Component layout on PCB's

[cksa]

I've been doing my component layout for my PCB in Eagle on 1mm grids. I now notice that Sparkfun uses 0.05 inch grids.

Is there a difference? Will board houses readily accept designs made on 1mm grids (I can't see why not, but I'd like to make sure)?

[EEVblog]

It does not matter, you can use no grid at all if you want (bad idea though), it makes no difference to the board maker.
What grid you use will depend on what type of components you are using and the design requirements.
Sparkfun would use an imperial grid because they design a lot with through hole components and connectors which are mostly still 0.1" pitch or multiples of.

Dave.

[RayJones]

Yep, imperial grids are far more common due to a lot of components still conforming to sub multiples of inch spacings. eg 0.1 for DIP, 0.05 for SOIC etc.

It is always amusing to have your data sheet tell you the pin spacing is 1.27mm, which is really 50thou (mil).
The imperial measurement is of course missing on the datasheet.

It is always worth converting oddball metric dimensions, they may be nice imperial values in disguise. 

[Zero999]

In my experience, through hole is generally imperial and SMT is more often metric so choose a grid which fits most of the components on the board.

I once used 1.25mm pin spacing for DILs until I needed to fit a 40 pin socket which gave me a few problems, fortunately it was fixed by drilling the holes slightly larger.

[jahonen]

Whatever coordinate system is used in PCB design, one must ensure that it is maintained through the process. That means if you use mils, then your gerbers and drill files should be in inches. I use grid when placing the components but PADS router works best without grid.

All newer (TQFP, BGA, QFN etc.) component cases are in metric dimensions, so I usually use millimeters (I can't remember when I made the transition from mils to millimeters). Mils convert exactly to millimeters (with hefty number of decimals but anyway) but not other way around.

Regards,
Janne

[Zero999]

I find mils confusing because many people abbreviate millimetre to mil. The correct term is thou, I think mil is an Americanism.

[RayJones]

mil is standard terminology within the PCB industry, and yes it is a thou, not a millimetre.
I learnt this from the early days of Protel, and that was when they were a small fledging company in Hobart, Tasmania, well before the Americans bought them out, and Australia is allegedly a metric country (just try getting any metric bolt though at your small hardware outlets)

As for a mil being an exact multiple of millimetres, what drugs are you on? Decimal places to billy oh do not count 
I do agree that you should stick to the one type of grid throughout your design as it sets the expected track clearances consistently.

When it comes to connecting to an off grid component (eg TQFP to a 0.1" header), Protel at least allows you to pick the unrouted net at the off grid component and will snap to the centre of the pad. You can then radiate from the pad and eventually step onto the usual grid with a 45 deg transition without too much hassle.
Amother odd bal spacing involves DB9 DB25 etc connectors. They don't match either grid nicely 

As for using a metric or imperial grid, well it really is determined by the bulk of your components footprints.
If you only have one or two oddball spacings, use the other grid as it matches with the majority of the components much better and you'll have less hassle pushing tracks between pads.

[alm]

As for a mil being an exact multiple of millimetres, what drugs are you on? Decimal places to billy oh do not count 

The inch is defined as exactly 25.4mm, so a mil/thou is exactly 25.4um (=.0254mm). It's not possible to represent 1/25.4 in a reasonable number of decimals. He didn't claim it was an exact multiple, but that you can exactly represent it in the decimal system (in just three digits). Although you could argue that an exact binary fraction is more important, and neither of the conversion factors are nice binary fractions.

[RayJones]

As for a mil being an exact multiple of millimetres, what drugs are you on? Decimal places to billy oh do not count 

The inch is defined as exactly 25.4mm, so a mil/thou is exactly 25.4um (=.0254mm). It's not possible to represent 1/25.4 in a reasonable number of decimals. He didn't claim it was an exact multiple, but that you can exactly represent it in the decimal system (in just three digits). Although you could argue that an exact binary fraction is more important, and neither of the conversion factors are nice binary fractions.

If you think that is workable you too must be on drugs  25.4 um - get off the grass, were designing PCBS, not microchips!

As I said, use the grid most appropriate for your design.
To battle along with a metric grid with predominantly imperial component spacings is foolish, as is doing the opposite.
Only the designer knows what components they will be using.

Each has their purpose when used correctly.

[alm]

If you think that is workable you too must be on drugs  25.4 um - get off the grass, were designing PCBS, not microchips!

What is it with you and drugs, projection? But I'll get of your lawn with the other kids and leave you alone with your drugs .

[RayJones]

LOL, it is a common Aussie phrase for "you can't be serious?", as is "get of the grass", and if you failed to notice, the comments were associated with grin. ie just a bit of gentle ribbing.

I was simply stating practical facts and also offering alternatives, perhaps you could try offering some constructive, practical comment on the actual topic?

ie to use an absurdly high grid resolution defeats the purpose of using a grid in the first place, which is to quickly and efficient place tracks which sufficient clearances without needing to stop and think all the time.

[alm]

I know you were joking, but I felt it did get a bit repetitive, you should vary those expressions a bit more .

I wasn't suggesting that someone should use a .2um grid or something like that, but metric vs. imperial is a related choice. For example, things like Gerber and drill files can be both in mm and inch. In that case, being able to express both exactly in just a few digits is useful. I do agree that choosing the grid that corresponds to the pitch of most of your components makes the most sense (whether it's 50mil or 1.27mm), but I don't usually think that adding AOL-style me-too posts add much to the discussion.

[jahonen]

Yes, indeed I meant that 1 mil = 0.0254 mm, which is accurately represented by "reasonable" number of decimals, whereas other way around is not, 1 mm = ~39.37008 mil. Or it is exact if using fractions, 1 mm = 5000/127 mil I did not mean that would be good choice as a grid pitch, as it would not be even fine enough as demonstrated by the following example:

There is a similar problem for using imperial grid even with imperially dimensioned components if the pitch is small enough. Take a IC with 0.635 mm pitch (or 25 mils if you prefer) and even number of pins per row, now if you have the component origin at the geometrical center of the package (for the pick and place machine), then you would need a 0.3175 mm (12.5 mils) grid to route the pins using the grid. So one would fallback to decimals anyway at some point. PCB Matrix IPC-7351 PCB land pattern calculator defaults to millimeters, so "go figure".

Mentioned "optimum grid"-hassle is one of the reasons why I personally won't use a grid at all when routing, since the software I use will push away the neighborhood traces anyway according the design rules (unless I protect them, in which case it just won't let me route too close) when I route too near which would otherwise violate the rules, and traces will automatically snap into pads regardless which end I start the traces. It might be necessary to fiddle with grid with less advanced PCB software which can't do the aforementioned "plowing".

I believe the "professional" layout guys at my work do this similarly, or at least the grid is most usually completely off whenever I open the layout file for checking OTOH, I use something like 0.5 or 0.2 mm grid when placing the components, it is then easier to make them in nice rows. I never have had any problems doing that.

Regards,
Janne

[RayJones]

That is interesting about pick and place machines.

I have always set the origin of the device to be the centre of pin 1, thus always keeping the pads on grid. I'm pretty sure this is the recommended standard for the Protel component libraries (or at least it was).

Are the pick and place machines that fussy?
I always thought the use of feducial markers was far more important as you cannot rely on the board's outer dimensions for referencing the machine.

The Protel packages I have used tend to use a 20mil component placement grid by default (though I must admit have not tried their latest). This is pretty much equivalent to 0.5mm - but not exactly 

[EEVblog]

mil is standard terminology within the PCB industry, and yes it is a thou, not a millimetre.
I learnt this from the early days of Protel, and that was when they were a small fledging company in Hobart, Tasmania, well before the Americans bought them out, and Australia is allegedly a metric country (just try getting any metric bolt though at your small hardware outlets)

The Yanks didn't buy out Protel. I can assure you the original Tasmanian founder still owns and runs the show. But now they have a new name, a stock market listing, and had a habbit of gobbling up smaller overseas companies.

Dave.

[RayJones]

LOL, OK I got the American bit wrong. 
But if I recall correctly they certainly set up shop in the US at some stage in the 90's so they could acquire a bigger exposure?

At some stage I certainly got tired of needing to keep shelling out thousands more for an upgrade (with fresh bugs).
It then became better to skip versions as the incremental costs were more palatable.

I'm not ashamed to admit I still use Protel99SE as it does all I need.
This includes an 8 layer board I designed, but most are only DS.
I have used a later version at times to check a design file, and they appear to be nicer, but based on past experience the user interface / concepts have probably shifted again and another learning curve would be required. 

[Zero999]

It does annoy me when software vendors expect their users to upgrade very often.

I still use Windows XP, I don't see any point with Windows 7 which will just slow my PC down more. Heck, Windows 2000 is still fine for most applications.

[EEVblog]

LOL, OK I got the American bit wrong. 
But if I recall correctly they certainly set up shop in the US at some stage in the 90's so they could acquire a bigger exposure?

Yes, and they still have a big US office. But it's just sales and support, the US was and still is the biggest market, so a presence there was necessary to look legit to the Yanks.
Most development is done here in Sydney, with some in the Netherlands. There is a still a small token contingent in Tasmania who work on the libraries.

At some stage I certainly got tired of needing to keep shelling out thousands more for an upgrade (with fresh bugs).
It then became better to skip versions as the incremental costs were more palatable.

I'm not ashamed to admit I still use Protel99SE as it does all I need.
This includes an 8 layer board I designed, but most are only DS.
I have used a later version at times to check a design file, and they appear to be nicer, but based on past experience the user interface / concepts have probably shifted again and another learning curve would be required. 

Much of the industry is still using 99SE. It's the defacto industry standard in China for example, most unpaid of course!

Dave.

[Zero999]

I didn't know Farnel owned Eagle now.

 I've tried Eagle before but didn't get on with it. Now, I've downloaded and am having another go at and it's just as horrible as ever. I've used other CAD and SIM packages before and none of them are this horrible to use - I really find it hard not to swear! One needs to toolbar to do everything from deleting an object to moving objects and I've still not figured out how to move multiple objects simultaneously.

How long have Farnel had Eagle for?

I hope they finally do something about this, there again, if they do  there'll probably be loads of people who complain because they prefer the horrible user interface.

[TheDirty]

The interface has been cleaned up slowly.  I don't think they'll ever change the odd group copy procedure, though.

[charliex]

i rarely ever use the toolbar in eagle, just type the commands. the cut n paste thing is just different to how normal cut n paste is, its definitely whacky.

[Zero999]

Yes, I've just figured it out but I shouldn't have had to, it should have been obvious. It's still far from perfect, I'm now trying to stretch a group of wires, I'm sure I'll do it eventually.

One should draw a box to select objects and drag to move them in the same way most drawing programs made over the last 20 or so years have worked.

[David]

Does anybody here use Cadence Allegro? I've not really seen any talk of it on the forums...

[Zero999]

No, I've never heard of it.

On the Eagle front: I've got further than I had when I last used it, I've managed to draw an astable multivibrator!

Hopefully I'll manage to do the PCB.

EDIT:
Eagle has just told me that my schematic is wrong because there are missing net junctions, which I've found out are those little dots where wires join, which every other schematic editing program I know of places automatically! WTF?

[nessatse]

I am always astounded at how popular Eagle is.  Over the years I have played or worked with most of the 'free' as well as some of the $$$ professional packages like Altium and Orcad (Altium still beats all others IMHO!). Eagle fails probably in every aspect when compared to all the others. 

The UI sucks - it looks like they purposely tried to come up with the most un-intuitive design possible.
The library design is fundamentally flawed - combining footprints and schematic symbols as one is simply silly
Editing PCB tracks - well to be honest most PCB packages have their quirks here, but Eagle simply outdoes them all.  Its just bad.

The list goes on and on, but why bother?  There are far better free packages out there:- KiCad, Diptrace are two that come to mind immediately, and if you are brave - gEDA :-)