I’ve just finished a little step-up DC/DC converter board based on the Ricoh R1283 that takes a single battery 3.7V input and outputs dual +12V and -12V rails:
https://climbers.net/sbc/dual-rail-12v-dcdc-converter/
I’d welcome any feedback or advice. Does the oscilloscope trace look correct? I had some challenges along the way, including hand soldering a “no-leads” DFN12 SMD package which required a custom KiCad footprint. And I had fun with my thermal camera 
For any switching converter (or for just about anything but the most trivial designs) I recommend a 4-layer board, with contiguous planes on the inner layers to provide image planes for HF return currents and low-impedance connections for decoupling and filtering caps, as well as for better thermal performance. The cost increment of 4 layers vs 2 is negligible in all but high-volume low-cost applications. Connect grounded caps to the ground plane with vias as close as possible to the cap pads; this will generally yield lower impedance than traces or outer layer copper flood.
I’d try stacking some 100 nF and 10nF caps on the 10 uF to lower that big spike a bit. The lower value caps are typically required for higher frequencies, although the ceramics you use are way better than any other option like tantalum or electrolytic. Try that, and then build another one with a ground plane and compare the difference. (Both electrically and thermally)
But the main thing is you got it working! (ps: C3 is upside down, but the cap you used is non-polarized so moot point)
Many thanks @JuliaTruchsess, that is very helpful 
I’ve not done a 4-layer PCB before, and some of those terms are unfamiliar (image planes) so lots to learn!
I’ve just watched the CE tutorial on using KiCad for 4-layer PCBs at:
https://www.youtube.com/watch?v=OBY1GuCmgPY
but can you recommend any tutorials about the benefits of 4-layer boards and how best to use them to reduce impedance?
Many thanks @PeeJay for the encouragement and the suggestions. I’ve just updated C3 on the schematic!
Now I’ve got something that works, the next project will be to see how much I can reduce the output noise. I think I’m going to do a new PCB layout that minimises the critical current paths, and I’ll add extra caps on the input and both outputs as you suggest.
JLCPCB charge the same for 5 pcs of 4-layer PCB as they do for 2-layer, so it might be interesting to order both and see what difference it makes to an otherwise identical circuit, electrically + thermally.
The noise on the scope could come from the probe pigtails. Check that probing is OK before adding extra capacitors
Thanks. I’ve now watched that Bob Pease video, but it went straight over my head. They obviously had fun, but it wasn’t aimed at beginners.
Was the takeaway message that you want the ground lead coming off the probe to be as short as possible?
Can you recommend any tutorials about probing basics? (with the proviso that my sub-£200 oscilloscope is by far the most expensive bit of kit I own, and I can’t afford to buy fancy probes).
Yes. Here’s another video, maybe more straightforward:
Thanks, that video is much easier to understand I’ve also found this great video from EEVblog which is exactly what I should have watched before I started!
The other thing I forgot to mention is to keep the input capacitor, inductor, diode and output capacitor in the switching circuit as physically close to each other as possible, and use big oversized tracks to connect them. See Pg 9 in the datasheet https://www.mouser.com/datasheet/2/792/r1283-ea-1770612.pdf
This is where the high instantaneous currents and other nasties originate from. Try and use diodes and inductors that are physically small as well. 1.3 MHz isn’t quite as lenient as the old 200 kHz stuff was!
Thanks @PeeJay. I’ve tried to position those components closer together in an updated PCB layout, and with much wider tracks (0.8mm). I hadn’t thought of using smaller inductors to reduce noise: is the trade off that a smaller inductor will be worse for heat?
The R1283K001 I’m using actually comes in 300kHz, 700kHz and 1.3MHz switching versions, so will it be easier to get a smooth output if I use the 300kHz version instead?
@sheffield_nikki Thank you for this thread. It is really timely because I want to do something quite similar, but to plug onto the end of a breadboard. I plan for USB / banana plug 5V input and +/-15V 100mA outputs that go to the upper and lower rails. I’m also looking to put BNC sockets on for injecting and monitoring signals.
My design is based on a LT3467, but now I am wondering if the Ricoh part is better as it has regulates both rails independently.
There’s some great tips here, that I will try to take heed of!
Smaller inductor has smaller volume, so Flux density is higher for same field intensity
A smaller inductor needs more turns for the same inductance, so you will not be able to push the same current into a smaller core
Additionally, losses rise faster with Flux density than core size, so more core loss
Smaller volume means higher thermal resistance, so higher hotspots
Yes, smaller inductor has less surface area, so less heat dissipation.
A smaller switching frequency tends to have more ripple, and the inductance of the coils need to be significantly larger. I would suggest just build a better version of your current design and see how much better it works. Then if you like build a low frequency one to compare the two. It will be a good learning exercise!
Do you have any use in mind for it?
So an update on the output of my circuit, after learning a bit about how to use my oscilloscope and better probing, etc:
This is what I was originally getting with the wrong oscilloscope settings, a USB-powered electronic load and standard 6" ground tail clipped to the circuit output header:
v2 is with oscilloscope 20MHz bandwidth filter ON, peak detect mode, load now from a simple fixed resistor, but same probe + 6" tail. Numbers are now PkPk with average, instead of Vamp:
and finally v3 is the same oscilloscope settings but now using a ground spring to directly probe across the output capacitor:
Does that look correct? What have I missed? Many thanks
Yes, looks good 
Excellent, you’ve learned about scope ground lead inductance 
Not sure why you would turn on bandwidth limiting on the scope, though - that only prevents you from seeing stuff that really is there.
I was following the advice of Dave Jones in his video (above, starting at +4:16 mins) about measuring power supply ripple & noise? I’m pretty sure he said that was standard?