The need of silkscreen on the board during SMT assembly

[hulaomr]

Hello,

I just received a reply from an assembly shop that the assembly process is taking additional time because is difficult to place the SMT parts on the PCB if there is not silkscreen on the boards. I'm not sure this could be an excuse to get an additional couple of days or could be really an issue.

The question are

• What is the advantage of having silkscreen in the board for a run of 10-20 units of with ~1500 parts, including from 0201 up to BGA 624 pins?
  What is your experience with assembly houses regarding this situation?
  

Personally, I was expecting that the assembly drawing and the pick & place files were enough to assembly the board.

I prefer to keep the board without silkscreen at all and create documentation for further configuration and or maintenance. But if the silkscreen is really an issue during assembly of few boards, I will need to change to another direction.

Thanks

Lao

----
Below some random example of similar PCB:

[Siwastaja]

On modern high-density manufacturing, silkscreen is meaningless. It could only fit maybe 10% of the components anyway - at least in a readable way. Even if you can fit most of the designators, they end up so far away from the actual components that everything is ambiguous again. It's slow and uncertain to read, and hence, serves no purpose of "clarifying the design intent" for automated P&P.

For manual soldering of a few large components, it's useful.

On the other hand, a so called "assembly drawing" is still very much needed. It's like the silkscreen, and serves the same purpose silkscreen served some decades ago; but without the physical limitations: components are drawn as boxes, and the designators are unambiguously marked inside those boxes: you can use small enough font so that everything fits nicely. You also mark orientations in this picture. Everything fits and it looks nice.

They can print it out and do a manual check against the centroid data. This is very important because the centroid data itself is poorly standardized (actually, it's now quite well standardized, but because no one follows the standards, everything needs to be double checked manually).

[brabus]

They told you Total BS, hulaomr.

Never had any problem with silkscreen, it's just a graphic layer, ignored at best.
But wait, there's more! One should never rely on silkscreen for the assembly! Component names are all over the place.

Assembly only Needs P&P files and assembly drawing.

[Richard Crowley]

You did not mention HOW are they assembling your boards? 

If it is such a small run ("10-20 units") then are they placing components manually?  As others have mentioned, if they are using automated pick-and-place machines, then the machine doesn't care whether there is any silkscreen on the board or not.  But if they are manually placing the components, and especially if they are small ("0201") then you can see why it would speed things to have something that the human operators can see.

[ar__systems]

Many (I would say half) of SMT assembly places do not use XY files at all, the prepare placement programs from the BOM and silkscreen. (No, I'm not talking about manual assembly!) For the board that complex you should absolutely avoid places that use silkscreen for placement.

[Siwastaja]

Many (I would say half) of SMT assembly places do not use XY files at all, the prepare placement programs from the BOM and silkscreen. (No, I'm not talking about manual assembly!) For the board that complex you should absolutely avoid places that use silkscreen for placement.

This simply cannot be true except for the tiniest designs.

Maybe at some hobbyist-dedicated place who deal with super small "MCU and a LED" designs all the time, and cannot get correct centroid files anyway because beginners don't know how to produce them? Are "half" of the "SMT assembly places" really like that?

I'd expect most of the SMT assembly places need to deal with professional customers, where a 2000 component design is nothing special, and even the average would be well above 100. Such a design would be prohibitively expensive to build the way you describe it, so you must be wrong.

I have never heard about such a case. Centroid file is always in the requirements list as the most basic item.

If this indeed is the case, and you can confirm such a procedure, I think you should never ever consider using such a place and should warn others. This is like fixing an airplane with hammer only. It is possible, given enough time and skill, but it's either impossibly expensive or dangerous.

[ar__systems]

I personally don't employ assembly services often, and in my own shop I go straight from CAD file to placement file for the machine I use. But my customer, which is a several 100 mln comany, uses several assembly contractors. Guess what, I send them gerbers and BOM. Sometimes they come back and ask for XYA, sometime they don't. Just now they asked me for an XYA file for a board that's been over 2 years in production. Presumably they are changing an assembly supplier for that product

[Jeroen3]

Many (I would say half) of SMT assembly places do not use XY files at all, the prepare placement programs from the BOM and silkscreen. (No, I'm not talking about manual assembly!) For the board that complex you should absolutely avoid places that use silkscreen for placement.

This. All assembly houses we used do this if you do not use a board tool they have licensed. They manually lean the pnp/vision using the silkscreen.
Somehow they prefer not to use centroid files. I've never asked why. Maybe I should.

Judging on using Visualplace for assisted manual assembly, I can guess why. They are often rotated wrong, of just offset since the origin is at pin 1.
We often have wrongly placed 0603's on prototypes when they're placed in a grid where there is no silkscreen to indicate which pads match.

[langwadt]

Many (I would say half) of SMT assembly places do not use XY files at all, the prepare placement programs from the BOM and silkscreen. (No, I'm not talking about manual assembly!) For the board that complex you should absolutely avoid places that use silkscreen for placement.

This. All assembly houses we used do this if you do not use a board tool they have licensed. They manually lean the pnp/vision using the silkscreen.
Somehow they prefer not to use centroid files. I've never asked why. Maybe I should.

Judging on using Visualplace for assisted manual assembly, I can guess why. They are often rotated wrong, of just offset since the origin is at pin 1.
We often have wrongly placed 0603's on prototypes when they're placed in a grid where there is no silkscreen to indicate which pads match.

crazy, how many weeks is it going to take to do 1500 part manually?

[hulaomr]

On modern high-density manufacturing, silkscreen is meaningless. It could only fit maybe 10% of the components anyway - at least in a readable way. Even if you can fit most of the designators, they end up so far away from the actual components that everything is ambiguous again. It's slow and uncertain to read, and hence, serves no purpose of "clarifying the design intent" for automated P&P.

For manual soldering of a few large components, it's useful.

On the other hand, a so called "assembly drawing" is still very much needed. It's like the silkscreen, and serves the same purpose silkscreen served some decades ago; but without the physical limitations: components are drawn as boxes, and the designators are unambiguously marked inside those boxes: you can use small enough font so that everything fits nicely. You also mark orientations in this picture. Everything fits and it looks nice.

They can print it out and do a manual check against the centroid data. This is very important because the centroid data itself is poorly standardized (actually, it's now quite well standardized, but because no one follows the standards, everything needs to be double checked manually).

Exactly, I thought that silkscreen is meaningless, apply to 10% of the components and sometimes ambiguous

They told you Total BS, hulaomr.

Never had any problem with silkscreen, it's just a graphic layer, ignored at best.
But wait, there's more! One should never rely on silkscreen for the assembly! Component names are all over the place.

Assembly only Needs P&P files and assembly drawing.

I thought so, but you know, in the manufacturing arena we see too many things. As mention Siwastaja, "everything needs to be double checked manually" and they probably rely in the silkscreen to support their decisions.

You did not mention HOW are they assembling your boards? 

If it is such a small run ("10-20 units") then are they placing components manually?  As others have mentioned, if they are using automated pick-and-place machines, then the machine doesn't care whether there is any silkscreen on the board or not.  But if they are manually placing the components, and especially if they are small ("0201") then you can see why it would speed things to have something that the human operators can see.

I didn't ask then. I will. I visited the assembly line one time (before the first project). For this project with ~1000 parts and 90% in reel, I doubt they will not use the available tools.

Many (I would say half) of SMT assembly places do not use XY files at all, the prepare placement programs from the BOM and silkscreen. (No, I'm not talking about manual assembly!) For the board that complex you should absolutely avoid places that use silkscreen for placement.

This is absolutely part of my concern.  It's already too late for this project but certainly, I will request such info in the future. I hope they make their best and verify all the BGA alignment using x rays vision.

For small quantity runs, some parts are not package as expected. For example, some BGA (i.e. DDR, SOC) are not in reel, just a small cartoon box or cute tape. I was expecting they probably use a holder for their pick and place machine or perform a manual placement assisted by x ray vision. How these parts are placed at the assembly house you already work with?

Thanks all Siwastaja, brabus, Richard Crowley, ar__systems, Jeroen3, langwadt, for your input.

[mrpackethead]

I personally don't employ assembly services often, and in my own shop I go straight from CAD file to placement file for the machine I use. But my customer, which is a several 100 mln comany, uses several assembly contractors. Guess what, I send them gerbers and BOM. Sometimes they come back and ask for XYA, sometime they don't. Just now they asked me for an XYA file for a board that's been over 2 years in production. Presumably they are changing an assembly supplier for that product

Me to, but prior to having my own line i've always supplied the XY file.  In my own process i export the XY data from Alitum to a CSV file, that i run through some custom scripts to generate my machine files.       Not sure why you'd want to do it by hand?   

[ar__systems]

In another recent project they ordered the boards that came back with two pairs of parts swapped, because silkscreen left some room for guessing and assembly people did not bother to clarify

[mrpackethead]

This is yet another reason that Contract Manufacturing is so expensive.   Theres just so much infomation that needs to be transferred. Its so much cheaper overall to do it in house.

[mairo]

...
crazy, how many weeks is it going to take to do 1500 part manually?

If you have the right process in place and a good manual P&P, a single trained operator could do 1-2 such boards per day (6-10hours). Assuming that the chips do not go smaller than 0402 and have less than ~10% ICs in lets say abit difficult to handle packages - BGAs, double row QFNs, large fine pitch QFPs.. Having large amount of 0201s or smaller will greatly increase the assembly time if handled manually.

Regardless of the use of an automatic or manual process, for a job like this you may need anything b/w half day to two days in set up alone.
And regardless of the use of an automatic or manual process, if you do have the right processes in place you also do not need the silkscreen on the board, but only the XY and orientation file (assuming the designer has done his job right). A silkscreen may help a manual assembler, but they will still get over 1-2 boards/day, maybe on a double shift (8-16 hours)  .

[Jeroen3]

...

...

crazy, how many weeks is it going to take to do 1500 part manually?

1500 parts? At max 600. Usually it's an order to prototype of 8 weeks. And then first 100 in two weeks.

[Mattylad]

If you have not provided them with an assembly drawing could be one reason for needing it.
Also they could be making the placement file using Graphicodes GC-Place which scans the pads, identifies the footprints, identifies the components from the legend, imports your BOM and can then output a placement file in their chosen format.

Has the Op asked them why they need the legend layer?

[SMTech]

Regardless of whether you supply an assembly layer & centroid file as part of your gerber/cad data, having no Silkscreen whatsoever is generally speaking, a bad idea. At the very least the silkscreen should contain useful markers for device orientation, connector names, voltages & test points, maybe a nice spot for a serial number and firmware to be marked with a pen and generally enough to make the board "navigable" without reference to drawing or documentation. Don't forget that once in the field, someone may encounter your board who doesn't have everything to hand but does need to interact with/repair it.

In theory if every CAD package and SM machine defined their components according to IPC standards you could send a centroid file and that would be that. However in practice people do not define their component orientations according to those rules which means a contractor needs to verify orientations visually against either the assembly layer or the silkscreen. If you define all the key items on the silkscreen you can also carry out a secondary check as part of your NPI process using the fiducial camera to step through each component location and overlay the device outline&orientation over the PCB image. For low volumes where some parts might end up being manually placed a silkscreen is helpful to find those locations, and if your board is weirdly symmetrical it might even help with getting it the right way round.

IME using Essemtecs centroid import tool most people define 2 pin non polarised devices the same way. But notably two recent clients do not, this doesn't seem to be a CAD package thing either as one uses Altium (as do we) and the other Cadstar which hasn't in files I have previously encountered. Presumably both of them have defined their own libraries from scratch and for whatever reason defined these 90 degrees out. By contrast ICs etc almost always require a 270 rotation in the import utility, which suggests that here it is Essemtecs templates that are rotated (and there is little point checking seeing as there is not much I can do about that). Thats why the Essemtec software (and one assumes everyone elses that is remotely focused on NPI & small batch) builds up rotation rules for components and packages which are stored, in my case, on a per client basis.

It is entirely possible (quite why you would is beyond me) to spec some SM machines with no CAD import facility meaning they cannot import data from a centroid file instead relying on either manually teaching the entire board using the overhead camera, which is basically impossible without silkscreen as well as being painfully slow and prone to error or by using a 3rd party tool like PCB synergy to go directly from CAD data to machine specific format. I have encountered contractors for work we considered too much for our process who set the board up using manual teach, I have no clear reason for them doing it this way but it seemed to basically boil down to the operator having no clue how to handle data that wasn't in exactly the format his Mirae line would take as input, given he had the file as both XY and full CAD data there is no excuse for it. PCB Synergy and Libreoffice are both free and both could probably have carried out whatever "conversion" he thought was needed.

It is also possible (as mentioned previously) to create a centroid file from the gerbers using GC-Place or  FAB3000 or Unisoft, again with these tools you need either a silkscreen or assembly layer as part of your output, again these tools take time and introduce the opportunity for human error (typo or misread) so I consider these tools "last resort" weapons when the client cannot or will not provide a centroid file.

[SMTech]

Also, they can be used as distributed fiducial points for the PnP machine.

No never, Silkscreen is not aligned with any precision on the PCB nor is it part of any quality check to define a pass/fail on the board so it can be a fuzzy misaligned mess and the board would still leave the PCB house.

[langwadt]

I don't think part designator on silk layer is necessary, but part outline is. Otherwise they can't manually rework failed PnP/reflow parts.
Also, they can be used as distributed fiducial points for the PnP machine.
Let alone that parts that require hand placing (extra large, extra heavy, parts requiring mechanical alignment or parts with special aesthetic requirements, etc.) need silkscreen to guide operators.

fiducials are copper because that is what matters

[mrpackethead]

Have peopel seriously tryed to use the silk screen as a fid?

[hulaomr]

I went to the assembly line and talked with the engineer.
Effectively, they set the PnP manually, meaning, the board need to have some alignment lines to be sure the BGA will be in place. The data I sent, was used as reference only. Obviously, once they programmed the machine, they are free to go.
Unfortunately, they didn't perform a X-rays to inspect the board. It seem to be a service by request. Once I received the boards, was too late, so, I said, let's first debug the board.
While debugging, I started to thing if "there is a reasonable x-rays inspection machine that I could place in my desktop"? Well, that's another history...

[mrpackethead]

I went to the assembly line and talked with the engineer.
Effectively, they set the PnP manually, meaning, the board need to have some alignment lines to be sure the BGA will be in place. The data I sent, was used as reference only. Obviously, once they programmed the machine, they are free to go.
Unfortunately, they didn't perform a X-rays to inspect the board. It seem to be a service by request. Once I received the boards, was too late, so, I said, let's first debug the board.
While debugging, I started to thing if "there is a reasonable x-rays inspection machine that I could place in my desktop"? Well, that's another history...

I'd be looking for a new assembler. That time they are wasting setting up the machine manually is time YOU are paying for.   Thats just dumb