does anyone interested in 20 cent RISC-V MCU?

[legacy]

RISC-V SoC that can run Linux.
With a decent coherent cache and PCIe!
oh, and a book "see RISC-V run"

[legacy]

There are 3 types of buyers:

Now, there are 4 types of buyers

• Hobby
• prof short life production
• prof long life production
• educational and academic

 

[brucehoult]

There's been quite a lot of dreaming in this thread!

Yes, we'd all love a 1+ GHz multi-core Linux-capable RISC-V SoC for $1. That's just not going to happen in the near future.

The Kendryte 210 is a good example of what's realistic, though I'm amazed they managed to get it as cheap as $8.

The SiFive FE310 (32 bit, 320 MHz, 16 KB SRAM, 16 KB iCache, some GPIOs/PWMs) is retailing through CrowdSupply for $5 each (actually $25 for a 5-pack). You can be sure CrowdSupply are making a tidy markup on those. And those are still being made on MPWs. If the FE310 was put into volume production on dedicated full-reticule masks with a good number of wafers made then it would certainly be well under $1, and possibly approaching $0.20 -- though I think $0.20 is getting close to the cost of merely packaging the die.

That's a kind of ballpark that a 20c SoC would be in from sipeed. You could get different peripherals in there (but not a lot *more* as pins are expensive), or a bit more RAM. 64 KB? Maybe, but to be honest, probably less than 16 KB, to make room for onboard flash.

Someone mentioned CH554 at 25c. Good grief! 14 KB of flash and 1.25 KB RAM. And an 8051 CPU which is awful to program by hand, and absolutely appalling code density if you use a C compiler -- at a guess that 14 KB flash would be as useful as 4 KB (8 KB, tops) on an ARM or RISC-V unless you have an army of code ninjas writing everything by hand.

[zepan0]

It will have 16~64KB SRAM,  64~256KB Flash, and all interface that STM32F1 have.

[ebastler]

Now, there are 4 types of buyers

• Hobby
• prof short life production
• prof long life production
• educational and academic

 

But academic institutions (hopefully) want to teach their engineering students skills which are relevant in real, professional life. And they have an interest in keeping their curriculum low-maintenance, i.e. they do not want to overhaul their lab courses every other year. So their needs should be very similar to those of professionals working on designs with a long production life.

[GromBeestje]

all interface that STM32F1 have.

Do you mean it will be pin-compatible?

[FlyingDutch]

Hello,

in my humble opinion better if the price would be higher, let say 5US $, but have most of the stuff that 64-bit MAix-1 has. I mean FPU, AI booster, DSP blocks and interesting peripherals like counters (also PWM) and comunication interfaces: I2C, SPI, UARTS. The Flash size best 256 KB. I also in my work use mainly ARM-Cortex (from M3 to M7).

Best Regards

[lucazader]

If this

It will have 16~64KB SRAM,  64~256KB Flash, and all interface that STM32F1 have.

I this is the case, then thats great. The F1 series has a lot of useful general purpose peripherals.
I suppose this is quite a good market to go for, as lots of companies currently use the very cheap F103 in a lot of products.
Having the same peripherals and a possibly higher performance core for cheaper will definitely be great.

As others have said having a good debugging experience and code development tools that are useful will also go a long way to seeing wide adoption

[brucehoult]

Something worth mentioning here in case people aren't aware of it is that the price of a chip is strongly dependent on the number of pins coming out.

Obviously a bigger package with more pins and more wires to the die will cost more, but even more than that, the cost of the die itself (which is directly proportional to its area) can be dominated by the area for the pads with the MCU, SRAM, flash a relatively small blob in the middle of the die.

That's not such a big issue with a 32 bit MCU on a 180nm process (or 350 or 500 of course) such as the one below, but if you made that same design in 28nm you'd be pad-dominated and not get anywhere near as much area savings as you might think.

This of course is why so many cheap MCUs have a lot of peripherals but nowhere near enough pins to use them all at the same time, and a pinmux.

[Doctorandus_P]

I'm interested, but it does not have to be 20ct.


I've bought some 10 "Blue Pill" boards and about 20 STM32F103C8T6 chips "just for fun" to play with.
I choose this processor because I was getting a bit bored by the Atmel chips, and I very much like peripherals such as a shitload to timers, native USB, built in motor controller timer, DMA, multiple interrupt levels and other goodies.

I dislike the ESP8266 and ESP32 mainly because a serious lack of documentation. I know the're popular in the hobby market, but a 20 page leaflet called a "datasheet" is so bad I have trouble with understanding that it's become popular at all.

For prices upto EUR 2 it would be a no brainer for me. Put my STM32 chips in a drawer and focus on Risc-V. But it has to have good documentation and tools.

[thm_w]

I dislike the ESP8266 and ESP32 mainly because a serious lack of documentation. I know the're popular in the hobby market, but a 20 page leaflet called a "datasheet" is so bad I have trouble with understanding that it's become popular at all.

ESP32 datasheet (60 pages): https://www.espressif.com/sites/default/files/documentation/esp32_datasheet_en.pdf
ESP32 technical reference (669 pages): https://www.espressif.com/sites/default/files/documentation/esp32_technical_reference_manual_en.pdf

[GromBeestje]

I've mainly experience with ARM-based microcontrollers  (STM32, nRF52, PSoC4). I prefer those chips mainly because they've decent debugging capabilities (JTAG/SWD).

One thing I like about the STM32 chips is that they're pin compatible between different series (eg, you can swap an STM32F1 with an F0, F3, L0, F1) (But the new G0 series use a different pinout, I made that mistake) so you can use the same PCB for any STM32 with the same footprint.

I've only looked at the ESP8266 but not the ESP32. The ESP8266, little I/O, seems there is some binary blob running in the background. I've tried to connect to an ESP8266 using JTAG using some openocd fork I found, without success. I've tried this gdbstub I found. but that doesn't work stable. I don't like the thing.

[SiliconWizard]

@zepan0: do you already have an idea about what process node you would use, max clock frequency and approximate power consumption per MHz?

[iMo]

I would expect a 55nm TSMC process, 250300MHz clock @250uA/MHz @3.3V with all peripherals on