Want to understand engraving geometry

[joeavaerage]

Hi,

Observation #1: When FlatCAM tells you to "include alignment pins" in the layout,

How do you plan to flip the board and achieve better than 0.1mm alignment between top and bottom? I use alignment pins, but others use a pocket milled into the spoil board,
yet others use an optical microscope to align to one or more fiducal marks....I don't really care what method you choose, but you must have a strategy to align the board when you flip it
or you will have a failure. The fact that I use alignment pins which are in turn suggested by the board geometry as represented by the CAM artwork is just what I find easy and convenient, no more
no less.

Observation #2: I never actually "made a technical selection" on the tape,

The specialised tape I bought for the thick copper boards I was doing looked much like that, in fact might have been as much as 2mm thick. You might call it an adhesive foam
almost. It worked well, in fact so well that I struggled to remove the PCB from the spoil board. Unless you are using coolant then I think it's overkill.
Just ordinary doublesided tape that you get from any stationery store will be fine. The only proviso is that the material of the spoil board have a smooth, firm surface for which the tape can adhere.
The brand I'm using at the moment is 'Pomona', but is in fact just a cheap generic doublsided tape, the sort of stuff kids use at school day in, day out.

Observation #3: The "vacuum holddown" has a certain appeal,

In the early days, about eight years ago, after having difficulties with clamps pulling surface distortions into the blank, I tried a few vacuum experiments. I use vacuum forming for model aeroplane
parts so I had a pump and a few useful vacuum parts to choose from. You need to be able to seal the PCB blank to the spoil board, and the smaller the board the more critical that sealing becomes,
as the area will determine how much 'clamping' force the vacuum will exert. If you are using just one or a few size PCB blanks then having a vacuum jig with an O ring around the periphery would
be doable, but if you are making random sized board it would become extremely tiresome very quickly.

There was a guy on the Autoleveller forum whom used vacuum very extensively, and he made many boards a day. He had a set of procedures that worked very well and I was very inclined
to copy him. Then I tried doublesided tape, and that worked so well that I never bothered following up on vacuum hold down any further.

Observation #4: I have enough time while waiting for the new tips (Wednesday), I printed out the instructions, I can ATTEMPT to get bCNC or Chilipeppr or Candle running (all have
internal levelling).

The last part.... '(all have internal levelling)'.......is the single most critical part to successfully using mechanical isolation routing for making circuit boards. You have to achieve a vertical height control
within say 20um to 50um over the entire PCB 100% of the time 100% reliably otherwise you will always be disappointed and have unusable PCBs. Whether you choose to use a software solution like
Autoleveller as I have, or another software, or some mechanical version of 'milling the spoil board flat',or combination, does not really matter....what does matter is that you achieve 20-50um vertical precision over 100%
of the PCB 100% of the time.

I have attached a pic of my spoil board. It's in the vice in my mill. Its really very convenient, I can remove it from the vice to fit a new blank, or flip a blank or whatever, and then put it back in the vice.
The spoil board jig has a screw head such that I can put it back into the vice in precisely the same location as when I removed it earlier. No need to re-zero or any of that rubbish, as I say, very convenient.
You may notice that this spoil board has been used several times before as evidenced by the grooves/divots/holes left over from previous PCBs where the tool/drill has passed right through the PCB
and slightly into the surface of the spoil board. I will use this spoil board for about 20 or 30 PCBs before the surface becomes so scarified that the doublesided tape loses adhesion. I will then flip it over
and use the other side. Once that side is 'used up' I discard it, after making about 50-60 PCBs, say a month or so depending on customer demand. The piece of acrylic, 200mmx200mmx 8mm costs
about $7.00NZD, say $5USD.

Just as a matter of interest I would use one or maybe two engraving tools over this time, unless I break them, and that does happen from time to time when I don't concentrate.

Craig

[jlawton]

Hi,

How do you plan to flip the board and achieve better than 0.1mm alignment between top and bottom?

Craig

OK maybe I misread that. I was watching a YouTube earlier today that demonstrated the "pocket approach", but that's not what I had. What I think you're talking about is aligning the
edge of the board ALONG the pins. What I did was to "take FlatCAM literally" and PUT HOLES IN THE BOARD FOR THE PINS. The problem with THAT approach is when you make the holes small enough to control the position of the board in both x and y, the board binds against the pins, hangs up, the pin placement warps the board, the pins stop the board from contacting the tape correctly and so forth. I'm really trying to understand this, it seems like the combination of "holes for the pins" PLUS using the clamps included with the kit instead of tape, THEN not having height correction PLUS apparently "soft" engraving tips (well I'll find out on that one soon enough) ALL TOGETHER basically made any kind of success impossible REGARDLESS OF HOW LOW I WAS WILLING TO DROP THE STANDARDS. Fair enough! Lesson(s) (almost) learned!

I "get" what you're saying (I think so anyway) about the KIND of alignment "strategy" that will work, there are a number of them that would be adequate. I wouldn't have trouble with aligning
along the edge of two pins, and maybe deliberately having a tiny drill hole at one side's origin, a "poor man's fiducial", to keep it simple. Right now my "spoilboard"is 1/4" masonite, I don't know how "unlevel" it might be, I'm certainly willing to go to maybe acrylic sheet which I have. I don't really have any fancy microscopes or anything, also I'm not selling these, they're just internal prototypes, I do have vias to put rivets in but if they're a little off-target no biggie. Now it DOES mean I'm counting to a certain extent on the board outline being pretty reliably straight, maybe that's what I was unconsciously (and erroneously) worrying about. But in the end you're absolutely correct, the "dimension" that you need to worry about first, last and always is the ACTUAL Z location, ESPECIALLY because of the tiny height of 1 ounce copper.

Next to go in the trash is the Art3D tape. If you think I still missed or misunderstood anything, please speak up! (There's enough material JUST IN THIS THREAD to make a pretty decent tech manual, or a couple of chapters of one at least, if you're inclined to write it, go look at Amazon for material on the various things you can do with the ubiquitous 3018 mill, there really isn't much available, maybe "this generation watches YouTube but doesn't buy books" but I'm just not sure about that, with everyone making noises about STEM education and so forth...)

[joeavaerage]

Hi,
I think you're making this hard for yourself, this is not rocket science.

To give you an idea, this is a small board I make a few examples of, it 50mm x 30mm.

I start by selecting a piece of blank PCB and applying doublesided tape.

Stick it to the spoil board. Note that the piece of blank board is not square or anything, it just needs to be big enough so that the entire board can be made.

Put it in the mill and drill the alignment holes. The Gcode to do that is pictured, two holes 50mm apart at (0,0) and (50,0). Note how the final move
goes to (0,5), and the I re-zero the Y axis only, effectively placing the two alignment holes at (0,-5) and (50,-5), and the (0,0) corner of the board is now 5mm 'North'
of the pins.

Fit the alignment pins, just short bits of 1.5mm welding wire. I used a 1.5mm endmill to drill the holes so its a neat fit. Certainly you need to bee assured that the board does
sit down flat on the spoil board, but its just not that bloody hard!

Craig

[jlawton]

I see what it is, I had the "order of operations" different. I was treating the pins as "permanent fixture components" of a semi-permanent spoilboard. FlatCAM WILL generate the .nc file
if you want it to, but they don't give any guidance on when that ought to be used! You're just using them as temporary elements for the "flip" operation. And in THAT capacity they aren't
in the way to provide torque, warp or ANYTHING because they're gone before the next operation proceeds.

Truth is it's not about "rocket science". If you give any 10 "fixturing techs" an assignment you'll probably wind up with about 12 fundamentally different ways to do that job. And the primary
reason you get so many is THEY'RE ALL OPTIMIZING FOR A DIFFERENT ORDER OF PRIORITY OF THE SET OF CONDITIONS. It doesn't make one "right" or "wrong" until you actually
get really precise about what you need to look out for and how, and you'll STILL get some variation. The difference here is YOU'RE GETTING RESULTS and I'm still just creating powder!
But I have to admit it's starting to make a whole lot of sense, and I'll keep rereading your contributions until I can't learn anything more, thank you so much!

[joeavaerage]

Hi,

If you give any 10 "fixturing techs" an assignment you'll probably wind up with about 12 fundamentally different ways to do that job

Yes, that's true.

I'm showing what I do, if you want the results I get then this is a viable means. That's not to say that you don't come up with your own ideas, but certain fundamentals must be met,
and one of them is registration between the top and bottom of a board.

1) The top etch is complete, a light rub with wet&dry and dust off
2) Apply the tape for the bottom side etch
3) Flip the board (on the pins) and stick down to the spoil board
4) Put back into the mill. Run an MDI to cause the machine to drive to the lower right hand corner, ie (50,0)
5) Zero the X axis, the the right hand lower corner is now the work (0,0)
6) Bottom etch complete, drill file run, mill file run. I use a 1.5mm endmill for the outer mill path.
7) Board complete.

Craig

[jlawton]

Tonight I spent some time typing in install scripts, and I managed to get bCNC, Candle AND Chilipeppr ALL running. They ALL have built-in autoleveller functions. (I can't use the graphics on Chilipeppr but everything else works.) But the "make" command alone for Candle on my machine took about 20 minutes to execute, so I caught blinkenlight in a flat-ass lie, but for him that's the norm!

[joeavaerage]

Hi,
hardly the sort of thing I would say to someone who is trying to help.

A couple of days ago I did a PCB where the top etch file, after Autoleveller had done with it was 3.8MB, and it took just over two hours to run.
I have just started a top etch file on a another PCB at it 11.6MB! I'm guessing 6 to 8 hours. Going to let it run overnight, I've been at this all day already.

Craig

[jlawton]

I'm sorry, this is a "private war", no offense intended!

[joeavaerage]

Hi,
the top etch did not take quite as long as I expected, 4 hours 40 minutes.

Pretty happy with the result, a hint of undercutting at the very to right hand corner, possibly due to the jig flexing under vertical load???

Quite a few of the signal traces are 0.2mm wide, finer than my more normal 0.3mm, but required in order to fit 5 traces within an SIOC outline. They have all turned out perfectly.

I have in the past tried 0.1mm wide traces and they do work but are very tender. If you even rub the surface with wet&dry too vigorously you can 'wipe' the trace off the substrate.

All these fine traces, becoming increasingly difficult as the width decreases, is..... that if you apply any heat, say soldering, the the glue holding the trace down gives up the ghost
and the trace dislodges from the substrate, ie a failure. A fine trace, say 0.2mm or 0.3mm, must be terminated in a pad or doughnut of reasonable area such that the softening of the
glue does allow the feature to lift off the board.

The PCB pictured is 210mmx150mm. When it was probed for 'level' it was found to have a variance of 0.705mm (705um) from highest peak to lowest valley. I have posted a pic of the
visual representation of the height variance as discovered by Autoleveller. I was rather surprised at that amount, I rather suspect that the spoil board/jig is not quite flat relative
to the vice. Autoleveller still did a very good job, if that were not the case the 0.2mm traces would not come out as well as they have.

Craig

[jlawton]

So if Craig happens to be there...or, well, I suppose this could be considered a "FlatCAM" question, or maybe not, let me go back to asking about geometry, yes a lot of what I was doing was incorrect, now I need to back and review what I was previously doing in light of the rest of this enlightenment, which I assure you was most helpful indeed...

When I went looking for a YouTube video with "reasonable" values that I could copy in order to set up FlatCAM, I DID find somebody who did one based on FlatCAM beta version 8.993.
So the current version (the last time I looked) was beta 8.994, so since that was the ONLY version giving instruction on ANY "beta" version, by definition that was the set of instructions
that in my mind was "most likely to produce results". I guess the part that I'm suspicious about is a setup issue that I didn't understand, the video author advised that if I were using
a 30 degree engraving tip with 0.1 mm tip width like his, to get the results that he did I should set up the isolation routing for three passes at 25%. The problem is I don't really know
what that means dimensionally (yes I know what "3 passes" means) or how to determine whether it's reasonable! How do I calculate how much track width "would be left" for those parameters given a specific "design" track width and the other parameters? Are the other 2 passes supposed to be OUTSIDE the actual track, and is that really happening? Is it really intended to mean that the total width of the isolation area on each side of the track is 0.15mm (0.1 x (100%+2x25%)), and the track width is "supposed to be" whatever the artwork says,
and is that really supposed to be happening? Or what would be a more reasonable set of parameters, and how do you get there? Or would someone argue there's some other free toolset that will produce the same or better results than FlatCAM going directly from a set of Gerbers all the way to Gcode? I just want to get a better idea that either these parameters are reasonable or not and what they possibly mean, the REAL problem with FlatCAM is the documentation doesn't come CLOSE to addressing any of this, but you're required to put in SOMETHING as if you knew what you needed and what it meant!

[joeavaerage]

Hi,
I don't know enough about FlatCAM to be of much help.

As I have posted I use EAGLE, or a version of EAGLE that is incorporated into Fusion 360. The Fusion 360 version has 'broken' the 'polygon pour' function of EAGLE which was intrumental
in generating Gcode user a 'user language program' that was provided by a long term user of EAGLE. As a consequence I use Fusion, the electronics part, to generate the artwork, then export
the board to EAGLE, and then use the 'pcb-to-gcode' user language program. Its a bit of a work around, but I am so familiar with pcb-to-gcode that I do it anyway.

The set-up of pcb-to-gcode asks the tool tip diameter, I have a value of 0.175mm set. I have a stepover of 0.09mm and an isolation value of 1mm.

The Gcode generated goes around the outline of the traces plus 1/2 the tool tip diameter. Thus the trace width as displayed in the CAD will be realised in the final etch by virtue of the toolpath
being displaced by the tool tip radius. Then it takes another pass 0.09mm further out, and then another at 0.09mm etc until it gets to the isolation value I want, 1mm typically. It takes about 7 or 8 passes
to get to this value,

If I'm doing a 230VAC board the I need more isolation, at least 3mm, but that take lots of passes at only 0.09mm at a time!

I have been told that Fusion can do a similar thing, but I have never devoted the time to investigate and experiment with it. I'm told for instance that you can have varying stepovers, that is to say
you might have a small stepover of 0.09mm to get perfect copper clearance around the very small features but as you get further away you can use a different tool, say a 0.5mm endmill, or a 0.8mm endmill,
and takes larger stepovers. This allows larger isolation distance WITHOUT using a very small tool like an engraving tool and having to take dozens of passes, and therefore take all bloody day.

Note that I chose the tool tip diameter carefully and after some experimentation. I have set 0.2mm as the MINIMUM distance between adjacent traces. Clearly the tool tip diameter must be at least
a little less than that distance to fit between the traces. After considerable experimentation I found 0.175mm with my 60 degree engraving tools to give perfect isolation down to 0.2mm between traces,
without taking an in-ordinate amount of time to etch.

Craig

[jlawton]

I can't change software toolsets. Both the parts placement (in KiCAD) and the autorouting (in freeRouting.jar) are DONE. Either I get FlatCAM working or I send out for etched protos, if I elected to do that I at least have "panelize" working to save costs. Support and documentation for FlatCAM are both (for all practical intents and purposes) NONEXISTENT. (And I guess I still can't rule out spindle runout as a possible contributing factor since I can't measure that either.) Holy sh*t.

But like you're saying you do have to input a MUCH larger # for "effective tool tip dia" than the "mechanical" one even in your "working" system. Guess I need to start there. Some of these boards are simple enough that there's "just enough there" to use for a test platform and not "blow away" a lot of clad material while you're learning how to make it work. Anyway this is the LAST time I allow myself to "get suckered in" to the racket that "all I have to do" is follow the instructions in the video. Like he**! Total MISinformation!

[joeavaerage]

Hi,
my apologies, when I used the word 'etch', I meant mechanical etching, or isolation routing, not the more widely understood chemical etch.

Many manufacturers claim a 0.1mm tool tip diameter. Can you even see it, or more importantly measure it?. Even if you could and found that the tool
did indeed have a tip of 0.1mm.....how long would it last, especially those skinny little 30 degree tools? Even 10 um of runout would cause such a fine
tip to break or split the moment it touched the PCB blank.

My aim was to have 0.2mm between features. The IC pads of a 0.4mm pitch quad pack are 0.2mm apart. In order to etch (mechanical etch) such fine features requires a toolplath
that goes up the inside of the lefthand pad but down the inside edge of the righthand pad. Clearly then the tool tip diameter must be smaller than your intended minimum
separation between features. If I had nominated 0.22mm say, then the PCB-Gcode program would not generate a toolpath that went down the gap because it recognises that the
tool is too wide and would therefore cut away at the edges of the pads, which its forbidden to do.

What the actual tool tip diameter of my preferred tools is I cannot say, I cant really measure it anyway. If I had an optical microscope perhaps.....but who really cares?
A more important setting is the stepover. If, lets say my tool tip is 0.1mm in diameter, then the max stepover that did not leave a sliver of copper untouched would be
0.1mm or 100% stepover. In practice you want some overlap to ensure NO copper slivers escape the tools attention. I use 0.09mm, and remembering that I use 0.175mm
as tip diameter that is nearly 50% stepover.

I did not arrive at this compromise immediately, and suffered many failures or 'less than optimals' before I found it. I did not slavishly follow all the advice I was
given, but what I did, and take a certain pride in, is to glean from the advice that I was given 'what is the essential principle or objective' was in each case. By focusing
on the objectives you can devise your own techniques that suit your application. Its not necessary for you to use Autoleveller for instance, although that is my solution to
the ever present Z height control problem, so long as you find and use a technique that achieves the same result.

Craig

[jlawton]

Oh that's fine, not ALL of this came as a surprise, and certainly nothing was all that shocking...

I guess as sort of a "newbie" I expected there to be a sort of "controlled relationship" between the ACTUAL tip diameter, and the number you enter into the software, and maybe a way to measure/verify it. And yes I suppose that I thought there would also be "some sort of formula" you could use to figure out the "performance" you would get for the "parameters" you stick in, and in a rational world there MIGHT even be corrections for various types of runout and so forth. And I suppose one of the many reasons that I would have expected to have this "nailed down" has to do with the fact that my previous experience with "mechanical engineers" was with people who designed computer storage peripherals (like 250 in/sec tape vacuum column drives), and if they saw a hummingbird in flight they would be able to tell you what the magnitude of the fifth harmonic of the motion of the bird's wings was because...well that was their job, that was WHO THEY WERE, they were surprised by NOTHING. I suppose I was spoiled...

And really what this comes down to, I have some other projects that can easily "make use of" my brand new 3018 machine, in terms of circuit density they're nowhere near as demanding as the project I'm working on now, THAT is the one that is actually likely to make some money I think. Or maybe I'll take the 3018 and "stick on" that cheap 5 watt diode laser and use it to "faux die cut" some fancy cardboard product packaging (don't know whether that's reasonable yet either). It just HAPPENED that at the same time I got my first 3018, I started playing with FlatCAM, I also got KiCAD 6 (I had previous experience with KiCAD 5), and I started looking at that "Panelizer" software. Now I had initially THOUGHT maybe I could do panelizing INSIDE FlatCAM to engrave them, but I soon learned that's WELL beyond the package in its current capabilities, but "Panelizer" works fine if you're sending out PCBs to be etched. I never had any package which was particularly useful for panelizing, that certainly isn't one of KiCAD's internal strengths, but in most regards, for what I need KiCAD has just the set of capabilities that I need (I don't use its ngspice though, LTSpice is MUCH more consistent). I also learned that the guy who authored the YouTube video on using FlatCAM, well he really isn't an expert on anything, that piece of video needs to have a "toxic waste" sticker attached...and no in that program there is nothing designated a "stepover" but it sure needs it.

Anyway I'm going to spend my immediate future doing a "cleanup pass" on all the silkscreen layers, I would ignore them if I were just engraving a proto, but if I'm sending this out I may as well get some work out of them. I want to thank you for all your asssistance, and before I go I will note you ought to take a look at any of the following programs, bCNC, Candle or Chilipeppr, all three of them have INTERNAL autolevelling so you don't have to "reprocess" the gcode files (or worse keep two sets, compensated and not), it strikes me that makes the process much simpler and cleaner, or it seems like it would, FWIW.

[joeavaerage]

Hi,

I will note you ought to take a look at any of the following programs, bCNC, Candle or Chilipeppr, all three of them have INTERNAL autolevelling so you don't have to "reprocess" the gcode files (or worse keep two sets, compensated and not), it strikes me that makes the process much simpler and cleaner, or it seems like it would, FWIW.

I use Mach4, and it vastly outperforms any of those others, it just cannot be compared. Can you write code to do Automatic Tool Changing for instance, in any of them?
Mach4 has up to six coordinated axes, another six Out Of Band axes, and up to four slaves for each master axis, and the list goes on. That does not include Mach4 Industrial which has
parametric programming (Macro B) or Mach4 Expert that allows multiple instances of Mach running at once for things like mill-turn machines, with or without auto pallet changing robots etc.

I use Fusion 360 mainly, but also EAGLE as previously explained and a number of utilities, Autoleveller for example, but also a Gcode viewer/simulator, if you don't have one, get one, and a conversational
Gcode programmer. This is just the bees knees when your at the machine an you need a simple toolpath but cant be bothered with CAD/CAM in the usual sense, say a surfacing toolpath or a bolt hole circle.
In short I use multiple software programs, either individually or in combination to do the work I do with my machine, and I need them all. CNC is largely concerned with computers after all.

Craig

[jlawton]

Having "automatic tool change" in software doesn't do me any good because I don't have the hardware to support it. I guess Mach4 is an improvement on Mach3 which tied up a whole PC and all the machine interfaces had to be tied some way into the parallel printer port which hasn't been a standard part of PC architecture for literally decades, I don't know if Mach4 got away from that. The machine control on my cheap machine is an Arduino 8 bit and it doesn't need to be any more elaborate than that because the stepper motors on this or any machine are insanely slow in comparison, since it's Arduino you can upgrade the control program in flash through the standard interface. There's a 32 bit upgrade including offline controller you can get from China for $30 but the only people who bother are the ones using a laser instead of a spindle, and even THAT'S only because the beam control then upconverts to 16 bit pulse width modulation which allows for finer control.

I've had my own computer system since 1979, and been studying them formally since 1968. Computer hardware nowadays is unbelievably cheap, but precision mechanisms and LICENSED software (which I avoid) cost through the nose. My netbook is currently running Linux Mint MATE 20.04, and all the software I mentioned I have built to run on Linux, even in those cases where the language it is written in is somewhat obscure (Candle was written in qt5).

[joeavaerage]

Hi,

I guess Mach4 is an improvement on Mach3 which tied up a whole PC and all the machine interfaces had to be tied some way into the parallel printer port which hasn't been a standard part of PC architecture for literally decades, I don't know if Mach4 got away from that.

Both Mach3 AND Mach4 have had the option, for a decade now, to have an external motion control board like an Ethernet SmoothStepper and therefore render the parallel port null and void.


My machine runs servos and I require Step pulse streams of 416kHz to get to 25m/min g0's, and I require 466kHz to get my C axis (AC servo based) spindle motor to full rpm. The parallel port
couldn't get even close, while my SmoothStepper is scarcely taxed being capable of 4MHz.

LICENSED software (which I avoid) cost through the nose.

Mach4 cost me $200USD, and I've used it ever since, seven plus years. Best investment in CNC I ever made.
I'm not a fan of subscription software but I pay $500US/year for Fusion 360 and consider it great value. If you don't want to pay for software don't be surprised if the freeware is not up
to snuff. If you are prepared to pay modest amounts then there are some very good solutions at fair prices. I paid $440USD just for one servo......why would I whinge about $200 for Mach4?

Craig

[joeavaerage]

Hi,
how are you making out?

I have to make some MOSFET switches, and this is the board for one of them. The copper layers are 12oz/sq.yd or 0.42mm thick. The MOSFET output
circuit has to handle 50A peak and therefore the thick copper is required. The nominal voltage is 600VDC, so I have allowed 5mm isolation.

Ordinary engraving tools are no good for this stuff. I used a 0.5mm two flute endmill running at 24000rpm under flood cooling. The first pass, which is in effect, a slotting
toolpath, always a challenge when the cut depth is 90% of the tool diameter, I slow down to about 200mm/min, but the second and subsequent passes are at 400mm/min.
The thick copper requires something like flood cooling or air blast in order to clear the chips from the cutzone or the tool will load up with swarf and break.

Tool life is largely determined by the amount of abrasive swarf that comes amount by cutting into the fibreglass board. In this case I nominated a 0.475mm cut depth, with a 0.42mm
copper layer, and therefore potentially I'm cutting 55um of fibreglass. I was able to cut both sides of this board with one tool, but its now so blunt that I'd not use it again, about 1 hours use,
not great. If I were being really picky about tool life then I might reduce the cut depth to 0.44mm and therefore cut as little as 20um of fibreglass. This clearly places high demand
on you strategy for 'leveling' a PCB blank.

In previous years I was able to extract 8 hours use from a tool cutting this exact same material....mind you the cut quality had degraded badly by then because the tool was so blunt....but it just goes show that it
can be done if you are very particular Z height control. After eight hours use I removed the tool, 0.5mm endmill, and it was still intact!

The cutting parameters I used are:
Tool diameter=0.5mm
Step Over=0.25mm
Isolation required=5mm
Cut Depth=0.475mm
Cut Speed (XY plane)=400mm/min
Cut Speed (Z direction) =200mm/min
RPM=24000
Flood coolant engaged.

Using a 0.5mm endmill means that the minimum clearance between features is just a fraction more than that, say 0.51mm. That's why I've used some 1812, 2012 and 2512 capacitors and resistors
so that I can fit a trace between the pads, which is just not possible with a 1206 package AND maintain 0.5mm clearance.

I still use doublesided tape to stick the PCB blank down (and the alignment pins) but I put duct tape around the periphery to try to exclude the coolant from the doublesided tape and degrade
its adhesion. For the half hour or so while the toolpath runs its fine.

Craig