WCH $0.10 USD RISC-V MCU

[tim_]

Problem is the STM8S is a wider vcc part, so the STM32C has no hope to kill any 5V uses !!

Do you still see that many use cases for 5V outside automotive? Anything powered by a battery will be 3.3V or lower. There is also not much else that would be only available in 5V, but not in 3.3V. Ok, if you directly want to drive white LEDs, 3.3V is maybe somewhat close.

What I find more annoying is that the lower limit is 2V, which degrades low power performance. (Ok, we have STM32L for that...)

[josuah]

Do you still see that many use cases for 5V outside automotive? Anything powered by a battery will be 3.3V or lower. There is also not much else that would be only available in 5V, but not in 3.3V. Ok, if you directly want to drive white LEDs, 3.3V is maybe somewhat close.

USB powered devices? Even if they lack USB data connection, I sometimes see a Type-C connector as power plug for all sort of devices that are not computer peripheral.

[GromBeestje]

USB powered devices? Even if they lack USB data connection, I sometimes see a Type-C connector as power plug for all sort of devices that are not computer peripheral.

I'd say *only* if they lack a USB data connection, as USB's I/O voltage is at 3.3 Volt.

[SiliconWizard]

Do you still see that many use cases for 5V outside automotive? Anything powered by a battery will be 3.3V or lower. There is also not much else that would be only available in 5V, but not in 3.3V. Ok, if you directly want to drive white LEDs, 3.3V is maybe somewhat close.

USB powered devices? Even if they lack USB data connection, I sometimes see a Type-C connector as power plug for all sort of devices that are not computer peripheral.

So because USB provides a 5V supply you need to use 5V logic?

According to https://en.wikipedia.org/wiki/USB#Power (and backed by experience), VBUS may be as low as 4.4V in some situations (and that's not counting cables), so many strictly 5V parts may actually not function properly if directly powered by VBUS. One exception would be those ATMEGA MCUs, which, while many people take them as "5V devices", actually operate on a pretty wide range of 2.7V to 5.5V (at least if you restrict the clock to 8MHz), so VBUS should not be a problem as long as you don't use any other part that would require a tighter tolerance for the 5V, and as long as you don't use VBUS as any kind of reference voltage either.

[DavidAlfa]

Hell, let's get back to 15V logic then!
From my point of view, the 5V argument seems a bit silly.

[PCB.Wiz]

Problem is the STM8S is a wider vcc part, so the STM32C has no hope to kill any 5V uses !!

Do you still see that many use cases for 5V outside automotive?

It is not me making all those new wide Vcc parts, it is the semiconductor suppliers, and they must know the targets.
Also note that CMOS LOGIC highest volume parts, are the wide Vcc 1.8~5.5V 74AHCxxx etc
Like CMOS logic, small MCUs are now commodity parts, and buyers and distributors want flexibility.

Atmel tried releasing a 3v3 only AVR and that is now relegated to a niche part, (ie it flopped) and their latest AVR's released are wide Vcc ones.

Anything powered by a battery will be 3.3V or lower.

? There are many 3.7~4.2V lithium battery packs

There is also not much else that would be only available in 5V, but not in 3.3V.
Ok, if you directly want to drive white LEDs, 3.3V is maybe somewhat close.

Exactly, a 3v3 part simply lacks the headroom to drive White/Blue high threshold LEDs, as well as the common Serial RGB leds.
It relegates them to a niche use, they are not general purpose use parts.

For larger MCU's, > 48 pins and >> 100MHz and  > 500k Flash, then 3v3 (and multiple power voltages) are more common, and makes sense thermally.
- but low pin count, bottom end, commodity MCUs - nah.

What I find more annoying is that the lower limit is 2V, which degrades low power performance. (Ok, we have STM32L for that...)

Yup, see how you now have to pick from many families....

[ralphd]

STM just released their new low-cost STM32C MCU series. For reference, it seems that the die size of the STM32C011D4 is 2.4mm². You can read this directly from the size of the WLCSP package. The device has 32kb of flash, so I assume that the 16kb STM32C011J4 device could be around 2.xmm², very comparable to the CH32V003. We have to take into account that they most likely use slightly different technologies. The STM32C is in 90nm, I beliebve the CH32V003 to be 130nm. Nevertheless, they seem to be quite comparable in area and hence cost.

A few months ago Digitimes had an article about TSMC wafer prices, and pegged 90nm at $2000.  Not sure if that was 200 or 300mm, but I'll assume the worst and say 200.  Smaller manufacturers will be cheaper, probably around $1600 now.  That puts the cost of a 2mm^2 chip at around 10-11c.  I'd guess 130nm wafers are around $1000-1200, putting the cost of a 2mm^2 chip at about 7c.

[josuah]

I'd say *only* if they lack a USB data connection, as USB's I/O voltage is at 3.3 Volt.

So because USB provides a 5V supply you need to use 5V logic?

I figured my mistake too late, I misunderstood (or maybe it was implicit) that it was referring to the I/O.
It looks like, implicitly, "it does 5V" refers to I/O whereas "it can live off 5V" revers to power supply, and "it can get 5V" might mean 5V-tolerant I/O or something like that.
I'll get used to it.

[JPortici]

There is also not much else that would be only available in 5V, but not in 3.3V.
Ok, if you directly want to drive white LEDs, 3.3V is maybe somewhat close.

Exactly, a 3v3 part simply lacks the headroom to drive White/Blue high threshold LEDs, as well as the common Serial RGB leds.
It relegates them to a niche use, they are not general purpose use parts.

Also direct drive of higher voltage mosfets! once you pick VDS > 100V it's really difficult do find mosfets that wil switch reliably with 3V3 at the gate. I wanted to migrate a project to a newer "better" (bigger, more pins) mcu but i had no board space nor BOM margin for all the then required mosfet drivers

[tim_]

A few months ago Digitimes had an article about TSMC wafer prices, and pegged 90nm at $2000.  Not sure if that was 200 or 300mm, but I'll assume the worst and say 200.  Smaller manufacturers will be cheaper, probably around $1600 now.  That puts the cost of a 2mm^2 chip at around 10-11c.  I'd guess 130nm wafers are around $1000-1200, putting the cost of a 2mm^2 chip at about 7c.

These numbers are off. 90nm is 300mm only, 130nm is almost always 300mm.

You also need to take into account that the actual wafer cost depends on the number of process features being used in a specific product. To reduce product cost, it is often possible to use simplified processes with reduced mask coutn to reduce wafer cost. For example, small digital designs may need less interconnection layers. Design without demanding analog or power do not need as many different front end devices.

[brucehoult]

Plus, at these die sizes costs are being dominated by cutting up the wafer, packaging, and testing.

The area is also getting dominated by the I/O pads, so even though the active circuit will be smaller (and cheaper) at 90 nm than at 130 nm, the total area may not change much if at all -- this is the main reason it can be just as cheap to make a 32 bit microcontroller as an 8 bit one.

[SiliconWizard]

Plus, at these die sizes costs are being dominated by cutting up the wafer, packaging, and testing.

The area is also getting dominated by the I/O pads, so even though the active circuit will be smaller (and cheaper) at 90 nm than at 130 nm, the total area may not change much if at all -- this is the main reason it can be just as cheap to make a 32 bit microcontroller as an 8 bit one.

Yes, something we have pointed out quite often. Apart from MCUs which natively (meaning, on the die itself, and not just as a package variant with most GPIOs unconnected) have a very small number of GPIOs, they are usually pad-limited. The exception will come from the amount of internal flash/RAM rather than the core itself, but small, low-cost MCUs usually have a very small amount of internal memory, so that's not a factor either.

[brucehoult]

Dave just reviewed it.

https://www.eevblog.com/forum/blog/eevblog-1524-the-10-cent-risc-v-processor-ch32v003/

[tim_]

Plus, at these die sizes costs are being dominated by cutting up the wafer, packaging, and testing.

The area is also getting dominated by the I/O pads, so even though the active circuit will be smaller (and cheaper) at 90 nm than at 130 nm, the total area may not change much if at all -- this is the main reason it can be just as cheap to make a 32 bit microcontroller as an 8 bit one.

The dicing, packaging (bond wires!) and test cost usually also scales with area of a die through various direct and indirect scaling relationships. So, comparing die sizes is quite a good first order indicator. Of course there are outliers.

I/O limitation is less of an issue when you use circuit-under-pad. We have seen that with the Padauk MCUs, and possibly this is also used by WCH. I have seen some stray test pads on the die.

[Adi]

Sorry if this is a little off topic ... But just to show that 0.10$ chip can perform decently

"CH32V003 controlling ILI9341 over SPI with DMA at 24 MHz"

[josuah]

I have seen more expensive controllers failing where this one did succeed!

Did it have enough memory for a full framebuffer for drawing?
On another MCU, we had to generate the output line by line (buffer of 400 rows * 640 pixel * 2 bytes per pixel (YUV422) => 500 kB of RAM needed, only 512 kB on the MCU).

[brucehoult]

Did it have enough memory for a full framebuffer for drawing?

Not even close.

The VDC has 172800 bytes. The RISC-V chip has 2 KB.

[brucehoult]

Good catch. Braino.

[Adi]

I have seen more expensive controllers failing where this one did succeed!

Did it have enough memory for a full framebuffer for drawing?
On another MCU, we had to generate the output line by line (buffer of 400 rows * 640 pixel * 2 bytes per pixel (YUV422) => 500 kB of RAM needed, only 512 kB on the MCU).

Only small buffer(512B)is used for single character to send it
with DMA.

[betocool]

I watched Dave's video with the 10c chip yesterday. It reminded me heavily of STM32IDE. Seems like a low learning curve if you've done STM32's the last 10 years. I was also pleasantly surprised to see they have a Linux edition.

I'm wondering though, that USB - Serial converter he uses, is it just USB - Serial or USB - Serial + Some debug interface?

I'm also a bit wary of having that thing running on my main machine... but perhaps on a VM or a secondary PC? And since it's all GNU-Risc-GCC, perhaps even just have a Makefile and see how that goes.

Cheers,

Alberto

[thm_w]

I watched Dave's video with the 10c chip yesterday. It reminded me heavily of STM32IDE. Seems like a low learning curve if you've done STM32's the last 10 years. I was also pleasantly surprised to see they have a Linux edition.

I'm wondering though, that USB - Serial converter he uses, is it just USB - Serial or USB - Serial + Some debug interface?

I'm also a bit wary of having that thing running on my main machine... but perhaps on a VM or a secondary PC? And since it's all GNU-Risc-GCC, perhaps even just have a Makefile and see how that goes.

Cheers,

Alberto

Its discussed in the thread: https://www.eevblog.com/forum/blog/eevblog-1524-the-10-cent-risc-v-processor-ch32v003/

[MT]

Risc V is fodder for horses and has horse like speed and strength, says WCH Patrick.

[SiliconWizard]

 

[ataradov]

They must have some westerner on staff. They should really let them proofread stuff before publishing gems like this.

[SiliconWizard]

Chinese marketing is different. That's funny to us but it's just a different culture. It doesn't translate very well, but not sure they even care. I don't think they are running after western clients for their MCUs. So we take them because they're dirt cheap and we basically shut up.