Semiconductor technology progress for car controllers (ABS, ESP, ....)?

[Simas]

Hi!

How does progress in semiconductor transistors such as Si to SiC or GaN, MOSFET to FinFET, 10nm to 2nm, etc. improves the performance of a controller such as the ABS, ESP, ESC, ... of a car?
And how much is the impact of transistor technology progress on such a controller (compared to progress of mechanical components)?

My question might be very broad, I am just trying to understand with some quantification the impact of semiconductor progress, hence any quantitative example (even non-car related) would be very helpful.

Thank you

 

[anorlunda]

I think recent experience has car manufacturers going in the other direction. There has been a shortage of CPU chips after the pandemic. Some car makers were unable to deliver new cars. Other makers continued deliveries becasuse.

1. The design used non-custom chips. Sometimes the same chips we use in laptops.
2. The makers were able to switch to other semiconductor suppliers, and/or other designs, during the shortages.

So at least in the short term, car makers are more interested in security of the supply chain than in the performance of the chips. Other industries, such as defense and porn, can make better use of cutting edge technology.

Notably, Tesla invested big in advanced technology for training neural networks. But the training happens at the factory, not in the cars on the road.

https://electrek.co/2021/08/20/tesl...worlds-new-most-powerful-ai-training-machine/

 

[Borek]

I don't think it does at all. The only thing that is really important when it comes to controller is how fast it can deal with the data - and necessary processing powers were already available well over 20 years ago.

New technologies can make the chip production cheaper, can open new areas that require more processing power, can - as @anorlunda wrote - help secure supply chains, but they don't change anything when it comes to the basic principles of the way controllers you mention work.

 

[Rive]

semiconductor transistors such as Si to SiC or GaN

This affects the capabilities and performance of power electronics in general, not the 'controllers': not directly

MOSFET to FinFET, 10nm to 2nm

Requirements for electronics used in cars are stricter than usual and also, proven designs are more preferred, so these kind of things are slow to appear in the automotive industry, and especially slow to appear in the really sensitive parts like 'controllers' (likely, you mean ECUs). For example the Tesla autopilot hardware available from 2019 was manufactured on a process introduced in 2013, while the best available that time was already two generations ahead.

For ECUs this gap usually even wider. SPC5 still produced on 90 to 40nm, and that one still counts as a completely relevant chip, as far as I know...
It's safe to say that 3-5 generation distance is completely normal, and this also shows that for this purpose you don't actually need all that computing power what chips made on new processes can offer.

Also this means, that since the available FAB capacity for these kind of devices is already not in the range where new investments usually happens, the supply capacity is fixed and lagging behind the growing demand by several generations.

 

[Simas]

Thank you very much anorlunda, Borek and Rive!

 

[Windadct]

SiC is being used in the powertrain dive inverters in many or most large scale truck and passenger vehicle cases. It allows roughly 5x reduction in losses - which reduces size, weight, heat load etc.

SiC is also on the order of 3 to 5 x more expensive "per amp" of capacity, and has other technical challenges relative to the more common ( at this power lexel) IGBT.

But the power levels here are remarkable - a six pack (inverter) module about the size of half a sheet of paper can handle 600 to 700 KW. Previous IGBT gen product for this scale would be about 200KW

 

[Simas]

Thank you Windadct for the quantified examples!