My first DIY modules - starting a Kosmo format build

[juggle]

The big problem is it’s a bitmap. And PCB artwork really should be vector.

Forcing my printer into B&W only actually makes it even worse because more of the barely visible dots show up.

If you zoom in in acrobat it’s just barely possible to see them. They’re there in the PDF - just very light:

If you import the PDF into photoshop and adjust the levels it becomes even more apparent:

Vs. vector artwork from Eagle (this isn’t actually the right layers for etchign a board…just a quick test print to show that the printer isn’t adding any halftoning when it’s fed good vector artwork)

I could get a better print from the yusynth files…it would take some cleanup in photoshop first but it really would be time better spent to just redo it properly as working from bitmap for something like this just really isn’t ideal.

[fredrik]

@juggle

The big problem is it’s a bitmap.

It’s not. Your tools are messing with you.

(Well, at least the random files I checked are extremely clean on my machine at all zoom levels, and an assortment of PDF analyzer tools couldn’t find any images in them.)

[eric]

I agree PCBs should be published in a vectorial format, but these bitmaps aren’t that bad if it’s the only thing you have…

Convert it to black&white in photoshop before printing if your printer enhances these non-visible but very slightly non-white pixels.

If that conversion does the same crap to the image (I don’t use photoshop, only the software I’m developing… so I don’t know if it does a simple threshold or halftoning), use the histogram to get rid of these pixels.

[analogoutput]

@juggle

I was going to try one of
North Coast’s projects since they share kicad sources…but…I can’t get them to open Kicad keeps saying my version is out of date to open their files…but I downloaded the most recent version so I’m not sure what’s up there.

I’d never tried looking at one of their KiCad projects so I gave it a go, and got the same error. Based on what I googled here, I was able to get it working as follows: I opened the pcbnew file in a text editor (by which of course I mean emacs). I also opened a pcbnew file that my KiCad can open. I deleted the first line of the North Coast file and replaced it with the first line of the other file. Then in the North Coast file I deleted these lines:

I saved the modified file. I was able to open that version.

Strangely, Eeschema doesn’t complain when it opens the schematic but all the symbols are question mark boxes. They come from a library that’s included in the download but that isn’t loaded. I had to go into the symbols editor and add a library, mskala.lib in the kicad-symbols folder.

[fredrik]

@fredrik

Well, at least the random files I checked

E.g. does this have the same behaviour?

PCB-3096.pdf

(if you open that in a text editor, you’ll see that it’s pure vector data)

[eric]

That’s the Steiner-Parker VCF.
The Arp VCF’s pdf is a bitmap…

[analogoutput]

If I open that YuSynth PDF in GIMP at 2000 pixels/inch resolution (which is probably way more than needed but I wanted to make sure that wasn’t the issue) and zoom in it looks like this:

Then applying a threshold it becomes this:

I would guess the latter would give cleaner results.

[eric]

That’s exactly what I said, they aren’t that bad.
A little cleanup with either threshold or histogram (as we can see, threshold does a great job, no need for histogram…), done in less than 2 minutes, nothing major.

The files are at 300dpi, which is enough, and should print fine, at the right scale.

[juggle]

@fredrik

It’s not. Your tools are messing with you.

(Well, at least the random files I checked are extremely clean on my machine at all zoom levels, and an assortment of PDF analyzer tools couldn’t find any images in them.)

The ARP was apparently a bad choice as a test subject. I just checked the Steiner and it is vector, and so is the diode filter, and the wavefolder…I stopped checking there because it was enough to convince me that enough of them are vector that the ARP was an outlier.

And yes - a quick threshold or histogram tweak will clean the ARP image up. But you’re still left with softer less defined edges than if you were to print from original vector art. It’s totally usable and I probably would have done it if I had noticed the noise earlier in the process. I unfortunately picked a worst case for a first test and didn’t notice it until fairly late in the process. As a one time photographer who helped switch two different labs from traditional to digital darkrooms I have no problem working with bitmaps…in fact illustrator and inkscape tend to drive me bonkers when creating artwork. But…at the same time I very much prefer vector artwork when it’s appropriate for the application (as long as someone else creates it for me!)

For the size of traces used here it really doesn’t matter. My endgoal is to be able to etch some boards for SMT parts - for those finer traces it will be more important to get good clear detailed lines. But for through hole stuff like this - there’s a lot more room for error.

Also just to be clear…I’m my no means putting down Yves’ designs and contributions! He could share his layouts as photographs of crayon drawings and I’d still appreciate them! I fully get people not wanting to release original design files because it makes it too easy for cloners to abuse their IP. My preference is to share openly and freely, but I don’t hold it against others if they don’t feel the same.

[juggle]

So…my transparencies showed up a day earlier than expected. So…figured I’d give the UV dry mask a shot.

I printed out the Yusynth Steiner filter twice on the same transparency. I was worried that one layer wouldn’t be dark enough by itself to sufficiently mask and most sources suggest using two layers. Rather than compose two copies before printing I just set acrobat to print the artwork on side side of the paper - then fed the transparency in again facing the other direction to get a second copy on the same sheet. This may have been a mistake and I probably should have just composited it before printing so I could print it in one go.

Why? Well. I think the heat of the laser process may have slightly distorted the first print. Because when I went to line them up I had a hard time getting them to register exactly. I thought I got it…but I later learned I didn’t. It wasn’t the only mistake I made though so it wasn’t the deal breaker.

Note how the upper left corner is a little doubled. I thought it was just from how I was holding it to take the photo as it had looked good when I taped the two pieces together. I figured once it was in a contact frame it would line back up again.

Against an improvised light box (my lighted magnifier) you can still see that corner out of registration.

But…I forged ahead. I did have a suspicion that I was making a big mistake at this point…but I couldn’t put my finger on what was giving me that feeling. Well…I did have an idea (and it was right) but I talked me self into believing I was wrong and that wasn’t the problem.

I opened the roll of UV dry resist in subdued lighting and cut off a piece a bit bigger than my board. I thought I peeled one side of the protective plastic off and taped it down to a piece of paper with the resist side facing up. I then put my board against that side and taped the board down to the resist. I put another layer of paper over the top of this and then ran it through the laminator on the 5mil setting 4 times - 2 with each side up.

I then opened the paper and found…that I hadn’t actually peeled the protective film like I thought I had. Whoops. It did peel now but I wasn’t sure if it would still bond so I just ran the two pieces of paper through without the PCB. One pass through was enough to get it to fully bond with the paper. Ok…so this stuff is a bit more forgiving than I expected.

Cut off a second piece…made SURE I peeled the protective plastic this time and repeated. It bonded to my board quite nicely this time! I then realized I couldn’t find my good contact frame and had to make due with my backup. Which I didn’t get quite as clean as I would have liked. But…this is still test conditions so I wasn’t sweating it too much. Plus again this artwork is slightly larger than the boards I have and getting the art and board to stay aligned as I put this frame together is almost impossible due to the design of the frame. So…I figured I’d just go with it.

Put it under the UV lamp I bought for curing my resin prints for 30 seconds.

I could see that it had definitely changed color from light blue to purple so was probably quite well exposed.

When I pulled it out of the frame and removed the artwork I could definitely tell it was well exposed…and realized that unless there was a big surprise in the developer my gut was right all along and I did make a big mistake:

To develop the resist you use Sodium Carbonate. About $5 at walmart in the laundry isle sold as “Washing Soda” (NOT Baking Soda - though you can make sodium carbonate from baking soda by baking it in a 200f oven for an hour or so.) for a box that will last approximately 10 lifetimes for this use since you use “2 heaping 1/4 teaspoons” per 1.5 cups of 89f water to make the developer.

I mixed up the developer…dropped in the board…and confirmed my suspicions. This made a great mask…but I had made two fatal mistakes…and another less critical mistake or two.

The biggest mistake…I should have inverted the artwork. I knew this…but convinced myself I was overthinking it. Nope, I was overthinking the overthinking. You need a negative of the artwork to expose the resist. Whoops.

The second mistake. That registration I was concerned about between the two layers of the artwork. You can see how the upper right corner still has the obviously slightly out of alignment look to it. If the artwork wasn’t inverted it would probably still make a usable board with this size/style of traces and pads. Wouldn’t work so well for smaller traces…but for this…should have been ok.

Looking at the lower left where I got the registration best things look much better:

Though - the dirt on my contact frame did cause a few issues here.

And you can see the other issue. I didn’t get the UV film on super evenly. One site suggested floating it in water when applying it to the board…I just kind of taped it down and taped the board to it. So I got a few wrinkles. Probably still a usable mask though and I’m not overly concerned about this little flub.

I won’t be etching this though since it would result in an inverted board with copper where there are no traces and bare substrate where there should be traces. Not very useful! It would tell me how well the resist does…but…I don’t want to waste a board to find out. Instead I’ll clean this board off and try again.

But I’ll either find a design that will fit on these boards - or take a day or two to design my own board to try. I haven’t done any board design in about 5 years and would like to try it again. KiCAD has come a long way…and I have a full license for Eagle now since I broke down and paid for a 3 year subscription to Fusion with their recent changes to the free license and since Autodesk bought Eagle they now include it with Fusion. And Eagle and Fusion now integrate really well so you can bring boards from Eagle into Fusion to design around them very easily which I kind of want to experiment with.

So while I didn’t get a board out of tonights test I can’t call it a failure since I learned so much. I confirmed that with good artwork I can get good prints that make great masks. I confirmed I can apply the UV resist to a bare board - expose it - develop it - and get a nice looking mask. And I learned not to send the same transparency through the printer twice.

This method does have a few more steps than the toner transfer…but overall I found it easier and more likely to give a higher quality more accurate mask.

Honestly my biggest frustration is that I can find 100x70mm boards from various vendors on amazon all day long…but the ones I got are actually 1mm short in the 70mm direction and yusynths designs are 100x73mm leaving me 4mm short of being able to make them on these particular boards without compromises. I only see on vendor selling 100x75mm boards and the reviews on their blank PCB’s aren’t very encouraging. So I may have to hunt around for a better option on blank boards. The 100x70’s I have are nice enough quality and I can certainly use them…just not for Yusynth’s layouts.

[juggle]

Oh…and before I clean this board off…figured I’d put it under the scope and take a close look at the resist:

The dirt on my frame:

Here’s where I had registration issues between the two layers of artwork:

A spot where everything was “just right”

And I took a look at the transparency as well, even this spot where I had good resistration looks like it was a bit off:

Though this spot looks dead on…no wonder it came out so well:

And some of the worst registration:

Also separated them and looked at single layer:

Honestly…looks like my printer does a good enough job I would have been ok with a single layer…I’ll probably try that next. If so it should give me the sharpest most accurate results.

[juggle]

Note to self…might want to try a bit less time under the UV lamp next time. Acetone barely touched that resist. Took some serious elbow grease with a scotch brite and a razor blade to get it off. Little worried it could lift traces using that much force to clean an etched board!

[juggle]

Did a bit more reading…apparently acetone isn’t the preferred method of removing this type of mask. Instead it’s suggested to use Lye to remove the mask. So…looks like I need to make another trip to the store

[juggle]

After a few days of re-learning kicad it looks like tonight I’ll be ready to try etching my first board I did my own layout for.

Honestly…I’d rather go with a proven design…but due to the size of the blank PCB’s I got I haven’t had any luck finding anything sufficiently small enough to fit. So I picked the M4TM 808 Kick as a simple enough circuit that I’d like to have in my modular and gave it a go.

Re-drawing the schematic was easy enough. I did have to learn how to deal with multiple symbol units for the TL074, I hadn’t done that in kicad before. The schematic is the same as the original - though instead of one cap across the positive and negative rails of the TL074 I changed it to one one each rail (not sure which method is more correct for decoupling an opamp…I’m guessing someone here more knowledgeable than me will chime in.) I also added some 10uf caps on the power rails since I had to add the 10 pin header to the schematic anyway. Only other real change was adding some headers for the in/out.

Due to the small size of the boards I have to work with I decided to just go with headers for the pots and jacks instead of board mounting anything. I had initially wanted to board mount the pots - but quickly realized that I just couldn’t make that work due to the board size. I could have done it by cutting some boards up and doing a few smaller boards…but that was an uglier solution than just doing it this way I felt.

Routing out the board was where I had to really re-learn kicad. I had mostly learned it 5 years ago when I was making power distribution boards for quadcopters and a hexcopter - but switched to diptrace because at the time kicad didn’t deal well with the board outlines I needed.

I’m hesitant to even show the layout I came up with as it’s pretty obviously a first attempt and can certainly be improved on in many ways. I’m really tempted to just start over again now that I’ve got the basics of kicad figured out…but…the boards I have were cheap enough I don’t mind wasting one to see how this goes:

Of course that’s just the top view that doesn’t show the traces I wound up with. Other than the half dozen top traces that I’ll be doing with jumper wires. Those were one of the trickier things for me to figure out in kicad. I wound up just doing them as normal top traces…but then modifying the vias to have larger holes and pads that match the through holes so I can solder wires into them easier.

I used the plot functions in kicad to generate an inverted mask for the bottom along with a non-inverted mask for the silkscreen (which I’m petty sure will do what I’m looking to achieve) and composited them in photoshop so I can print two of each on a single sheet of transparency material:

With the UV masking anything that’s black on the printout will not get masked once the mask is developed. So for the traces I need the traces and floods black so they will be protected while the white areas around them will be etched away. And…now that I look at this…I think I got it backwards…again. Ugh. I did get the silk correct…The black bits where the exposed pads are will be left unmasked while the rest of the board will be masked acting as a solder mask.

Thankfully since I’m using live links in photoshop fixing the reversed mask is as simple as re-plotting in kicacd and photoshop picks up the new version automagically:

There…I think that’s right. I’m still not going to rush to print it on transparency until I think this through a few more times

I also prepared a mirrored print of the silkscreen for the top - since this is a bare board just doing a simple toner transfer of black toner should result in a readable usable silk to aid me in assembly:

Neither the soldermask or the silk are at all necessary for this board. I’m just curious to see if I can do them at this point and it’s a small and simple enough board to experiment with it won’t use up a lot of materials to experiment.

I also gave up on finding my old PCB drills. I know I have them somewhere…but new drills are easy enough to find now and cheap enough that I just ordered a new set off Amazon. Well, 5 sets of 10 sizes for <$20 since it’s so easy to break drills this small. Gave a got at drilling freehand with my dremel on the first test board I etched and it went easier than I expected:

I didn’t drill them all, just enough to get a feel for it. And I really only flubbed one hole pretty badly…but even so I was still able to fit a dip socket in without any major hassle. Still debating if I want to drill this whole board and try building it or not…

Considering trying drilling with my drill press as well which should give me better control. But…the drill press doesn’t have as high of a spindle speed and for these tiny holes apparently that’s pretty important. Plus the press is a lot more hassle to setup since I don’t have room for it in my backroom and have to go out to my shop to use it…and even there I have to clear out space for it when I use it.

So tonight I’ll be etching again. If anyone spots any glaring errors before I do so I’d appreciate the heads up

I also got some double sided boards. But…let’s get a usable single sided board that works before I dive into trying to register two sets of artwork!

[fredrik]

@juggle

The schematic is the same as the original - though instead of one cap across the positive and negative rails of the TL074 I changed it to one one each rail (not sure which method is more correct for decoupling an opamp…I’m guessing someone here more knowledgeable than me will chime in.)

You should decouple the rails to ground, not to each other, so your approach is correct.

Datasheet:

[juggle]

Thanks for the confirmation @fredrik that’s how I’ve always understood it but the original M4TM schematic showed it across the rails and doing some searching I found a few discussions in various synth DIY groups that suggested that since the opamp has no reference to ground it can work just as well to put the cap across the rails.

The people who supported that method also claimed that caps between the rails and ground could couple noise from ground into the op-amp…but…that seems to misunderstand how decoupling actually works in the first place since the whole point of decoupling caps is to stop that.

I probably could have done a better job of getting them closer to the pins…and have heard a number of people who are far more knowledgeable than me say you may as well not even include them if they’re not right at the pins. (That was in regards to high speed digital circuitry though and they were arguing that even putting decoupling caps on the other side of the board but right under the component they’re associated with on the other side of a via is far enough away to defeat the point of adding them.)

Decoupling caps are one of those holy wars I really don’t want to get into since one thing I do know is that my knowledge is limited enough that I can’t take a stance on any side of the debates I was caught off guard by seeing something I wasn’t familiar with and since I’m new to working with op-amps wondered if there was something I wasn’t aware of that more experienced people knew from experience that contradicted general best practices.

[analogoutput]

This is audio, as long as someone mentioned bypass caps in the IC’s hearing it’ll be fine.

[juggle]

So…anyone spot the mistake in my “fixed” mask image yet?

I’ll get it right one of these times…

[eric]

The pads are “full” ?

[juggle]

Well, since nobody spotted it or wanted to call me on it. The mistake in the fixed mask image is that there are no holes. Apparently when you change kicad from SVG to gerbers and back to SVG it reverts to no holes (since gerbers use a separate drill file for holes) and I hadn’t noticed. I fixed that and exported again and got my mask then tried to make a board tonight.

Again, I learned so much from this I can’t call it a failure even though the board is even less usable than my first attempt at the yusynth board. This board I have basically no chance of building up due to a few of the mistakes. So to keep from calling this a failure let’s look at what worked, what didn’t, and what I screwed up so I can hopefully learn from these mistakes and not consider tonight a waste of time.

Speaking of time. It was just about an hour start to finish from when I printed the transparency until I “finished”. Note - I didn’t bother trying to do a solder mask or silkscreen. The error count here were just too high to bother. Those would add more time of course if I was to continue on to taking those steps.

First issue. Even printing both copies of my mask on one transparency that only went through the printer once - I had a heck of a time getting them to line up. I’m not sure if it’s these transparencies, or my printer, or just the reality of printing transparencies on a laser. I almost decided to try a single layer…but on the lightbox it was so obvious that two layers is MUCH darker I went ahead. I’m going to have to experiment with another brand of transparency and/or printing on two separate transparencies to see if I can eliminate this issue.

Second issue. Everything is just slightly undersized. I had seen a few writeups of this process that warned about this…but holding sockets and headers up to printouts and measuring them with my calipers it seemed that things were looking about right. About right isn’t right. The two mounting holes I placed are 90mm apart in kicad…but 85mm on the board I etched tonight. Ouch. But it’s hard to say how accurate that is since measuring the center of a hole is a bit tricky. But…everything is a bit undersized. The holes for example. The transistor footprint has 0.75mm holes. But my 0.7mm drill bit is clearly larger than the hole in the copper. Most of the other holes are 0.8mm and again my 0.8mm drill is clearly larger than the hole in the copper. The distance between two traces coming off of the DIL socket are 2.54mm apart. But on the etched board they measure 2.1mm

Measuring a single transparency shows that the error is definitely in the print. Layering the two and the slight difference between them only compounds both of these issues.

Next print I’ll try scaling up a tiny bit to compensate for the size issues. (I’ve seen 1.03 scale factor suggested and based on a few measurements that seems like 1.05 may be better.) And I’ve ordered a different brand of thicker transparency film with better reviews in hopes that it gives more consistent results.

I also see that my printer has a “reduce paper curl” setting. But I can’t tell what it does…might be worth trying though to see if it helps minimize distortion.

Anyway…

Third issue. The default design rules in kicad are probably a bit optimistic for DIY etching. Though…I think with issues one and two out of the way this would be ok. Traces where the two transparencies were aligned well came out great. But where they were off just a little bit I lost traces entirely. I need to resolve issues #1 and #2 before I can say just how big of an issue this actually is.

You know what though, that’s a wall of text. And I’m describing results I haven’t shared yet. Before I keep going through issues let’s take a look at how tonights process went.

Two improvements to my process. 1) I got a new lightbox that made lining up the two transparencies much easier. And my daughter loves tracing drawings so we both get to use it! 2) I found my good contact printing frame which makes it much easier to line things up and keep them from moving as I put them in the frame:

See that line across the inside of the frame below my board? That’s where the back is split in two with some hinges. So you can put something down…clamp half of it with one side and then fine tune alignment before clamping the other side. I didn’t need that tonight. This new design easily fits on my boards with room to spare (more room than I expected due to issue #2) so I was able to be a bit more sloppy than with the yusynth layout.

If you look closely at the image you can see a few flaws in the mask even at this point. Which is issue four. The bit of UV film I used wasn’t the best. It was from the very start of the roll and was a little damaged. This led to a few flaws in the mask. But I wasn’t overly worried at this point since due to issue #1 I was already expecting this board to be a “failure” and was mostly looking to see how well the mask did and if the traces this small were actually realistic.

This photo was just after exposure under my UV light. Which was also issue five. I need to start taking better notes. I thought I had only done a 30 second exposure last time. So that’s what I did again. But…I got radically different results and todays test really looked underexposed. It wasn’t nearly as purple as the previous test and I was worried the mask may wash away in the developer!

It didn’t…but the flaws in the UV film (issue #4) were starting to become very visible:

This just wasn’t a great piece of film - it had some knicks in it and apparently 2 passes through the laminator weren’t quite as effective as I thought they were. Also need to keep better track of my exposures. Better piece of film + 4 passes through the laminator + more exposure should result in a better mask.

You can see near the bottom where some traces are missing. I honestly expected that at this point due to issue #1. But…there was enough that did stick that I felt it was worth sacrificing a board to see how well the mask would hold up to these thin traces.

Issue six - I should have just made up fresh etchant. I read mixed things about re-using this particular etchant. The Muratic acid / hydrogen peroxide etchant apparently gets stronger as copper dissolves into it as long as it gets oxygen replenished. The acetic acid…not so much. Etching went a LOT slower than last time. After awhile it was apparent that it was taking too long…and the mask was starting to look like it was starting to fail. So I made up a fresh batch to finish it off. Next time I’ll just make fresh unless I try switching to the Muratic acid since I found I do still have some from our pool.

But it did etch…and the mask was much easier to remove this time. Probably partly due to the apparent underexposure and possibly the time in the etchant. I still had to do some scrubbing with a scotch brite but acetone got most of it off. And no traces were damaged by the scrubbing even though there were some very fine traces. So I may not have to pick up Lye afterall.

How were the results?

The top half of the board is pretty great. That’s where the transparencies aligned well. The middle has a few flaws which are from the damaged parts of the UV film - wouldn’t have happened if I had used a nicer piece of the film instead of the ragged end of the roll. The bottom…I lost some traces. Which I expected since that was where the alignment issues were worst on the transparencies and the traces were all but disappearing.

Still…I figured I’d go ahead and drill the board to see how it went…At this point I was tempted to try building it anyway since I could always just use some bodge wires to replace the missing traces.

Drilling was much trickier than on the Yusynth board. I had to be a lot more accurate because even my 0.7mm drill seemed too big for the 0.8mm holes. This was when I broke out the calipers and really discovered issue #2 from above.

I tried fitting a few components to see how bad it was…and I was able to get them to fit…but it was obvious that the scale was off just enough to cause some issues…and if I was off just a tiny bit on drilling the tiny pads meant I lost some traces. I was definitely leaning towards not bothering to try and build this one when I realized the biggest issue and had a flashback to my second module the noise drum.

It’s backwards. This is the bottom copper I’m looking at from the bottom…but it looks like I’m looking at it from the top. I had read that it was good to print the transparency mirrored so you can have the toner right against the UV emulsion for best accuracy…And I did have the toner right against the UV film…so…I must have got this backwards at some point.

The yusynth design is cruder…but…that means it’s more forgiving as well. The big pads and traces make it a lot easier to get a usable board even with some bit failures.

But - this test tonight did show that the UV masking is capable of extremely fine detail, and other than the mirroring issue none of the failures were really unexpected. The biggest issue is working out how to get accurately sized prints that are dense enough to make good masks. And it’s a problem that others have solved before so I’m fairly confident I can solve it as well. I rather strongly suggest that the generic cheap transparencies are part of the issue…and my workflow in plotting and printing is the other.