Re:EU countries team up for semiconductor push
by ducktective on Wednesday December 09, 2020 @11:33AM
(#25360410)
> micro-controllers do not need advanced nodes...They don't have a lot of logic
Why MCU-bound application logic are simpler than PC apps? If that is the case, we shouldn't have seen improvement in embedded MCUs in 20 years, yet every year the clocks are increasing, the flash/RAM size is doubling, ... I have a theory/fear that in future, the line between embedded-dev and web-dev would be blurred...
Re:EU countries team up for semiconductor push
by baybal2 on Wednesday December 09, 2020 @12:18PM
(#25360974)
Given how eagerly all those GCP/AWS/AZUR "IOT frameworks" run out of memory on the most beefy microcontrollers a mortal can buy, I'm afraid, you are right.
Re:EU countries team up for semiconductor push
by petra on Wednesday December 09, 2020 @12:45PM
(#25361309)
As long as mcu's with less ram/flash will be sold for less(which is standard business strategy to optimize revenue), in high volume designs people will use c/c++ to optimize.
As for low volume designs, if reliability is important, rust would be a good candidate.
Re:EU countries team up for semiconductor push
by petra on Wednesday December 09, 2020 @01:09PM
(#25361588)
ST does 22nm FD-SOI at globalfoundries , not their own fab.
The thing is, micro-controllers do not need advanced nodes. They don't have a lot of logic (it's all relative, but compared to a PC CPU or smartphone SoC), so would be "pad limited" on advanced nodes: the area of a chip depends on the logic (and memory), but also on the I/O pins or pads, which take area. So if you used a too advanced node for a chip with (relatively) little logic and still enough I/O pads, some die area would be wasted as the minimum area would come from the pads. It's a situation you want to avoid, as there's no point paying for a more costly finer nodes (with more leakage) to waste it by not using the silicon area.
Another factor is that micro-controllers also embed Flash on their die, to have a very integrated solution with logic, RAM and Flash storage on a single die. This puts a limit on the node density. Flash is OK down to 40nm. Then from then one must move to more advanced, and still young and costly, alternatives (MRAM).
So if your business is driven by micro-controllers, there's no real point in pushing toward node density ("low" nodes). ST does process optimization as everyone, but at "medium" nodes best suited for µC. There's no business drive for them to chase TSMC.
It's similar for other Eurpeans silicon makers: they tend to be on specific segments where going for "low" nodes like 5nm simply doesn't make sense. The relatively advanced fabs in Europe are from Intel and Global Foundries (both US). And GloFo stopped the race to advanced node at 14nm, as it was too expensive.