mounting a pot core

On 2026-01-21 11:45, john larkin wrote:>

Yeah, putting the copper under much tension isn't a growth strategy. Stress due to bending is quadratic in the curvature, so a little bit is no problem, but how much is OK, I don't know.

IBM's famed Thermal Conduction Modules (TCMs) originally used sintered alumina/refractory metal bricks (dim memory says niobium) holding either

They were brutally reliable--not a single field failure in 30 years--but had a dielectric constant of almost 10, so they were slow.

The packaging folks at Yorktown and East Fishkill came up with a faster technology based on copper metal and glass-ceramic bricks with an epsilon of 5. That sped things up by 40%, which was a win as far as it went. Trouble was, they were hopelessly unreliable.

In the original model, the metal had a lower CTE than the ceramic, so that once the brick cooled from red heat, the metal was always in compression, which prevented cracks from propagating. However, copper has a higher CTE than glass ceramic, so that the copper was always in tension. That's for a good time, not a long time.

A pal of mine figured out a fix that my caveman side found very appealing: they dunked the hot substrates in oil, and the resulting thermal shock cracked all the copper loose from the glass, relieving the stress. That got them back to nearly perfect reliability.

(Later they went to copper/polyimide, I think.)

Nowadays people just cut holes in the hood. ;)

Cheers

Phil Hobbs

On 2026-01-21 11:45, john larkin wrote:>

Yeah, putting the copper under much tension isn't a growth strategy. Stress due to bending is quadratic in the curvature, so a little bit is no problem, but how much is OK, I don't know.

IBM's famed Thermal Conduction Modules (TCMs) originally used sintered alumina/refractory metal bricks (dim memory says niobium) holding either

They were brutally reliable--not a single field failure in 30 years--but had a dielectric constant of almost 10, so they were slow.

The packaging folks at Yorktown and East Fishkill came up with a faster technology based on copper metal and glass-ceramic bricks with an epsilon of 5. That sped things up by 40%, which was a win as far as it went. Trouble was, they were hopelessly unreliable.

In the original model, the metal had a lower CTE than the ceramic, so that once the brick cooled from red heat, the metal was always in compression, which prevented cracks from propagating. However, copper has a higher CTE than glass ceramic, so that the copper was always in tension. That's for a good time, not a long time.

A pal of mine figured out a fix that my caveman side found very appealing: they dunked the hot substrates in oil, and the resulting thermal shock cracked all the copper loose from the glass, relieving the stress. That got them back to nearly perfect reliability.

(Later they went to copper/polyimide, I think.)

Nowadays people just cut holes in the hood. ;)

Cheers

Phil Hobbs

Pot cores aren't the only shaped bits of ferrite you can buy. RM cores are another version of the same idea, and the cores that are designed to be used with printed circuit windings do seem to be flatter and wider.

The Siemens/TKD ferrite core catalog might be worth reading. Here's a small chunk.

The first core pair on that list - PQ 107/87 - is 87 mmm high when assembled. It seems to be the biggest in that bunch.

Putting the ferrite core under any kind of mechanical stress is something they advise against.

Any PCB that mounts on multiple spacers likely has some flex stress. A little is probably not hazardous.

Our PCB fab notes specify flatness 0.075 mm per cm for some reason. That's close to 1% twist I think.

We could cut a hole in the top cover. That would be fun.

John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics

My intern tried several experiments of making the windings from kapton flex, but they weren't as fast as using coax.

A regular pot core with connectorized micro-coax wound on a bobbin is nice, and easy to make in production. No soldering required.

We have samples like that. Seems to me that they would store less energy than a full pot core.

John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics

Aren't you worried about the capacitive coupling between the windings? A piece of coax wound on a ferrite core is basically a balun. You

Jeroen Belleman

I like to think of it as a transmission line transformer, but it's the same thing.

When people say "balun" they don't always mean balanced-to-unbalanced.

Coax has continuous capacitance from shield (primary) to center (secondary). And shield-shield on the neaby windings.

We plan to buy a length of micro-coax with an SMB connector on each end. We'll wind that on the bobbin and plug it into pcb SMB connectors.

Like this, but higher voltage.

John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics

The coax wound on the pot core is unbalanced to unbalanced, but in an odd way -- shield and center are decoupled from one another, and so one can reverse role of center and shield.

The capacitance between the outsides of the shields is irrelevant so long as the shields are thick enough and the frequency is high enough that the inside currents are independent of the outside surface currents.; this takes a few megahertz to accomplish in solid copper shields.

Joe

Our goal is to make high voltage pulses with few-ns rise and fall times, hundreds of MHz equivalent.

John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics

The speed of electromagnet propagation is lower in Kapton that it is dielectrics like teflon and polyethylene, which have lower dielectric constants, but your Kapton flex probably wasn't designed to be a planar transmission line.

What matters to you is the total volume of ferrite, rather than its precise distribution. If you could find a wider, flatter ferrite core you should be able to store more energy in the same vertical headroom.

Semi-rigid coax does. Regular coax - with a braid outer conductor - has around 95% shieding.

Whoever mounted those pot core pair without lining up the gaps threw away some of the energy storage you claim to be interested in.

The point is that the ends of the coax are separated from each other (for RF) by the common mode inductance of the winding. The capacitance between the windings shunts this inductance, so that's to be avoided.

Its not always an issue, but it's something to be aware of.

Jeroen Belleman

I differ. The purpose of winding a coax on a magnetic core is to insulate the ends from each other by the common mode inductance of the winding, This lends you the freedom to choose different RF GND points at both ends.

Even though we call these things baluns, it does not necessarily transform only from balanced to unbalanced. It can as easily altogether invert a signal, just by choosing the right RF GND point. Interwinding capacitance shunts the common mode inductance of the winding, deteriorating the insulation between the ends.

Jeroen Belleman

Toss the double pcb layers.

RL

Mount the transformer on the box wallwith on a layer of vibration absorber.

Connect your flying leads as required.

Energy storage ~ minimum xsectional area - not mass. . . . so align your slots.

RL

True. The product works fine in production, but the misalignment might matter a bit in my new GaN pulser. We'll experiment and see how much.

That pic is from the isolated HV option of a digital delay generator:

John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics

I don't think we are saying contradictory things. But the viewpoint and wording do differ.

Joe

We like to put parts on PC boards. But why vibration isolation?

We will use a bigger pot core on the new product. I measured a 2% change in inductance as the core halves were rotated, so slot alignment is no big deal.

John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics

We made several and we sure intended them to be txline windings. We don't understand why the planar windings were slow, but we need to move on with life, and my intern has gone back to Cal Poly, so we'll go with connectorized coax. That is known to work.

We are tight on PCB area, and we need to move on with the project. We can experiment with different ferrite materials once the boards are built. There's lots available in this pot core size.

John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics