Smallest Possible Feature ?

Question

I can't sketch anything smaller than about 100 nanometers. Is it normal ? As I wrote in an old thread "Modelling space size" :

The format for floating points used in SW is the C
>"double" format, which guarantees 15 decimal digits of precision. This means
>the largest and smallest dimensions that can be added and still give a
>correct result must not have a ratio larger than 1/10E15. If you want to
>measure nanometers (1E-9 m) and still have some margin for geometric
>operations, you can't allow dimensions much larger than 1E3 m.

So I thought I could "easily" design sub-micron features, especially when setting microns, nanometers or angstroms in the document options, but it seems the workspace is "fixed" to a few cubic kilometers, limiting the resolution at the bottom of the scale... What limits sketching to 100 nanos while numbers would allow to go down to about 0.2 nanos ? What's the use of using angstroms as units if you can't use them ? Did I miss something ? For those who ask why I care : some swiss watchmakers model gears with a sub-atomic precision. The fact that atoms can't be cut in two flat pieces is considered as a machining error... Philippe Guglielmetti -

formatting link

Corey Scheich · Accepted Answer

When SW goes 64 bit you will have that precision if it ever does.Corey

P. · Answer

Philippe,There are upper and lower limits. Even working at the size of a pin head SW can be a bit awkward. It isn't just the ratio inherent in C doubles. SW has certain tolerances built in to help it decide when elements are or aren't coincident, etc. These would seem to be much larger than a few nanos. Take the .005 inch minimum involved in certain sweep geometry.Philippe Guglielmetti wrote:

Philippe Guglielmetti · Answer

no Corey, sorry, it won't change anything. Doubles are already 64 bits (see

formatting link

) but with 52 bits mantissa. The only way to extend precsion would be to move to "extended" formats. AFAIK the 80 bits format with

64 bits mantissa is supported by Intel hardware, but it would push the limit "only" by a factor 1000. Don' expect too much from 64bits procs : few geometry or CAD related algorithm (if any) require manipulating intergers larger than 32 bits (about 4'000'000'000) Philippe Guglielmetti -

formatting link

Jerry Steiger · Answer

Where does that one show up? I haven't seen it yet (that I know of) and weuse sweeps a lot.Jerry SteigerTripod Data Systems"take the garbage out, dear"

P · Answer

If you search back on the newsgroup there was a discussion long ago (99, 2000).

When you do a sweep that just skins an existing face you want to embed the sweep at least .005 into the face. You can find some discussion in the KB. Also if your generatrix has a construction line to a directrix I think there is a minimum length for that construction line of about .005".

There are other examples of tolerances, for example the 3 face / 1 face behavior on an extrude protruding from a flat face. If the extrude protrudes just a very, very small amount and approaches being coplanar with the existing face, then it depends on which side you approach the existing face whether you will get one face or three when coplanarity is achieved. I would probably have to post the example to show this one.

Jerry Steiger · Answer

Ah, that brings back some very old, (and, unfortunately, very vague)memories. Thanks!Jerry SteigerTripod Data Systems"take the garbage out, dear"

Jeff Howard · Answer

Don't "know" a thing about it (fuzzy stuff to me); just some general bs and curiosity questions:

I'm sorta surprised you can get something as small as 1 E-7. Somewhere in there there's an absolute tolerance below which two points are considered to be coincident. I'd have figured this is 1 E-5 to -6.

(What were model units? Does SW scale absolute resolution or set it to match model units; e.g. 1 E-x meters, 1 E-x mm or is it a static value?)

Probably can't get the full 15 significant digits of accuracy. ACIS uses

10 digits, lopping off 4 to 5 digits as for "noise", round off errors, etc. and 1 more to provide a buffer over absolute resolution. This range can supposedly be shifted, but not expanded. You can, in fact, specifiy values outside of the default 1 E-5 to 1 E4 range, but shouldn't be surprised if problems are encountered.

Somewhere up above someone mentioned .005 for swept features (?). Polynomial surface calculations are never as accurate as analytic surface calcs, and .005 is probably a good buffer. Surface modeling functions are sometimes more appropriate vs. a "solid" function that's trying to make a swoopy face coincident with another one.

-----------------------

Smallest Possible Feature ?

Join the Discussion

Didn't find your answer?