Close

P2 - The vision meets reality

A project log for DIY Space Grade PCBs

3D print circuits using silver ink and ceramic substrates to make high performance, multilayer, fine line < 3mil, PCBs

chuck-glasserChuck Glasser 05/30/2016 at 03:131 Comment

A standard poker chip just fits within the boundaries of this test PCB. Six internal signal layers with 2mm pads on one side and 30 mil pads on the other. Using the absolutely wonderful open source layout tool Kicad, it took three days to layout. 26 mil annular ring with 16 mil drill. Traces are 9.8 mils wide. By any definition, this is pushing up against present day limits.

One thing that's going to take some time to wrap my head around is that the layout rules of LTCC are far, far, friendlier than conventional FR4 PCB circuitry.

There is no cost burden for buried and blind vias! So, all the zigs and zags are unnecessary. If that hole is not on your routing layer, go right over it! Weird!

Here's the reality for a single layer, in this case the first inner layer L1.

Disappointing resolution at best. Keep in mind that the best this thermal print head can do is 96 DPI. Technologically the print head used, a HP C6602 is ancient. There is a cloud as well caused by adjacent channels firing. The exact mechanism may be grounding or just the limits of the technology for this era of Ink Jet.

This is what I was hoping to see.

So, that's 96 x 96 DPI vs. 5760 x 1440 DPI a factor of 900, almost 3 orders of magnitude difference. In this case the printer was an Epson C88. Impressive performance for a printer that goes for < $100.

Incidentally, the C88 uses servos for the page traverse.

So, what does this all mean?

1. It appears that the low performance head will be adequate for printing electrodes on a ceramic substrate. The present test board is well beyond reach.

2. It will be necessary to use higher performance print heads, AKA piezo along with servo motors instead of steppers to reach the desired quality.

3. It should come as know surprise that the software that is available does not match up well to the present hardware.

To use the present CNC system, or any CNC system for that matter one uses G-Code. Most maker related PCB software utilizes PCB milling as the operant technology and are subtractive in nature. LTCC is additive. The artwork for Kicad was exported as SVG images and processed with InkScape and the InkScape extension GCodeTools. However to make an 8-layer board it is necessary to generate no less than 16 seperate files. Eight (8) to drive the laser for the drill files and eight (8) printer files. To have any hope of having a usable system it will be necessary to have the software parse gerbers and partition each of the layers into strictly placed regions so that only one alignment between the laser and printer will be necessary. FlatCAM may be a suitable starting point for some hacking. It's going to take some very serious coding to make this work.

Better to fail often. One cannot reach the goal without constantly correcting the course.

Closing on an upbeat note.

This is what the long bed really looks like.

The rotary table will more than likely have 8 positions. Shown are a 400 watt, 30K RPM spindle motor, conventional FDM print head, peristaltic printer head for depositing colloidal suspensions, and a piezo print head. To be added are a camera, vacuum pickup, touch probe, and TPD. In any case, 7 of the 8 positions will utilize 80/20 mounts so they are universal.

Might as well start here for the X traverse servo.

So, all in all, it looks like for this second round I've managed to dig up a huge can of worms. Not bad for 6 weeks of effort and lots of 16 hour days!

Discussions

Tobias wrote 06/03/2016 at 23:35 point

Is there a resolution at which this method would already work reliably to make PCB's? I am constantly isolation-milling single-sided low-res PCB's for prototyping with my Zen Toolworks CNC, would much rather print them if I could. In other words, having a print head that I can attach to my existing CNC would be awesome!

  Are you sure? yes | no