Hi Bob,

I guess I can help a bit more to understand whats really going on when using
inkjet printers with different drivers and settings.

But first of all, I `ll tell you which Hardware I use as a reference.
The "upper" model of our small format printers was based on an R1900 Printer
until last year; i`ve buildt and sold some more than 100 units of these last
year.
We changed to the R2880 with the beginning of this year. Actually, I donīt use
these models for the pcb tests, as these printers are more intended for
graphical prints and therefor dont need to print that precise like it is
necessary for 3 mil traces.
We already sold some for pcbs, but for these people traces/spacing of 10/10 was
more than enough, so the standard models were o.k.

For the pcbs test I use a prototype, based on an Rpson 2880. The tray is made of
steel / aluminium, milled to very fine tolerances.
Its guided by bearings over a lenghth of 23 inches to get it absolutly straight
and in the correct angle through the printer.
The printer is modified and calibrated to be really precise in feed direction
also; for the starting point there is a fixed stop made of steel - I am able to
push the metal tray against that stop to start from the the same reference point
again and again.


So lets see the problems of the windows driver relating to its
screening/dithering methods.
Looking at the photos in the "Don`t trust photos" album, you can see that the
dots are not placed in a way we would have expected, and the finer the line is,
the more difference you see between the input (file) and the output (print).
The reason for that is, that Epson tuned the driver not for pcbs, but for text
and photographic printing. For text and graphics, the optical precision is more
than enough, and for photos (and graphics) a very high goal is always to avoid
banding. Thats the reason, why the dots are not placed exactly where we would
expect them, but beeing spread about a somewhat larger area. If the printer
would really print exactly one dot as an output for every pixel as an input,
that would result in:
1. Terrible banding
2. Unbelievable pale prints especially onto photographic papers etc.
(You will find an explanation for that later)

Obviously, that couldn`t be an aim of Epson, and they did a very good job to
avoid problems like these. Be aware, that Epson had to find solutions for these
problems, which work out of the box for people who haven`t any idea, that
problems like these even exist in inkjet printing.

So lets have a look, how input and output are related, and how - for example-
the thinnest line theoretically could be printed. For that, we begin thinking in
a "perfect world" without dot misplacements by hardware tolerances, misalignment
etc.

As we want to produce pcbs, traces may be horizontal, vertical, diagonal or
circles etc.

Looking at the printer/driver resoltion, the limitin factor is than the lower
number of dpi in a given setup; so 5740x1440 doesn`t help (in theory) much more
than 1440x1440dpi.

So we start to think in a 1440x1440dpi matrix. The only thing the printer can
do, is to print a dot at a crossing point with this grid, or not (simplifed:
actually, it can print a dot of variable sizes, or it can print multiple dots
depending on the setup)
This means: in horizontal and vertical direction the minimum distance between
two dots can`t be any smaller than 0,7mil. As we also want to have diagonals,
the minimum distance grows to about 1mil.
We dont want to have a dotted line, so each dot have to be large enough in
diameter to get enough overlapping between two dots.

So a quite good compromise for the desired accuracy is a dropsize with a
diameter of about 1,6mil.

I have posted two drawings which explains the basics of this in a separate photo
album.

So the thinnest line which is printable would consist of a single row of dots -
resulting in a linethickness of about 1,6mil in each direction.

2nd thinnest line would be about 2,3 mil horizontal and vertical; 2,6 mil in an
angle of 45 degrees.

1,6 mil is quite exactly the diameter of an inkdrop you get by the NGen ink,
using a dropsize of 3pl, and you get it repeatedly. (If you keep in mind what I
have written and drawn so so far, it may get more understandable, why it is
absolutly necessary to be able have each part of your setup under absolute
control, when you are trying to get
real etched and measured traces below 5mil.)
It was one of the most demanding problems to design the ink in such a way, that
it behaves like that.
At this point, you may ask yourself, why it wasn`t formulated in such a way that
it gives such a diameter for a 1,5pl droplet, which is mostly used in Epson
inkjets as the smallest dropsize: easy to answer: thats simply not possible.

Sounds sad to you as your printer doesn`t use these 3pl drops? Fortunatly, the
way the printers/drivers work are more complicated than described until now,
giving us additional possibilities, and which may also be helpful for the people
who want to use other inks.

Lets go back to our input file. It has some lines with the width we want to
have, and we define it in a solid black, as we want to have it to be printed
with 100% black (or one of the other colours installed).

But what is an 100% output?

Is it the maximum what the printer/printhead can spit out? Is it the blackest
black what we can get on a paper? Is it the 1440dpi matrix, which is coverd by
one drop at each of its crossing points - what would mean that each input pixel
would result in one output pixel?

All of that is true.
And all of that is wrong.

Its possible to drive an Epson in a way, that it really spits out all the ink
the printer itself can do. Imagine 10 input fields, first filled with 10% black,
last with 100% black. The output than would result in something like that: 10
percent input would optically look like about 50% black. 20%input would look
like even more than black.
100% input would give you (with NGen inks) a field beeing noticable thick, maybe
about 30 mil thick.

Now imagine photographic printing onto photoglossy, highest resolution, smallest
dropsizes. If each input pixel would result in one output drop, than the very
light shades would look quite well, but the darks would be by far to light. 100%
input would maybe just look like a 50% gray.

So obviously, there has to be some correction that limits the output to a useful
amount and which tweaks the relation between input and output in such a way,
that 10% input looks like 10 % gray and 100% input looks like a 100% black.
Thats a linearisation curve which looks a bit like a exponential curve for the
example with the photographic paper, as you need much much more additional ink
for the step from 90 to 100% than for the step from 50 to 60%.
That is (simplified) one of the many steps in the buildt-in colorcorrections,
which the manufacturers of the printerdrivers created as presettings for you to
get out a photo which is about quite right.

But what do we need for pcbs?

If we have a solid line, we want to have a dot at each position, where there is
an pixel in the input file. We don`t want to have any dots less than that -
which would happen in a setup, where the input of 100% is defined with an output
of lets say 80%. In that case, 20 % of the dots would`t be printed...not good.
We also want - not like the window driver - beeing the drops at the position we
want them to be, and fine widhts not spread over a larger area to reduce
banding.
So covering each crossing point with one drop seems to be the goal.
But what about 1,5pl drops with NGen inks? What about inks which gives you
smaller diameters, not filling the gaps between the crossing points of our 1440
grid? And what if the NGen isn`t printed thick enough to withstand the etchant?

For that, we need an output which not only prints one, but several small drops,
until the resulting output gets close to that what we have in our input files.

We can get close to that using gimp or photoshop and gutenprint for windows.
There is no general setup for all printers, as the models differ in the way of
their output. But I can give you an general idea about the workflow in a
separate post.

If you have read the whole post and looked at the different files of the
photosection I guess you should notice that the traces I have "microscoped" are
quite near to the limits which are possible with the tools we are using here, so
the hints for the setups should`t be so wrong also.

Ingo








--- In Inkjet_PCB_Construction@yahoogroups.com, "Bob" <bobd@...> wrote:
>
>
>
>
> Hi Ingo,
> Thank you for your post, it's very interesting. You taking the time to take
and save the example photos is much appreciated.
>
> How could I take your sharing of hard won info as offensive? I'm learning and
theorising about what I have found not pushing a belief. So please continue to
take the 'risk'.
>
> I want the Gerber form of the group PCB on the homepage to use as a test
coupon to guage my progress. I understand that it may print with different line
widths but that is what I want to look into next. At this stage of my
experimentation I'm trying to set up the process so that it can be easily
checked and reproduced by others. So standard input files are a means to that
end. At the present time I doubt if I could print it to just look that good
never mind under a microscope!
>
> Talking about microscopes. I have set aside a USB one to use when I get to
that stage of my experiments. I could use some advice on the best paper to use.
What do you use for getting a nice print to examine? Somewhere, I have one of
those 'Peak' brand optical PCB microscopes with a measuring graticule I could
use as a size reference.
>
> I like to understand what I'm doing when I use some process, like making an
inkjet PCB. I'm not really a 'try this' and 'try that' kind of person. I have
settled on the ESCP/2 data as my reference point. If the image in the file is
poor than the print could have been better. I'm trying at this stage to just get
a good image in the file!
>
> I think your comments about DPI are good. I see the printed DPI as just an
upper limit of what is possible. After all 3mil lines and spacing at 100 DPI
just isn't going to happen.
>
> I'm still too busy developing my 'PCBprint' program to actually take advantage
of all that it tells me! One thing that looks exciting will be the ability to
make lines of an exact number of pixels, and of different dot sizes, by patching
the print file.
>
> The the R2880 is an 'up market' model. I will look up it's specs. How do you
like it?
>
> I never thought of the drop size varying with the type of ink until you
mentioned it. So now, I will have even more experiments to do!
>
> Bob.
>
>
> --- In Inkjet_PCB_Construction@yahoogroups.com, "Ingo" <Ingo.Schuldes@> wrote:
> >
> > Hi Bob,
> >
> > please don`t take it as an offensive comment, but photos like that are
completely worthless. Of course you will get out a fine line when you set it up
as 1 mil, and yes, it will be thinner than a 2mil one - but in reality you
neither know by looking at it or take a photograph how thick such a line is in
reality. The author of that photo you mean already said that he has no idea
about the real dimensions of the traces in an older post.
> >
> > To show you what I mean, I have made a simple example for you:
> >
> > I have setup a similiar file in a vector file with lines from 0,001 to 0,01.
Printed it out using an converted Epson 2880 at highest resolution, photographic
settings with the standard Epson driver.
> >
> > Printout looked absolutly perfect and flawless to the naked eye - even the
0,001 line!
> > Took a picture of that and put it in a new folder in the photosection in a
folder "Don`t trust photos".
> >
> > Next, I made a microscopic capture with the same settings I used for the
ones in my other folder.
> >
> > The perfect 0,01 line showed to be about 0,013, and the 0,001 turned out to
be completely useless, beeing about 0,005 and totally jagged.
> >
> >
> > I have made the same faults in the beginning...was absolutly enthusastic
that I could get a one mil line even in my very first try..without any effort.
Than I started to take measurements...and learned that extra efforts have do be
done to come close to the lines you want, especially the thin ones, without such
jagged edges.
> >
> > You should not only consider the amount of dpi. That only tells you the
distance between two dots. If a dot itself has a diameter of 2 mil, you can`t
print anything thinner - than 2 mils is the theoretical thinnest line
(consisting of a row of single perfect matching dots - extremly hard to get). If
you print at 5760x1440dpi, than the smaller number is the limiting factor.
> > So the second thinnest line thats (theoretical) possible with dots in
diameter of lets say 2mil, placed in a distance of 0,7mil gives you a line of
2,7 mil. (2x2/2+1/1440).
> >
> > Due to imperfections like misplaced dots thin lines consisting of less than
2 dots should be avoided, so at the moment I see a limitation at about 3mil,
regarding the fact the Ngen gives you at the moment diameters of about 1,6 mil
with a 3pl drop. Until now, I avoid using smaller drops, as these are misplaced
much easier in less than perfect setups.
> >
> >
> > Ingo
>