For the last two years, I’ve been supplementing that with a formal background from my university’s EE department, taking first one and recently two classes per semester. The education is interesting and enlightening, but it does take its toll — at least forty hours a week of work, six hours of classes, and (say) twenty hours of study and homework, plus some volunteer work unrelated to those, doesn’t leave me with a lot of free time; and you can see it by the decline in my hobby activity.

I don’t want to lose sight of what I love, though; and I hope to make a small business of electronics and make a few products available for sale. So this summer I’m devoting every spare moment to get some projects off the ground. And my ally in that plan is BatchPCB.

Why Have Boards Manufactured?

In the past, I’ve done a lot of circuit prototyping on breadboards. For some types of circuits, though, the needed prototyping has more to do with physical form factor and less to do with circuit validation. (I hope to show some examples in the coming weeks.) I’ve etched my own circuit boards; I’ve imposed on friends to mill prototype boards for me; and I’ve hoped to build my own milling machine to prototype my own boards at home.

The drawback of all of these methods is the lack of plated through-holes. I’ve heard of DIY hole-plating methods, but I found them to require a prohibitive setup for chemical processing. I’ve asked everyone I know whether they can think of any source for 1.5-ish mm (60 mil) OD copper or silver tubing, thinking of making a small riveting press to flare tubing onto the PCB surface both top and bottom — and even found very small silver crimp tubes used in beadwork and jewely-making, but none as small as 1.5 mm OD, nor in a consistently appropriate length.

I’ve worked around the lack of plated through holes by laying out boards that don’t require them, always carrying a signal from top to bottom using a component lead that can be soldered on both sides. But this means no vias (soldering pins top and bottom just to change layers is a pain) and only crossing layers at resistors and diodes. It means always routing connections to electrolytic capacitors on the bottom, ’cause you can’t get to the top side of the board to solder unless you stand the capacitors way up on their leads. It means routing traces to headers only on the bottom, or sliding the plastic guide up on the pins to solder the top side and then sliding it back down. It means a dozen little design compromises for a prototype board that don’t need to be made for a board I’m going to have commercially manufactured later. It means not only extra effort to accommodate my prototyping methods but also extra effort to undo that work before going to manufacturing.

SparkFun Electronics created BatchPCB as an offshoot of their own PCB prototyping contract. They aggregate orders from multiple users, tile them together onto standard-sized panels, upload the panels to Gold Phoenix, get the boards manufactured, receive the shipment from Gold Phoenix, sort out the boards, and send them back. They charge $2.50 per square inch, which is higher than you’d pay if you were ordering 100 square inches — but far less than you can pay anywhere else if you only want a few square inches of prototype. And they charge a flat $10 handling fee per order, regardless of how many designs you include in your order.

They suggest it’ll take about three weeks to get your order. My experience has been two weeks. It sounds like a long wait, but as they say:

As we develop projects, we always get at least one PCB design onto the week’s batch panel. While one design is being fabbed, we have new PCBs for another design already arriving from a previous batch – we always have new PCBs to play with!

My Summer with BatchPCB

I’m trying to place an order every two weeks and to order boards for multiple projects each time. I’ve received two batches so far and I submitted a third this weekend.

I’ll say more about these designs as I work on the projects, but starting at the top and progressing in reading order:

• Logic gate boards to help bridge the huge gap between what happens in the digital design lecture and lab. My classmates with no prior experience did not gain much academic value from poking opaque ICs into breadboards.
• A Freeduino-derived board for my personal use, manufactured as a proof-of-concept that I have the right components and schematic to embed an Arduino-compatible core into projects of my own. (I will of course publish all source files for designs that I distribute to anyone other than myself.)
• The first half of a pair of boards to breakout a headphone cable and jack for breadboarding, to plug an iPod’s headphone output into op-amp filter circuits and listen to the results.
• A board for testing component lead fit against through-hole sizes. To date, I’ve used calipers for measuring lead sizes for PCB design. I’m curious whether testing against a physical board gives me any different expenditure of time or quality of results.

The logic gate boards were my first batch. I ordered four each of two variants of the boards, assembled a few, and discovered that I don’t like soldering 0603 SMT as much as I thought I did; so if I make more, I’ll be changing the boards to use 0805 components.

Lesson learned: Order only one of your first prototype, regardless of how sure you think you are that you’ve finalized the design.

So why so many of the other boards? I did order only one of each … but SparkFun, bless their hearts, appears to fill wasted space in each panel with small customer boards that they give back to their customers as a bonus; and I happened to have small designs in this batch.

The idea is to expand on a simple LED tester by allowing the user to plug in an LED, dial in the LED brightness, and then read information on an LCD showing the LED voltage drop, the current current, and the value of current-limiting resistor to use in a target circuit.

A microcontroller determines this information by measuring the voltage drop across a series current-sense resistor to calculate the current and measuring the voltage drop across the LED to calculate how much voltage will drop across the current-limiting resistor in the target circuit and what that resistor value should be.

Variable Resistor Drive

Until now, all of my prototyping has used a variable resistor in series with the LED to set the current. After subtracting the LED’s forward voltage drop from the supply voltage, the variable resistor dominates the resistance of the remaining series chain (which includes the current-sense resistor), thereby setting the series current.

This does give control over the LED current and brightness, but the problems with this method are:

• A small-valued potentiometer doesn’t provide enough resistance to dial down to low enough LED currents. For example, a 1K pot with the circuit running on 9V won’t deliver less than 6mA, depending on the LED color (and voltage drop); and modern, high-efficiency LEDs are surprisingly bright at 6mA.
• A large-valued potentiometer has an extremely non-linear current response, with all the “action” at the very end of its rotation.

Here’s the response of two different LEDs with a 10K potentiometer:

Very slow response until near the end of the potentiometer’s rotation, at which point the response is so rapid that it’s very difficult to control
And of course this makes sense, as it’s the hyperbolic curve of I = V/R.

Transistor Drive

Last week I started looking at improving the range and linearity of the LED current. I’m not looking for a perfectly flat response curve nor for a true constant-current drive; I just want a somewhat better response. What came to mind was this simple PNP transistor circuit — actually an even simpler version without R1 and R3, but I’ll explain their purposes in a bit.

The theory is that R2 (or R1 + R2 + R3) acts as a voltage divider across the power supply, linearly setting a drive voltage. R4 (nearly) linearly turns this voltage into a current sink across the PNP transistor’s emitter-base junction; and because R4 >> R2, R2 presents a “stiff” voltage source to R4, meaning we can largely ignore R4′s effects on the voltage division.

Thus R2 provides (nearly) linear control of the emitter-base current. In the common-emitter configuration, the PNP transistor amplifies the current by the transistor’s β (about 150-200 for a small, general-purpose PNP like the 3906) for a correspondingly higher emitter-collector current

I_EC = β I_EB

which goes through the LED and the sense resistor, providing (nearly) linear control of the LED brightness by turning R2.

Well, that’s the theory, anyway. This weekend I dug out the prototype and built up the transistor control to test it in practice.

(70s decor courtesy Radio Shack.)

The first thing I noticed was a section at the CCW end of R2′s travel in which nothing happened, because R2 wasn’t providing more than the transistor’s cut-in voltage — that is, although V_B was less than V_E, it wasn’t enough less to overcome to emitter-base forward voltage drop and bias the transistor down into the active region.

I tried installing a small-signal diode “above” the potentiometer so that V_B would always be at least .6V below V_E and eliminate R2′s dead region, but the diode’s forward voltage drop was a little too high (it did too good a job) and the resulting minimum LED current was a little higher than I liked. I settled on adding R3 in that position, selecting 68Ω as a value that worked well with both traditional and high-power / high-efficiency LEDs and with both 9V and 7.2V supplies.

With a 9V supply and R3 = 68Ω, I tried three different values of the base resistor R4.

The table shows a similar effect at the other end of R2′s travel in which the LED current was pretty well maxed out and not increasing any further. I think I was hitting the knee between the transistor’s linear region and saturation, meaning increasing I_EB was no longer increasing I_EC. Experimentation gave me R1 of 200Ω keeps the transistor pretty well out of saturation and gives a satisfyingly more-linear response than what I measured here.

The 0mA readings at the beginning of the table, by the way, are a bit deceptive — some of my test LEDs are actually lit in that region. I’ve updated the Arduino code to show tenths of a milliamp when the reading is below 10mA, and I can see LEDs glowing with as little as .1mA. Probably not a value of interest for most people, but it could be effective for making flickering gas lamps for model railroads.

Choosing Values

R4 = 22kΩ looks like a pretty good compromise between providing a near-linear response and covering the range of LED currents I expect most people would be interested in testing, so I’ve tentatively settled on it.

I’m still fiddling with values to give good performance at both 9V (alkaline battery) and 7.2V (NiMH), because I use rechargeables almost exclusively and want to make this work well on rechargeables to encourage other people to do the same. The problem is,

V_supply = 7.2V
V_EC ≈ .8V
V_{blue LED} ≈ 3.5V

V_R5 = V_supply – V_EC – V_LED = 7.2V – .8V – 3.5V = 2.9V

I_LED = I_R5 = V_R5 / R5 = 2.9V / 100Ω = 29mA

In other words, running on a 7.2V battery, with the transistor saturated, a blue LED with a 3.5V forward drop maxes out at 29mA; and it gets worse with a battery that’s not straight out of the charger and some white LEDs with a higher forward voltage drop. I’d like to enable people to test up to 50mA, to cover high-brightness LEDs, so I’d like to push this maximum current a little higher.

R5 = 68Ω gives I_LED up to about 42mA, which isn’t as high as I like; but the tradeoff is that a smaller R5 gives me a smaller voltage range to sample in the A/D converter, hence lower resolution for the display. 68Ω seems like a good compromise. And I’m already thinking about a DPDT switch to change the resistor and alert the microcontroller about battery chemistry.

Saturday night the EasyBright already got its public debut! I played a pair of classic rock concerts Friday and Saturday nights, and Friday had trouble seeing my music (occasionally folded out to four pages) with the clip-on stand light I was using. Saturday after assembling the EasyBright, I built an LED “wand” music stand light that worked marvelously.

I didn’t have a lot of time for construction, so I cut a 1/2″ dowel to 3′ length and drilled eighteen 1/8″ holes through it (axially, not longitudinally) every 2″. Paint wouldn’t dry before the show, so I sanded the dowel and then colored it with a permanent marker. I then installed bright flat-top LEDs with a good viewing angle into the holes and bent the leads out in opposite directions.

I lap-soldered teflon-insulated (heat-resistant) wire from LED to LED with heat-shrink tubing preinstalled — but didn’t shrink the tubing until after I had tested the LEDs, in case I needed to repair any solder joints. I skipped LEDs to make an A-B-C-A-B-C pattern so if a chain failed, I’d lose light evenly along the whole wand instead of all in one section.

I crimped connectors onto the wires, connected everything to the EasyBright, put an appropriate connector on a 24VDC wall wart, and fired it up perfectly on the first try. (Such luck!) I disconnected the wand, reseated and shrank the heat-shrink, zip-tied the wires in place, and then powered up the wand to burn in for an hour before leaving home for the concert.

On stage, it delivered a very even wash of illumination across my music, giving me a great view all through the show.

The circuit board is so lightweight, it was comfortably suspended in mid-air between the keyboard rack and my music stand by the power and LED wires. For the long term, I’m trying to decide whether how it should be mounted to the wand — perhaps attached near the end inside a sleeve of giant heat-shrink.

Last week while watching Mannequin (a very young and fresh Kim Cattrall, a goofy plot, and music by Starship — what could be better? okay, if it had John Cusack and were set in Shermer, Illinois, yes, that would be better) I split all the EasyBright components into a parts bin for easy access and portability.

Saturday afternoon I put together the first sample.

This is waaaaaay too much solder paste for 0603 parts and 1/40″ IC pin spacing. I had to remove several solder bridges from the IC, and the passives had solder mounds instead of fillets. I took the picture specifically to record how much paste I used so I could adjust on the second attempt.

Here’s the cleaned-up board, front side.

Back side, with hand-written labels for the current rating and the serial number (S00). The “permanent” marker comes off easily with rubbing alcohol — I need to get some clear nail polish to seal it in.

Changes

Even before assembly, I had made notes about (and started implementing) things to fix whenever I print the next boards:

• Change the IC’s ground connection from a via outside the IC footprint to a trace going straight in to the heatsink pad. I had routed that connection before I confirmed with Maxim that the pad is okay to connect to ground — it’s just not okay to be the only ground — and then forgot to go back and change it. Removing that via gives me a little more room to route the bottom-side LED power traces cleanly, and also:
• Increase the pad size on the optional through-hole current-sense resistors. This, believe it or not, is EAGLE’s default pad size, and I think it’d be challenging to solder without a good, narrow-tipped iron.
• Increase the trace isolation on the solder-side ground pour. There’s no reason to have it that close to the pads.
• More subtle, I spaced the 2-pin connector pads an extra .02″ apart to see whether I could get the connectors to friction-fit for ease while soldering. They don’t quite. Either change the library footprint to space the pads a little further apart or just get used to pinching the leads together before stuffing the parts and soldering, which works better than I had expected.

I’m still delighted!

While shopping around for board houses, I had narrowed my choices to two. Here’s how I made the final selection of Gold Phoenix — not, as it turns out, whom I thought at first I was going to pick.

PCB Express

PCB Express was initially my strongest contender. They offer a range of options, including copper only (no solder mask or silkscreen), two-layer with silkscreen, and four- and six-layer boards. They also partner with Screaming Circuits (read their blog for amazingly insightful, real-world ruminations on how to [and not to] design boards for manufacturability) for PCB assembly — something I don’t need now but might want in the future.

PCB Express’s pricing is a matrix of board size and quantity, with pretty appealing prices for moderate quantities of largish boards. They don’t mind if you panelize (tile a plane with multiple copies of your board — some board houses refuse to accept panelized orders) and will even route perforated lines to break apart your smaller boards, but won’t route complete cuts to separate your boards for you.

The EasyBright came out to be 1.75″ x .8″, and I wanted either about twenty (test/samples) or about a hundred (first production run) of them. My “E2″ order would have cost $166 to get two boards without panelization and with double-sided silkscreening ($83 per board) or the same $166 to get 12 panelized boards that I’d have to break apart ($13.83 per board). $478 would get me 90 panelized boards, again with double-sided silk — about $5.31 per board.

Gold Phoenix

Gold Phoenix was my underdog, in part because I hadn’t studied their pricing and capabilities closely enough. They have a prototype offer of 100 square inches for $100 or 155 square inches for $110 — each for two-sided boards with solder mask and top silkscreening. This rate includes shipping.

It also includes free “step and repeat” — that is to say, they will automatically panelize your design for you; and although this took me a while to figure out for sure, they cut them apart for you, too. Bottom-side silkscreen is an additional $20, as is electrical testing to discard boards with manufacturing defects impacting the copper connections.

So for $130, I could get 155 square inches worth of boards, plus $20 for electrical testing so I didn’t have to worry about clinkers. I guesstimated a kerf of .1″ to cut apart the boards, so the 10″ x 14.5″ panel they mention should give me 11 x 7 = 77 or 5 x 16 = 80 boards — in the neighborhood of $1.88 per board. That sounded fantastic!

Ordering from Gold Phoenix

To place my order, I ran an EAGLE CAM export and zipped up the files along with an index listing what layer each file represented. I then emailed the zip file to goldphoenixtech@yahoo.ca, indicating that I wanted 155 square inches of panelized boards with standard PCB material and copper weight, green solder mask, and double-sided silkscreen.

I got back a prompt quote for 110 boards instructing me to PayPal them, which I did, and my order was underway. 110 boards — obviously they cut the boards apart with a smaller kerf than I had envisioned.

And then after my order was placed, Asmodeus noticed that I had dropped a trace on one of the current-sense resistors while I was playing with eagle3d. Drat!

I quickly corrected the mistake in EAGLE and emailed Gold Phoenix asking whether they could restart the process with my fixed file. They had already “processed the job,” but for a $50 fee were willing to start over. That was a no-brainer — 110 boards I’d have to fix by hand or fifty bucks to fix them — so I sent the money and they started over.

Boards from Gold Phoenix

Just over a week later, they sent me a note indicating that my boards had shipped, and I received the package a few days after that.

113 adorable little boards, all cut apart and stacked and sealed up in plastic (and packed carefully in a box, not shown here). Woo-hoo!

Wait, 113 boards? I was estimating 77 to 80, they quoted me 110, and they delivered 113? What’s up with that???

Entirely ignoring the kerf and envisioning the 10″ x 14.5″ panel they list as their maximum size,, the .8″ x 1.75″ boards could fit 12 x 8 = 96 one way or 18 x 5 = 90 boards the other way. Neither one of these gets me 110 boards, much less 113.

Let’s go about this another way. 113 is prime; so if we assume that they made me a single rectangular panel, they must have discarded some boards after electrical testing. How would nearby numbers factor and what panel sizes would that imply?

• 114 boards = 6 * 19
  6 * 1.75″ = 10.5″ and 19 * .8″ = 15.2″
  10.5″ x 15.2″ = 159.6 sq-in (plus kerfs)
• 115 = 5 * 23
  23 * .8″ = 18.4″, which I rule out since they claim the maximum dimension is 14.5″
• 116 = 4 * 29, which gets us nowhere useful
• 117 boards = 9 x 13
  9 * 1.75″ = 15.75″ and 13 * .8″ = 10.4″
  15.75″ x 10.4″ = 163.8 sq-in, which is quite a bit larger than promised

So I don’t know quite what to think — other than that I like getting 113 boards at $1.33 each (plus my $50 screwup).

Quality

All of the boards I’ve examined look really good, and I was delighted with the legibility of the silkscreen and my fairly small type. The top-side silkscreen registration was low by about .01″ (click one of the pictures above), which makes a bigger difference to the appearance of the board than you might guess, but it’s quite acceptable.

A sample board measures .796″ x 1.748″ — certainly close enough for me. Its upper left corner is cut in a bit of a diagonal, I’m guessing the last place the router touched when it cut loose. I should point out that I’m mentioning every little thing I’ve noticed while looking at the board with a critical eye for several days.

All in all, I’m tremendously pleased.

Thank you, thank you, thank you SparkFun for emphasizing how important it is to view your Gerbers before submitting (search in page for “Something I highly recommend”).

I had already rendered the board layout in eagle3d (pay no mind to the test holes I used to figure out that eagle3d doesn’t make holes through copper that’s part of a polygon, which its documentation clearly states if you bother to read it) and thought it looked pretty good.

But I followed Nate’s advice, installed Viewplot, and was rather startled to see what my board was really going to look like.

The silkscreened boxes for the plastic “keepers” on the pin headers are fairly faithful to the connectors I’m using, but what’s with putting the pins in the silkscreen? I don’t need a shadow of the pins on the board.

Also the 0805 SMT resistors and capacitors don’t have silk around them showing which pads belong to the same component. It’s easy enough to figure out on this small board, but I’d like to develop good habits.

EAGLE’s CAM processor lets you pick which PCB layers are used to generate each Gerber file, and there’s not much in my top silk layer. The pins must be in tPlace, so I used the PCB layout editor to preview what would happen if I turned it off …

I lost the header outlines as well as the pins, as well as the outlines for the optional through-hole resistors. That’s not gonna work.

Exercising outrageous optimism, I tried changing my header packages from right-angle to vertical, hoping they’d have more appropriate package outlines (still in the tPlace layer).

Better — no stray pins in the silkscreen. Now guardedly optimistic.

I turned off the tDocu layer that shows the physical outlines of components, and which isn’t (and mustn’t be) included in the silkscreen Gerber because it would interfere with soldering, then added lines in tPlace to indicate the edges of SMT components.

Silly me — I thought I was done!

Back in Viewplot, look at how the silkscreen around the power connector is now crowding the pin labels. Aaargh! There’s not room to move them far enough away.

I went back to my connector library and designed two- and three-pin versions of “locking” connectors based on the SparkFun locking connector concept. (See footnote about SparkFun EAGLE library license terms.)

The Gerber view looked pretty good now, except the ground symbol was too close to the power header and a little visually confusing.

I moved that down a bit and in my next trip to Viewplot discovered how to get the drill holes to show up: don’t load the drill rack file into Viewplot, just the drill file itself. Getting visualization with holes and confirmation that there’s really a mounting hole through the heatsink — outstanding!

Somewhere around this time I also printed out the solder-side of the board to make sure that the boxes I made to write in (visible in the next screen shot) were large enough for me to write in. They weren’t. I enlarged them.

Done now, maybe?

Oof, look at all the problems with the silkscreen on the back side. The top line looks like I’m incrementing V by 5.5-40V (C joke), and if I fix that I need to move I_LED‘s = further away also. The / in the URL is awfully close to the solder pad, and the box for me to write the “factory”-configured LED current could stand to extend a little closer to the “mA” label.

Props to Nate at SparkFun again (search in page for “Label everything, all the time”) for reminding me to put the input voltage range and output current rating on the board, BTW.

Fixed! Really! Done! If I stare at this thing any longer, I’m going to start hand-kerning the vector font.

I zipped the files and uploaded them to Gold Phoenix last night.

SparkFun EAGLE Library License Terms

I’m not using SparkFun’s EAGLE library nor a derivative of their library file because I haven’t got a response from them whether their cc-by-sa license is intended to be:

• like the GPL, meaning that if you use their library in your board design you have to open-source your whole design — which I will do after I’m confident the design is right but not immediately upon shipping — or
• like the LGPL and you can use the library in your product without open-sourcing your design but you would have to open-source derivatives of the library itself.

So my connector library is most definitely based on Pete Lewis’s idea to skew pin positioning from side to side to make a header friction-fit in a board for soldering, but (as far as I know) the idea is not patented and I’m in the clear. My library is a reimplementation from scratch of the idea, so is not derivative of their library.

I hate playing games like this and I would love to toss my library and use SparkFun’s if I can get a verdict from them on the licensing. Also as soon as I’ve got boards tested and working and I’m ready to publish the design files, I can switch back to their library too.

Update 07-May-2010: I heard back from SparkFun and it will be fine for me to use their library. I’ll look at switching back on the next iteration of the board.

I’d like something whimsical but which still relates to its function as an LED driver. Fun hobby electronics names I love: Adafruit, MintyBoost, BlinkM, SparkFun, MakerBot, and CupCake.

I’ve considered Illumerator, Illumifier, and variations Lumerator and Lumifier (which is probably ^TM and a bad idea). Whatever I settle on will have -3L appended, to distinguish this 3-string linear driver model from the -1S switching model I want to do next.

So I welcome suggestions for great names. I’ll be happy to send you a couple of drivers if you’re the first person to suggest something I end up using.

Warning: I don’t care for variations of my name that feel like they came from the “makin’ copies” sketch.

So for the maximum 100mA, R = 2.03Ω; for 20mA, R = 10.15Ω.

I’ve laid out the board to accommodate both SMT and through-hole sense resistors. I’d like to offer boards preassembled for at least 100mA and 20mA and boards without sense resistors so you can provide your own resistors for custom currents.

Should I offer other preconfigured currents besides 100mA and 20mA? Do I need to bother with 15mA?

Let me know in the comments what currents you’d like to have. Just because you ask for something doesn’t mean I’ll provide it; but if enough ask for the same thing, I very well might.

The project, to refresh your memory, is a linear driver for three strings of LEDs, up to 40V and 100mA per string. You supply your own power (5-40VDC), LEDs, and wire.

After spending two hours looking through 250 pages of connectors in the Digi-Key online catalog (I am not kidding), I’m distressed to discover that the least expensive connectors I can find are going to cost about $2 per board. Feh! But I want the power and LED connectors to be polarized, and that’s more expensive than plain ol’ .1″ headers.

Here’s a mockup (clickable for the big pic, like all the photos here) of an early prototype PCB with product photos of the connectors I’m leaning toward. On the right, a six-pin header for three LED strings and a representative two-pin plug that’s polarized by virtue of the semi-latching bump. On the left, a two-pin header for power and definitely-latching matching plug.

Note that the male connectors are on the board to reduce the opportunity for shorting what’s on the ends of the wires while they’re unplugged and flailing around.

Questions:

1. Is it reasonable to expect people to plug three two-pin connectors into the right places in a six-pin header?
2. Is it worth making the power connector different from the LED connectors, to make it harder to misconnect things?
3. How many people are going to crimp pins onto the wrong wires and have to start over?

Opinions in the comments, please.

If you think you can find me less-expensive polarized connectors with .1″ spacing (I want to offer the driver with straight male headers for breadboard friendliness) that I can buy from a supplier as reputable as Digi-Key, please provide direct links to at least the header and ideally the plug and crimp pins. Free LED driver if you actually come up with something I like better.

If I need further input, I’ll mock up another picture with the latest plan and start a new post. Thanks to you all, and comments are now closed!