density

Somebody was whining about PCB density.

I want to do a board with about 105 relays. I'd prefer 130 but that's not going to happen.

Here's a trial placement, just to see what might fit.

formatting link
We normally use the TPIC6595 shift register as our relay drivers. But it's big and I'd need 14 of them. So I think I'll add another Trion FPGA and have its pins drive a logic-level SOT23 mosfet under each relay. The T20 is only about $11, about half the cost of 14 of the TI things.

I do need to be sure that the relays won't interact magnetically.

The coils will generate heat too, brickwalled like that. I think we'll use 24 volt relays and run them at 16 volts quiescently, to cut the power dissipation about in half. We can bump the coil voltage up to 24 for a few milliseconds, whenever we change the pattern. [1]

My production people think they have ideas for placing 100+ relays efficiently.

I have about 10 square inches on the right for the rest of the stuff I'll need. And there's always the bottom of the board.

[1] sorta like the Austin-Healey overdrive solenoid.
Reply to
John Larkin
Loading thread data ...

Am 30.04.23 um 00:18 schrieb John Larkin:

What about bistable relays? That will be nice to the generated heat and will also reduce magnetic interaction when you switch only few at a time.

I have made a variable delay line from 3 ranks of 1:6 SMA coax relays and a pound of SemiRigid cable. Now that runs hot! I have provided an extra power switch for the relais so that they can be forced OFF when currently not needed, even when the phase noise analyzer is ON.

Gerhard

Reply to
Gerhard Hoffmann

Put the bistable relays in an 11 x 12 relay array and drive the row and column lines with tri-states out pits.This would activate a single relay at a time with 8 processor pins and two decoders.

If multiple relays need to be activated simultaneously, put them in the same column, this will require 12 row lines and 4 decoded column lines.

Of course the problem with bistatic relays is how to set the initial power on state.

Reply to
upsidedown

We have used latching relays to select thermocouples, because they have less thermal offsets. Relays have a lot of intermetallic junctions inside. We had a giant capacitor bank so that when power failed we has enough energy available to flop all the relays into their off state. I'd have to do that here too, with supercap at least. Non-latching relays are cheaper and easier to get.

One of the functions of this board is a cable tester, where 25 paths connect corresponding pins on the two D25's. I need each path to be closed when power is off.

You can probably play pullin/hold voltage trick, like I plan to do. Some relays and solenoids stay closed at 10% of operating voltage.

Thermal and magnetic interaction could be mitigated by relay placement and coil polarities. That would be an insanely complex puzzle. Fortunately, many of these relays specify a coil polarity. But I could substitute rotation to maintain the horror.

Reply to
John Larkin

Does that work, driving latching relays in a matrix? There would be sneak paths to unintended relays.

If that worked, one in principle could matrix-PWM non-latching relays.

One tiny mosfet and one FPGA pin per relay is sure easy. I drew a second FPGA as the relay driver but I probably have enough free pins on the usual FPGA.

The stuff on the left side of the board is our "template" for all the plugin boards in this family. It looks like that FPGA has 154 free GPIOs, and I'll only need about 125 or so.

That too.

Reply to
John Larkin

Sure, add diodes and do the relay version of charlieplexing LEDs.

Cheers

Phil Hobbs

Reply to
Phil Hobbs

Wouldn't that take two diodes (one dual) per relay?

Given non-latching relays, the drive duty cycle would need to go down as I allow multiple relays to be on simultaneously. And the voltage would correspondingly go up!

If I have 100 or so FPGA pins free, it will take just one avalanche-rated logic-level SOT23 n-fet per relay.

Reply to
John Larkin

søndag den 30. april 2023 kl. 20.46.16 UTC+2 skrev John Larkin:

or ULN2003 et. al.

Reply to
Lasse Langwadt Christensen

On 2023-04-30 14:46, John Larkin wrote:> On Sun, 30 Apr 2023 13:13:18

-0400, Phil Hobbs > snipped-for-privacy@electrooptical.net wrote: >

Charlieplexing relies on two features of LEDs: they only conduct in one direction, and they have this huge forward voltage, so if you have two in series driven from a lowish voltage, nothing happens. The latter is also true of relays if they start out inactive, but not necessarily if they're already active.

It might be fun to think about, but I doubt there's a genuinely useful application.

Cheers

Phil Hobbs

Reply to
Phil Hobbs

søndag den 30. april 2023 kl. 21.02.50 UTC+2 skrev Phil Hobbs:

but with enough voltage to make pwm work, row/column multiplexing could work

Reply to
Lasse Langwadt Christensen

Relay data sheets are universally awful. Among other things, they seldom provide the numbers you need to exploit the pullin/dropout differential.

Full voltage to energize and 0.7 of that, half coil power, to hold usually works, but isn't generally guaranteed.

The efinix FPGA costs about 5 cents per gpio and takes very little board area.

Maybe someone could start a business defining some useful functions and selling programmed serial flash chips to program an FPGA to do them. Or sell the functions in FPGAs that have internal flash.

Reply to
John Larkin

We stock them too, in SO16. They would probably go under the relay matrix, on the back side of the board. I was intending to use the TI shift registers as the relay drivers, but they won't squeeze between the pins of the relays.

The ULN2003 *does* fit between the relay pins on the bottom of the board!

Reply to
John Larkin

Don't some relays come with a builtin quench diode on the coil (in parallel)? Those have matching features to the LED case, if you look at the polarity-dependence of coil current.

Reply to
whit3rd

I haven't seen any. I do want a multi-sourced relay, in a standard

10x20 mm case.

I think some relays have a magnet inside, which makes the coil polarity matter. Latching relays seem to care.

The ULN2003, suggested by Lasse, has 7 darlington drivers with catch diodes. That's pretty good. I'd need about 16 of them.

I'll also use a dozen or so SSRs, probably the Ixys CPC1540. I can drive them directly from my FPGA, with maybe a series resistor.

Reply to
John Larkin

Not competitive against ~6 cent SIPO shift registers like SN74HCS595DYYR. You might have to push a couple of the MOSFETs to the side slightly, but at 3.3x4.3mm footprint they should fit under your relays. (other side of board)

The 5V capable outputs will also make more MOSFETs usable - possibly a further saving.

pre-programmed chips are a available from several vendors.

Reply to
Jasen Betts

Why are those "good'? Those do 100 mA low-side drive per channel, and have 1.2V loss; a biased transistor will drive a 24V relay, under 10 mA, with 0.3V loss.

Reply to
whit3rd

Because they are cheap and we stock them already.

They are in our PADS library.

Because they fit within the relay pins, on the back side of the board

Because they have 7 channels with catch diodes

Because they are multi-sourced and available. Digikey has over 100K available. Mouser has 400K!

More like 0.75 driving my 12 volt relays. Not an issue. The 12 volt power supply will be a programmable switcher anyhow.

If I use 12 volt (automotive) relays, I'd need about 33 mA. Must-operate voltage is 9.

What did you have in mind for a "biased transistor"?

Reply to
John Larkin

The little resistors-built-in things, like DTC114E

formatting link

should also work (up to 100 mA) for a 12V relay, but will get warmer for 12V than for 24V relays.

Reply to
whit3rd

onsdag den 3. maj 2023 kl. 11.14.13 UTC+2 skrev whit3rd:

it is less parts, loss doesn't really matter, a biased transistor will need a clamping diode

Reply to
Lasse Langwadt Christensen

Oh, a bias transistor. I still need to worry about the flyback kick.

Reply to
John Larkin

ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here. All logos and trade names are the property of their respective owners.