Sensing 300A on a SMD PCB

Well, I'll be damned. That actually does end up giving you a more-or- less accurate reading of the inductor current.

However, it'll be that only to the extent that the inductance stays constant -- if the inductance changes much it'll throw the "current" reading all out of whack.

--

Tim Wescott 
Wescott Design Services 
http://www.wescottdesign.com

Yep, that gets my vote. If there's a good, reliable way to connect a 300A cable to a sheet of copper glued onto fiberglass, I would appreciate any alternative to using it, twice.

Wurth 7461117, M10 screw, 90deg press fit, good for 340A.

7461100 is a M10 stud for 340A.

These are multi-pin press fit, basically 10A per pin.

< >

--
Chisolm 
Republic of Texas

Those would work with PC boards that have super-heavy copper, or many conducting layers. With reasonable PC boards, current density will get extreme around those things.

Current tends to come from one direction, so most of the copper that funnels into a thing like that isn't working. The inner pins are basically shielded by the outer pins, so they won't see much current either, unless you play some multi-layer tricks. Those gadgets can stand 10 amps per pin, but the PC board maybe not.

Press fit is scary here.

--

John Larkin         Highland Technology, Inc 
picosecond timing   precision measurement  

jlarkin att highlandtechnology dott com 
http://www.highlandtechnology.com

Ouch... good catch. George H.

The closed-loop type also have a flat response, if that's important to you.

On Wednesday, 10 February 2016 04:03:05 UTC+11, snipped-for-privacy@yahoo.com wrote :

The current sensors on the London Underground trains were pairs of C-cores wrapped around the bus bars.

The guy who designed the system - Colin Hunter - my boss from 1973 to 1976

- used the C-core as the core of a Royer inverter, that drove the core into saturation in alternating directions. When the current in the inverter win ding added to the current in the bus bar, the C-core saturated at a lower c urrent than when it opposed. Monitoring the DC content of the current runni ng into the inverter was a remarkably precise measure of the current in the bus-bar, divided by the turns ratio.

C-cores big enough to fit around a bus-bar are big, and can carry a lot of turns. IIRR (and this was told to me verbally, not revealed on paper, some forty years ago) 300A in the bus-bar equated to 100mA asymmetry in the driv e currents.

Spehro's magnetic sensors could be smaller and neater, but would be less fu ndamentally accurate if you didn't build them into a null-sensing system.

With the C-cores you could unclamp them and flip them around to check for a ny local magnetic fields.

--
Bill Sloman, Sydney

DIN-41612 mixed connectors are nice. Element-14 (Newark in the US) stocks this

You get eight high current inserts per connector. I can't find the high current inserts, but the Harting web-site suggests that each one is good to 40A (which seems high).

--
Bill Sloman, Sydney

I remember seeing a Clark board for an IBM 3090 mainframe, circa 1990. These were boards about 1 cm thick, which connected the TCMs (thermal conduction modules, ycliu).

I was working in the ManufacturinResearch department at the time, and they were having some issues with 'mouse bites' in the PCB traces causing reliability problems. Power was distributed via large copper 'angle iron'--iirc it was +3V and some smaller negative voltage, all at over 3000A. Dim memory makes me want to say it peaked at 8000A.

Cheers

Phil Hobbs

Ooops, yes.

Thanks. On the run here...

Cheers, James Arthur

Flux gate. Those are slick.

Now if they just came in 2512 .

(Really though, thanks.)

Cheers, James Arthur

A frain bart--you got me. Correction appreciated.

Grins, James Arthur

LTC mentions they have to be calibrated for the inductor, and LTC offers dR/dT compensation using a thermistor to gauge the L's temp.

Slightly hokey, but fast, efficient, and cheap.

Cheers, James Arthur

Even wave soldering?

--

John Devereux

Let me re-phrase that to "you may have trouble soldering". Granted we had a limited set of data points. The design was a rather long/narrow board. I dont remember the cu thickness but I think it was 6 power layers, 3 and 3. We went press-fit. I had used press fit in the past for some high speed / high density backplane connectors.

--
Chisolm 
Republic of Texas

Something like 96 layer, with multiple layers of twisted pair overflow and EC wiring on both sides.

Depending on the particular machine the voltages varied but I think you're talking about -3V. The "Wiccopee" levels were 0V and -.9V and "Wiccopee Long Line", which was a VBE (emitter follower) down from that.

A max of 8000A sounds about right. A CPU was 15 TCMs, each something between 1200 and 1500W (maybe pushing that some).

I was in the "Machine Technology" group at the time. We did most of the circuit designs (particularly any level translators), wiring rules, and test systems. The cooling guys were right down the hall from me.

What made these modules so power hungry? How many gates might there be in such a module? Clock rate?

I've never had the chance to peek inside or tear down a mainframe yet.

About 5K ECL gates per chip, 121 chips (including terminators) per module, IIRC. I don't remember the clock rate of the ES5000 series (if I ever knew) but the 308x, the first to use the technology, was

40MHz in production in ~1980. the 308x wasn't as bad, though (1kW/chip, 100 chips per module). It really was amazing use of brute force to get the performance needed.

Fascinating- especially the brute force part. Is it correct that the power supplies could have been really "stupid" as no special regulation would have been needed as the load would always be the same no matter what the processors/logic were doing?

Any idea where they assembled the MCMs? Where the chips from IBM's foundry in NY or VT?

It's really quite amazing that at one point IBM basically built everything in house.