Project Logs

Collapse

• #11: Design deficiencies, and what you can improve

  sky-guided • 08/28/2024 at 00:49 • 2 comments

  I'm considering this project Completed, at least for the time being. I'm incredibly proud of what I've created -- every design goal was achieved, and it looks even better than I'd dared to hope.
  
  However, this is far, far from perfect. If I were to continue this project, I have a real hit-list of what could be improved.

  Overall System

  The biggest issue is that controlling drive intensity by adjusting mosfet feedback bias voltage is janky and unstable. A proper system would leave a bias voltage set with a trimmer, and use high-side limiting on the primary input current for primary drive control. I realized this kinda late in the design process, and wanted to reach "done" rather than do another round of prototyping.
  
  Also, having a breakout board for the USB-PD trigger module is pretty inelegant. Having a trigger module on-board would be better, but when I started this project I was having trouble finding a suitable IC with decent availability in the US. Several folks have since suggested good candidate ICs. 

  MOSFET and driving

  A straightforward error is that the gate-drive resistors have been undersized. The design shown in the video uses two 10-ohm 1206-size resistors in parallel (per mosfet), and they get incineratingly hot if there's no ionization/plasma to load the circuit. For the published release version, I increased those to a pair of 1-watt 2010-size resistors which should hopefully fare a bit better.
  
  I probably ought to be using an asymmetric resistor / diode pair on the gate pins. Almost wish I'd investigated that sooner since the footprints are already there, but, hey, like I said I'm declaring this project Done.  
  
  

  More generally: I'm convinced my mosfets are running hotter than they really ought to be.
  
  I still don't feel like I have a sufficiently deep understanding of mosfet switching behavior, especially when driven from a sine-wave rather than a proper square wave. Using two parallel mosfets for the final version was a callous attempt at reducing heating. In prototypes, heat was observed as being heavily dependent on how much current I was drawing and that made me hope that parallel would help, but I also should have known that I was looking at switching losses rather than anything Rds(ON) related. I did some quick math before going parallel which made it seem promising, but like I said there's aspects of mosfet switching that I'm still learning. 
  
  The most-correct version of this circuit would probably use an actual mosfet driver IC, with something like a PLL as part of a feedback loop. Would be lovely to have both a nice square-wave drive and finely adjustable phase offset.
  
  Using something like a GaN mosfet might also perform a lot better!

  Arc Start

  For the first start when the glass vessel is room-temperature cold, button-press arc-start reliability isn't as good as I'd like -- sometimes it does take several tries to get the toroid running.
  
  This particular arc start method is too expensive. If I were designing for even small-run production, it'd get a major overhaul.
  
  Both arc-start transformers and the GDT are all what I'd call "kinda-specialty components" which isn't ideal even if their BoM cost is modest. More seriously though, the flyback controller chip LT3420 is both expensive ($6.75/ea at qnty 10) and has limited availability. The good news is that the fancy chip is probably completely unnecessary. I initially thought I might be using the advanced functionality like keep-charged and the "done" signal, but the system I ended up with would probably be just fine with a fixed-duty-cycle flyback charger.
  
  There's a lot of different ways of getting pushbutton arc start though. I know one person who reports success using a cheap aliexpress "arc ignitor" module with one leg of the output tied to ground and the other pressed up against the globe. Another promising idea which was suggested to me would be to have an entire autotransformer-style secondary winding driven by...

  Read more »

• #10: AND IT IS DONE

  sky-guided • 08/25/2024 at 22:31 • 0 comments

  Thanks for following along :)
  

• #9: Black FR-4 and transparent soldermask, by PCBWay

  sky-guided • 06/01/2024 at 21:33 • 1 comment

  For the final build of this project, I wanted something special.

  Typical PCBs use a olive-green fiberglass core, layers of copper, and a (traditionally green) soldermask layer to protect the copper and define areas where solder/component pads should stick.  This board instead has a dyed-black fiberglass core and a transparent soldermask layer. This means that all traces are visible as their natural copper color, which is delightfully high-contrast against the black substrate. Component pads and other areas left uncovered by the transparent mask are plated (ENIG) gold. 
  
  This order of circuitboards was (partially) Sponsored by PCBWay! They covered about 60% of the cost of these boards, and I paid the remainder out-of-pocket.
  
  There's very few photos out there of real PCBs made using this stackup, so I'm taking the time to write out a review.
  
  (Transparency: being partially sponsored will likely bias my opinions, but I did also pay enough to feel like I have skin in the game. I have not given PCBWay any editorial control and they will not have seen this review prior to publication. Photos have been lightly post-processed and are broadly representative of how things look irl under good lighting. An order number [visible on the unpopulated board] has been blurred out for privacy.)
  

  An in-depth look

  This is going to be mostly photos.

  Here's how the board looks straight out of the packaging

  That looks really good, yeah??

  The copper is a matte tan, not shiny at all. ENIG gold finish looks like it always does, and creates a neat kind of sandy two-tone.
  
  For scale, the smallest traces on this board are 8mil with fairly generous spacing and the finest pitch component is 0.5mm. Components are D2PAK through 0603.
  
  

  There's a sort of white ghostliness visible around the copper areas -- I'm guessing due to the transparent soldermask being not pressed quite as deep into the fiberglass. Under certain kinds of lighting, a faint weave texture of the fiberglass can be seen.

  I absolutely love the way that text looks in copper. (This example has a character height of 1.2mm.) Super sharp! There's also a subtle dimensionality visible due to the thickness of the copper and the drape of the overlaying soldermask. 

  Text in copper+soldermask also looks nice and sharp (although is some slight cruft visible in the notches on the N and W). Contrast is more variable than copper-only as lighting changes, as the gold catches ambient light and reflection differently than the semi-gloss transparent mask. 
  
  The grey silkscreen ink is a bit mis-aligned from the other layers. This photo above shows it especially clearly: take a look at the holes on the left side and the silkscreen circles are distinctly off-axis.  Silkscreen alignment has been better on my other PCBWay boards, which makes me wonder if the layer required some kind of hand-alignment that's less automated for their higher-order-volume stackups. Not an actual problem but definitely a bit unfortunate.
  
  EDIT, ADDENDUM: My PCBWay sponsorship point-of-contact was perturbed by the soldermask alignment and raised it as a possible issue with their production team. In email correspondence with one of their customer service reps, PCBWay stated that while white silkscreen is aligned by machine, other colors (including grey) are aligned manually and "there will be +/-6mil offset which can not avoid. QC is based on the standard that the silkscreen are not on the pad."
  
  Their Advanced PCB overview lists silkscreen offset as ±2mil, which is presumably representative of their alignment standards for white silkscreen.

  Here's the board backside with its large ground plane. They grey silkscreen QR code is plenty legible against the black FR4. (White silkscreen would be more typical against black, but I thought grey would be a bit more elegant and am happy I made that choice.) The bottom silkscreen is equally mis-aligned as the top....

  Read more »

• #8: A Unified Board

  sky-guided • 05/27/2024 at 18:08 • 3 comments

  It's been a busy couple of months behind the scenes -- this project is basically finished! I'll be breaking the final few updates into a couple posts, while I wrangle loans of camera equipment for the beauty shots. 

  Circuit Unification

  Taking the different prototyped modules and banging them together:

  (you might have to click to embiggen)
  I've paralled the mosfets for better heat dissipation. This doesn't help quite as much as one would like (since a lot of the losses are switching and not resistive), but it does help. The mosfet gate resistors are also paralled since they get quite hot before the arc is struck. I'm not sure why that behavior is occurring -- if you know, drop me a line!
  
  The 4.3V zener diode (D5) for regulating gate bias is a carefully chosen value for this particular mosfet and circuit. Cranking the knob full-clockwise guarantees oscillations start, but doesn't set the bias so high as to cause excessive power draw and heating. The use of full knob range also makes adjustments more precise and controllable.

  A bank of four 33pF capacitors is a balance between heat distribution and board space. While the board itself has plenty of room (see below), I ran into some odd layout constraints due to things like the under-board heatsink mounting holes. The cabs do get pretty hot before the arc is struck, but are plenty cool during operation. 

  There's a total input bulk capacitance of 30μF (shown as two caps on main power and one before the 12V regulator). Less bulk capacitance caused occasional power supply brownout/shutdown.
  
  The oscillation sense coil from Update #6 has been brought on-board, shown in the dotted area within the main driver section. 
  
  Arc start is basically the same as discussed in Update #7. I keep waffling on whether a GDT value of 300V or 470V  is preferable here -- 470V may be shown, but keep your mind open.

  
  The only real new addition is the overheat sense. There's nothing special here; it's almost exactly per the datasheet for the TMP708 IC. The one bit of oddness was having to add a small decoupling capacitor in parallel with the programming resistor, because the overheat IC was placed as close as possible to the mosfet drains and it was picking up EMI. Debugging that was fun -- while poking around with scope probes, I noticed that merely touching the sense pin fixed the problem.
  
  Here's a delightfully goofy photo from a prototype board, showing a 0603 capacitor bodged directly on to the IC pins:

  Programming a trip value of ~56C was about right for raising alarm when the mosfet case temperature hit 100C. 
  

  PCB Design

  An overview:

  Details might not be super visible at this far out a zoom, but here's the overall layout. Power input and status LEDs are far left, overheat sense is lower center-left, main driver is center (with knob lower center-right), and arc start is far right. The heatsink is mounted to the board bottom, outline indicated  by the cyan square around the center area.
  
  Most of this layout is relatively low-density. The minimum size was constrained by needing to fit a heatsink underneath which could accomodate a 40mm fan. There's a handful of components which are 1206-sized even though 0603 would have worked, simply because that larger size looked better. 
  
  The rounded traces were made using the KiCAD plugin "round traces". 

  
  MOSFETS were placed on a large copper fill, stitched to the heatsink on the bottom with a lot of thermal vias. The thermal tab on the mosfet is attached to Drain (and not GND), so rather than having truly a massive cut in my backside ground plane for the backside Drain copper pours, I split the drain pours into smaller sections with cruciform GND traces. 

  
  The TMP708 overheat sense IC was put within a small cutout on the thermal fills, to maximize thermal coupling. A SMD thermal jumper would have been "better" but I didn't want to add it to the BoM. 

  The main driver section...

  Read more »

• #7: how to ionize xenon: the arc start module

  sky-guided • 03/23/2024 at 19:24 • 0 comments

  The electromagnetic fields generated primary inductor aren't strong enough to ionize neutral xenon gas into plasma. Once any infinitesimal part of the gas is ionized, those charged particles almost instantaneously cascade into their neighbors and the whole globe lights up. But getting a bit of gas to ionize in the first place is the tricky part.
  
  The "standard" way to do this (as demonstrated by BacMacSci) is to forcefully twist the glass globe against the induction coil. The triboelectric effect causes miniature "static electricity" shocks which are enough to kick off ionization. Alternately, one can use some kind of external high voltage source like a handheld tesla coil zappy gun.
  
  Either manually twisting the globe or using an external device is a bit dissatisfying to me. I wanted to integrate arc start directly into the device, activated by the push of a button.
  
  This log is going to be longer than most, because it's been an adventure. 

  Read more »

• #6: status LED, galaxy brain style

  sky-guided • 03/22/2024 at 16:20 • 1 comment

  Since a key goal of this project is to run standalone -- no o-scope, no bench PSU, no amperage panel meter -- it's crucial to have some sort of indication of whether the circuit was in oscillation or just sitting idle.
  

  Overall power draw reliably reflects the circuit's state. When not oscillating, there's near zero draw -- just LEDs, leakages, etc. While running the draw is more like 2-3 amps, and can be reduced down to as little as ~1.2A by lowering mosfet gate bias. 

  The normal approach

  I had initially planned to use the pretty typical current monitoring method of a differential op-amp measuring voltage across a low value shunt resistor. Since I'm looking for a threshold current indicator rather than a continuous analog  signal, the op-amp is fed into a comparator, against a reference voltage.
  

  Yes, this diagram is kind of sloppy, but you won't need to linger on details here anyway. If the text is unreadable due to aliasing you can click to embiggen.

  The TSM102 IC seemed like a neat combo-wombo of packaging two comparators, two op-amps, and a 2.5V reference all in one. It's also relatively inexpensive and can run off of a 40V Vcc. Purpose-built current monitor ICs also exist, but I didn't necessarily want to add a 5V regulator just for one chip, plus I thought the extra op-amp/comparator channels might also be useful for other functionality.

  
  As mentioned in the previous project log, PCBWay sent me a sponsored board and solder stencil! Assembly was quick and easy. Thanks, PCBWay!
  
  

  Well?

  It didn't work.
  
  Output voltages from the op-amp do move in response to changes in the drive power, but not in any way that's intended. I suspect that the choice of 25mΩ shunt resistor is just way too low for this application, and is getting drowned out by common-mode influences and amplifier limitations. Using 130kΩ resistors as part of the op-amp network was definitely a warning sign that things might be getting a bit too off-road. 45mV seemed an ok enough differential when I thought a 1.8A current threshold would be good, but lowering that threshold down to 800mA / 20mV certainly didn't help. (Or maybe I'm just not very good at this, and there's a fundamental design error that's gone unnoticed.)
  
  That's all ok though, because I had a better idea.

  Read more »

• #5: Version 0.3 is calm, cool, and collected

  sky-guided • 03/11/2024 at 04:20 • 6 comments

  Revision 0.3 of the induction driver creates a plasma toroid that's stable and controllable, and does so without overheating.
  

  I'm closing in on a finalized version of the driver. Soon, both the induction coil and driver circuit will be unified into a single monolithic PCB.

  
  
  This round of prototyping was...
  
  

  Sponsored by PCBWay!

  Entirely to my surprise, a representative from PCBWay contacted me with an offer to cover the costs of a batch of boards.
  

  Four boards! Three solder stencils!
  
  (This project log will cover design updates to the induction coil and driver. The other two accessory boards will be featured soon, but there's still a few circuit details I need to refine first.)

  Some stuff I like about PCBWay:

  - Fast. Eerily fast, sometimes.

  - Inexpensive!

  - Their instant quote functionality can be super handy for design planning, even before a board layout is finished. For example, it's great to be able to test out possibilities for things like "how much would it cost to make the board 50mm wider? What if I want to use 2-oz thick copper?" Etc.

  - Quality is solid. Admittedly I haven't plumbed the depths of BGA or wafer-level-packages yet, but for everything I've done so far I've been totally happy with PCBWay boards.
  

  - If there's something malformed in the manufacturing file, a member of their production team will email asking for clarification of design intent. This has happened to me twice!

  - I like they that sponsored me. I'm saying that very sincerely, not just because the monetary windfall is nice (although it really is). It's incredibly endearing that PCBWay is seeking out and financially supporting small, independent projects like this one, and not just big-name makers and youtubers that already attract 100k+ eyeballs. 

  Assembly

  Just like previous versions, solder paste was smeared across the laser-cut stencil, components were placed with tweezers, and the board was reflowed on a budget hot-plate.

  (Kapton tape was used to keep the board in place during assembly, but not during reflow.)
  

  Minor soapbox time: I spent quite a while as an electronics hobbyist being intimidated by surface-mount parts. Secret is, SMD assembly (at least for relatively large components) is faster and easier than strip-board and through-hole. This is especially true with a solder stencil and hotplate (or presumably toaster oven) reflow. Through-hole parts are decidedly obsolescent, and these days most interesting ICs are only available in surface-mount. For this project, using an actual PCB is also hugely important for minimizing parasitic inductances.

  With a larger inductor coil (115mm center-span x 12mm width) and some neater connecting wire bundles, the assembly looks pretty slick!
  
  Here's what changed in driver version 0.3:
  

  Read more »

• #4: tweakin'

  sky-guided • 02/25/2024 at 21:36 • 0 comments

  Spent a big chunk of the weekend adjusting a variety of component values. 
  

  
  Here's the waveforms with the circuit unloaded (no toroid), at 18V supply:

  Yellow is Gate (5V/div), purple is Drain (50V/div), blue is feedback network bias input at TP5 (5V/div).

  Looks decently healthy to me. I suspect that the gate protection zener diode is doing its job and helping to keep drive voltages within operational limits.
  
  Here's the traces when the xenon is fully toroid-ing:

  Hm. Clearly the device is overall functional but these waveforms aren't as clean as I'd prefer. Maybe it's fine? [EDIT: as before, turned out a lot of the wiggliness was from bad probing. Shorten those ground connections!]

  
  All right, so let's walk through the component changes. I'll spare the step-by-step of each individual test -- most of it was poorly documented and I was going for more of a "better or worse?" approach than robust characterization. 

  
  
  First off, primary tank capacitance was increased 99pF to 141pF. This was the first change made, as planned in previous project log "i would like my circuit to not cook myself". 
  
  Increasing capacitor value was broadly successful -- overall power draw was reduced to ~2.9 - 3.2A depending on circuit configuration. Yay!
  
  Decoupling cap C4 was removed because it caused some kind of runaway oscillation on both gate voltage and power draw. A different cap value might be fine but for the moment, the pads are empty. [EDIT: the problem was that a cap here caused input power to ramp up faster than supply power could keep up with. The solution was more bulk input capacitance!]
  
  I spent quite a while trying to increase operating gate drive voltages. You can see in the second pic that gate drive peak is only 8V, when I'd have preferred something closer to 15. My understanding is that MOSFETS really prefer to be driven by a square wave rather than this sinusoidal feedback. With intermediate voltages near the switching threshold the mosfet is in a state of "kinda-on", which causes a lot more power loss (and thus heat) than being fully on or fully off. 
  
  I was hoping that larger values of R2 would strengthen gate drive and increase the differential between the gate voltage and the (still too wiggly for my taste) bias voltage. Initially I spec'd R2 for 10k in this iteration, but it turned out that overly large values resulted in too tenuous of a ground reference. The max stable value I tested was 3.3k for R2.
  

  Similarly, decreasing gate capacitor C3 to 6.8nF increases gate drive voltage a bit, but it's not as dramatic a difference as I'd have expected. 
  
  Playing with the gate resistor Rg was a bit odd. Looking at the traces I saw very little change with values ranging from zero ohm to 10Ω. Subjectively, 0Ω seemed to cause the least heating of the mosfet, but I wasn't taking rigorous enough measurements to say that with certainty. (Also, the 10Ω resistor almost immediately toasted itself; the gate capacitance is slurping far more power than a little 1206 smd can handle.) I wasn't able to observe any kind of difference in probable turn-on delay based on scope measurements.
  
  Looking back at the drain traces in the toroid-loaded condition, I see a couple of things:
  - Drain voltage peaks at only 55-65V or so.
  - Drain ringing after turn-on, including temporarily shooting negative.
  
  I don't even know if either of those is even really a problem. I may investigate using a higher inductance feed coil L1 -- other folks describe 10uH, I'm only using 2.2k. Might also look into a schottky diode across the mosfet to clamp drain against going negative.
  

  ---

  Returning to the topic of board thermals:

  The key discovery was that heating on the inductor coil and the primary capacitor bank goes way
  

  down when the xenon is actually ionized toroidal. I believe this lends some credence to the idea of framing coil heating in terms of "input energy has to go somewhere". 
  
  I also hooked up a little 40mm axial...

  Read more »

• #3: i would like my circuit to not cook itself

  sky-guided • 02/16/2024 at 03:22 • 0 comments

  First Light accomplished. What did I learn?
  
  The circuit in its current iteration,
  1: gets hot alarmingly quickly
  2: is trying to draw more power than I can actually supply. 
  
  Those seem related, yeah. 
  

  apologies for the excessively american temperature units
  
  

  Based on simulating the circuit in the condition tested, at 15 volts supply I'd expect to see something like 50W power draw, at around 3.5A. The tests discussed in the previous post sure looked like the USB-based bench supply was badly voltage sagging, so the next day I hooked up the beefier bench supply -- and very quickly started sagging, then blew another mosfet. I suspect that this very-budget bench supply has some sort of destructively un-graceful switch from constant-voltage to the constant-current mode it failovers to when it hits 5A, but it's also entirely possible that the mosfet blew first and I saw the power supply feeding a shorted chip. 
  
  My power budget is <5A @20V (the 100-watt maximum of USB-C-PD), but I'd prefer to stay <60W if possible.
  
  

  Read more »

• #2: v0.2 -- First Light!

  sky-guided • 02/10/2024 at 02:41 • 0 comments

  I'll put the cool part right up top: it works.
  

  (and i blew up zero mosfets in the process!)
  
  Version 0.2 boards arrived lookin' nice and spiffy. 
  

  Everyone else I've seen who does a variant on this project uses a primary inductor made either of regular wire or copper tubing. I thought a PCB inductor would be more elegant. This inductor is two stacked turns  on 0.6mm PCB, 100mm center span and 8mm wide. I measured the inductance to be about 1.9µH.

  
  This was my first time using an actual solder stencil and hot-plate reflow rather than daubing on solder paste and using the hot air station. Turns out, using an actual stencil is way easier. Who'd have thought. Also turns out a scrap MTG card is a great paste spreader -- thanks to my gf for the improvised tool 🌈✨
  

  Soldered up beautifully.
  

  Since previous test coils had their 18ga wires badly overheat, I used bundles of 4x18ga to connect the board and test inductor. This is either a hacky kludge to get more surface cooling area, or a way of making bootleg litz wire -- take your pick.
  
  

  Read more »

View all 11 project logs