Float precision for Ultra-Precision applications

More
23 Feb 2025 15:36 #322370 by Leo75Wolf
Context:
I want to use Linuxcnc for a Ultra-Precision machine.
For that i need Linuxcnc to tell my drives (via Ethercat) the waypoints with high resolution.
I can get 0.5 nanometer resolution out of the Ethercat bus with the 32bit Integer format at the travels that i need. Which is sufficient.

Now as for my question:
From what i gather the path planner and everything works in single-precision floats.
Doing the math on that at 10mm and 500mm positions tells me the resolution between bits of the float number is not enough.
10mm/2^24=0.59nm
500mm/2^24 = 29.80nm

I dont want to waste that much precision in my control for no reason (plus having to convert to the scaled integer value).

Is there a way to change that to double precision floats (64bit) without having to rewrite the whole control?
Also feel free to tell me that my math is wrong^^

Leo
 

Please Log in or Create an account to join the conversation.

More
24 Feb 2025 00:03 #322419 by unknown
Wow so you're aiming for sub lithographic accuracy, you got any pictures of your machine ?
How you handling temperature, a dedicated controlled room ?
I guess this is a retro fit ?

Please Log in or Create an account to join the conversation.

  • tommylight
  • tommylight's Avatar
  • Away
  • Moderator
  • Moderator
More
24 Feb 2025 00:18 #322422 by tommylight
Replied by tommylight on topic Float precision for Ultra-Precision applications
That is below "Extreme Ultra Violet" with it's 13nm wave length.
What is the current TSMC node, 4nm ???

Please Log in or Create an account to join the conversation.

More
24 Feb 2025 01:48 #322427 by GeneRF
The feature size is still well above a nanometer, but the edge precision and positioning precision have been well below a nanometer for many years.

The numbers quoted for various generations, such as TSMC's 4 nm are pure fiction; merely a label. Real dimensions have not been used as generation names for more than 20 years.

(I spent my career in state-of-the-art lithography. Now I use LinuxCNC to build model steam engines.)

Gene

Please Log in or Create an account to join the conversation.

More
24 Feb 2025 04:08 #322432 by unknown
.5nm, isn't that just a bit smaller than 4 gold atoms ?

With the radius of a gold atom being about 0.135nm ?
The following user(s) said Thank You: tommylight

Please Log in or Create an account to join the conversation.

  • tommylight
  • tommylight's Avatar
  • Away
  • Moderator
  • Moderator
More
24 Feb 2025 04:24 #322433 by tommylight
Replied by tommylight on topic Float precision for Ultra-Precision applications

The numbers quoted for various generations, such as TSMC's 4 nm are pure fiction; merely a label. Real dimensions have not been used as generation names for more than 20 years.

The first part of that statement is absolutely correct, but the second part i am not sure it's been 20 years.
4nm is feature size now, before it was transistor size. Last i checked when 7nm was in use, the transistor size was 14nm in total, or similar.
And personally, i am still not clear on how they get usable 4nm features with 13.5nm light source (molten tin hit by two lasers, all at 50000 times per second) etc, etc, but it sure as hell is astonishing!

Please Log in or Create an account to join the conversation.

More
24 Feb 2025 07:09 - 24 Feb 2025 07:09 #322466 by Leo75Wolf
Replied by Leo75Wolf on topic Float precision for Ultra-Precision applications
Hi,

Precision of the entire machine and resolution of commands are two very very different things.

The setup and build of the machine itself are not the topic of this discussion. Please keep it that way.

PLEASE stay on topic.

Thank you
Last edit: 24 Feb 2025 07:09 by Leo75Wolf.

Please Log in or Create an account to join the conversation.

More
24 Feb 2025 08:25 #322472 by unknown
Sorry sir, didn't know you were the boss of what could and couldn't be discussed, I'll crawl back into my hole

Please Log in or Create an account to join the conversation.

  • tommylight
  • tommylight's Avatar
  • Away
  • Moderator
  • Moderator
More
24 Feb 2025 15:46 #322497 by tommylight
Replied by tommylight on topic Float precision for Ultra-Precision applications

Is there a way to change that to double precision floats (64bit) without having to rewrite the whole control?
 

This is his original question, that we did not answer, so he is right.
While waiting for someone that can answer that, you might want to have a look at the source code, it is on github:
github.com/LinuxCNC/linuxcnc
Also, these might or might not help:
forum.linuxcnc.org/38-general-linuxcnc-q...ming-milling-machine
forum.linuxcnc.org/38-general-linuxcnc-q...cision-home-switches
The following user(s) said Thank You: Leo75Wolf

Please Log in or Create an account to join the conversation.

More
24 Feb 2025 16:26 #322504 by unknown
Yeah but someone starts a topic mentioning half a nanometre it's going to raise some interest. It's getting around to the level of the CMM NIST has in the US.
Precision is interesting whether it's a machinist hitting tenths on a manual lathe, a grader operator nailing a sub roadbed to the millimetre, the construction of the bridge over the harbour in the capital of my state or the work of Brunel in the 1800s.

Please Log in or Create an account to join the conversation.

Time to create page: 0.075 seconds
Powered by Kunena Forum