/s Hi-Speed USB 2.0. The STM32 is also significantly faster (216MHz), uses a Cortex-M7 instead of a Cortex-M0+, has a memory protection unit, and so on. The RP2040 is a toy in comparison.
It really depends on what you need for your project. Sometimes that weird RP2040 can be the perfect choice for a project, sometimes an even cheaper STM32 could be an even better choice, and sometimes you need a powerful chip like e.g. that STM32F730V8T6TR.
TI's new line of MSPM0 50-cent (in bulk) or less chips are pretty interesting for Cortex M0+ cost-optimized projects.
Though IMO, the most interesting chips here are the Cortex-M22 or Cortex-M32 chips (from everybody). These chips have outstanding features at under 5mA of power usage, albeit more expensive than your typical M0+ chip.
M0+ like RP2040 is always about cost-optimization. But capabilities wise, there's all sorts of other processors that simply does more in smaller power envelopes.
----------
Thinking things like Nordic Semiconductor's nRF5340.
Also, good luck GETTING an STM32H7/F7. I've had an order outstanding for ~13 months, expecting parts in 3 more moths. All this while i was able to buy RPO2040 trivially. There is something to be said for actually being able to GET the chip, even if it is not the perfect tool for your job.
This is using PIO not ARM (zero CPU time), PIO have whole 9 instructions and 32 instruction words shared between clusters of four PIOs https://blues.io/blog/raspberry-pi-pico-pio/
The RP2040 is definitely powerful compared to other controllers, so much that it is bordering "general computing" applications. It feels like it is one push away from being able to run e.g. ucLinux which would put it in another category entirely.
Today, there are "computers" that are being used as $0.05 timers running at 32kHz clock. RP2040 is aiming at the $1.5ish price point and is somewhat appropriate for the task. (I think that most of the Cortex-M32, M22, and M0+ competitors are a bit more pragmatic with regards to more capabilities in a single chip... but RP2040's niche is external flash, USB and a few other tidbits that hobbyists like)
-----------
"General Purpose Compute" is like $5 to $15. ESP32 still isn't quite there IMO despite being larger than RP2040. I'm thinking you want something like TI's AM335x or NXP's i.MX6 processors.
IMO anyway. But every price point from 5-cents to $15 has an option. The embedded world is extremely competitive. There will be that $6 processor with exactly the features you need to perform exactly the task you want. And a different processor at $3.50 that solves a different niche, and another processor at $11. Etc. etc.
> ucLinux which would put it in another category entirely.
A general purpose compute processor should support external RAM and external Flash for maximum flexibility for different tasks. That way, one die can be used in many different projects.
Cost-optimized processors, like what the RP2040 and its bretherin in the Cortex-M0+ cost tier, are here for people who "can't afford" an external memory interface. Either due to power-constraints (external RAM costs more power) or cost-constraints (straight up dollars).
The minute we hit $5+, we start seeing external memory interfaces supporting older 128MB DDR3 RAM and such and compute-capabilities per dollar shoots up dramatically. There's no point trying to compete against the $5 market with $1.50 chips. And forget uCLinux, these babies just run full on Linux (albeit with far less RAM/compute than a normal computer...)
-------
Like: 50-cents of NOR-flash just doesn't get you as far as a microSD card sorta thing. RP2040 is still on the "low-cost" side of the efficiency curve, if anyone cared about compute performance they'd spend $5ish to $10ish more dollars.
10Mbit Ethernet isn't really that far outside the capabilities of today's microcontrollers. In fact, a lot of these uCs regularly push out 10Mbit SPI aka 10MHz.
Its still impressive of course, for the level of effort to properly create Ethernet frames. But there's a reason why various Ethernet protocols are creeping up upon CANbus. People are realizing that these cheap $1 to $5 uCs actually have the compute power for 10Mbit Ethernet, and can talk full-on socket programming.
I don't know when this realization started exactly, but more-and-more "low power Ethernet" options appear on the market, hoping to dethrone CANbus.
Technically all of these protocols are bit-banging once you get low enough in the stack -- it's just that the PIO is particularly well tuned for arbitrary protocol implementation.
FWIW, I believe there is a real bit-banged ethernet stack on Atmel.
It appears they've overclocked the RP2040 quite a bit without also increasing voltage. First time seeing that. Eg, slightly higher at 250Mhz some RP2040's have been unstable without it: https://github.com/pimoroni/picosystem/pull/57
No, the checksum just moves to UDP; in the current system, the UDP checksum is set to 0. This is not allowed in UDP/IPv6, where the checksum, calculated over both the IPv6 pseudo-header and the UDP payload, must usually be set.
