CH32V003 Memory Retention

[Mike Warren]

I have been looking at the CH32V003 for a simple, very price conscious product.

I discovered this in the datasheet:



I wrote this to WCH:

I notice the CH32V003 datasheet says the the data retention is 10 years. Does that mean any product designed with this chip will likely start to fail after 10 years?

and got this reply:

Thanks for your support WCH.

“ 10 year”

This is the theoretical value of the internal flash of the chip.

Do you feel that this time is short?

 By the way, what product you are planning to design?

Hope my answer can help you.

Looking forwards your reply.

That seem like a very short time to me.

[hans]

Without any parameters it is quite an useless spec to work from. Flash Lifetime depends a lot on ambient temperature, and also accelerates with program/erase cycles.

So if without any spec I have to assume this is at 1 P/E cycle and Ta=25C, then that doesn't look great.
If its at 10K P/E and Ta=85C, then that looks pretty good.

Their answer is not very insightful unfortunately. Maybe you can ask them what the parametric conditions were for this theoretically derived/simulated value?

[voltsandjolts]

Quick look at the cheap-ish STM32L051 which I commonly use and it's 30 year retention minimum after 10k cycles at 85degC.

[HwAoRrDk]

The introduction to the datasheet's chapter 3 says the following:

All minimum and maximum values are guaranteed in the worst conditions of ambient temperature, supply voltage and clock frequency. Typical values are based on normal temperature (25℃) and VDD = 3.3V or 5V environment, which are given only as design guidelines.

The flash data retention period is specified as a minimum with no specific conditions stated, so therefore given the above, one must presume the retention period is at ambient temperature of 85°C (maximum operating temperature specified in table 3-1).

I found one source claiming a rule of thumb that flash memory data retention time halves for every 10°C increase in temperature. So conversely, if it doubles from a worst-case figure for every 10°C decrease (which I would assume it doesn't actually - it's probably a curve), then you might be looking at 30, 40, 50 years if your device never experiences temperatures greater than 30-40°C.

So, unless your device will be experiencing high temperatures, I wouldn't worry about it.

Besides, if you're aiming to make a "very price conscious" product, then there are always compromises. If you want a dirt-cheap microcontroller, you can't expect to get class-leading flash endurance or have its characteristics qualified to the nth degree. And do you really expect your cheapo product to still be in use 30 years from now?

[smokeless0864]

After 1year of uptime you could overwrite flash internally, it would last 100years+.

[SiliconWizard]

I doubt they have actually done any qualification on these figures, so they must come from conservative theoretical figures given by the process they use. I also doubt they have flash that is fundamentally different from what most other vendors use. What would be interesting would be to know the process node - if anyone has this information.

[HwAoRrDk]

What would be interesting would be to know the process node - if anyone has this information.

Zeptobars' estimate from examination of a die was around 90nm.

[Mike Warren]

Quick look at the cheap-ish STM32L051 which I commonly use and it's 30 year retention minimum after 10k cycles at 85degC.

Yes, I expect way more than 10 years based on other chips I use.

ATTiny44A and similar chips: Data Retention: 20 years at 85°C / 100 years at 25°C

The flash data retention period is specified as a minimum with no specific conditions stated, so therefore given the above, one must presume the retention period is at ambient temperature of 85°C (maximum operating temperature specified in table 3-1).

That would be fine, but not having it explicitly stated and the weak answer from WCH does have me concerned.

And do you really expect your cheapo product to still be in use 30 years from now?

Probably not, but it is a widget, so it is possible. It's not a complex product. I have always tried to design products for a long lifetime

[hans]

I agree with SiliconWizard.
I wouldn't be surprised if this spec is taken from another datasheet, e.g. their process foundry.

Cheap parts usually skimp on analog specs. Sure FLASH wear time is as much analog as it can be. But I mean more on IC design: highly linear 12-bit A/Ds, low power internal oscillators/PLLs, proper sleep modes, lots of power gating, etc. It also wasn't that long ago we saw MCU clones with a separate FLASH die bonded in.

ST have their own fabs so they control the whole development process of their PDK. But I also heard from engineers at other microcontroller manufacturers that they fine tune their own PDK.

I'm not sure how much work is being done from upcoming manufacturers, especially if they target cheap low-end parts. My impression is they have some catching up to do.