Driving large bank of single-coil latching relays

[Zbig]

Hi

I'm looking for the most effective (economically, mostly) way of independently driving a group of 24 single-coil 5V latching relays with a MCU. Here are the options I'm evaluating at the moment:

• The trick described here: http://www.maximintegrated.com/app-notes/index.mvp/id/3288
  Pros:
  
  • Maxim themselves say it should work so who am I to argue
  • Simple, doesn't involve any components other than 6 MAX4820's besides 48 pieces of 50 Ohm resistor
  Cons:
  
  • MAX4820/4821 are not exactly the cheapest things around and I'd need 6 of those
  • Considerable amount of power wasted due to the symmetrical resistor layout. But then again, it only needs to be energized for like 2 ms or so
• Wiring each of the 24 relays' coils parallel into a common H-bridge output and putting 24 full wave rectifier bridges in series with each of the  individual coils. Then sinking each of the now-rectified coil current via MAX4820 across rectifier bridge's +,- outputs.
  Pros:
  
  • Number of MAX4820's needed reduced by half (down to three)
  Cons:
  
  • Now I need an H-bridge and 24 rectifier bridges
  • Sum of the voltage drops across the H-bridge and rectifier bridge now calls for artificially increased input voltage in order for the relay's "Must operate VDC" to be met.
• Variation of the above, but using bilateral electronic switches in series with the coils, driven by simple latch instead of the bridge rectifiers "shorted" with MAX4820.

Is it worth to even evaluate any other option than the first one? Are 2 and 3 viable at all or am I talking nonsense here? Naturally, any other ideas not listed here are highly welcome. Of course 24 individual H-bridges would be ideal but I haven't found any combo IC with more than two of them inside and the price of 12 pieces is too high.

[c4757p]

How much current?

[Zbig]

You mean the relay driving current or the current across the relay's working contacts? The relays are the subminiature ones with 180 Ohm / 5V coil and the current they'll be switching is miniscule (0.5W of power switching tops, I guess). The thing is, the device I intend to build is the digitally-controlled resistance decade box so I have to have a full galvanic separation (I guess?) between my device and the DUT. I thought about using reed relays but I'd very much like it to retain its resistance setting while unpowered.

[c4757p]

I'd use the first solution, but use TPIC6C595 instead. Much cheaper. You'll need diodes, though, they're not built in. But diodes are cheap.

Who cares about the wasted current? It's a little pulse once in a great while.

You can save on wasted current a bit by using a 10V rail and dropping 5V in a 180 ohm resistor. That way the current is the same (27.8 mA) and the power wasted is only 138 mW in one resistor, 556 mW in the other. If you use that trick exactly, if for example the relay needs at least 4.5V to switch and you use a 5V rail, you'll need to use 18 ohm resistors, not 50, and the low side will draw 277 mA and burn 1.4 W. The exact "sweet spot" (not necessarily 10V) will depend on the relay's requirements, of course.

Edited to correct derpy math.

[Zbig]

Wow, TPIC6C595 is almost ten times cheaper than MAX4820. Thank you very much c4757p, TPIC6C595 it is, then! And thanks for the tip on the resistor values and the rail voltage, I'll do some calculations and measurements to get those optimized.

Thanks again and best regards,
Zbig

[c4757p]

Maxim parts are stupidly overpriced and more ordinary than they let on, you can almost always find something just as good for less than half the price. Number one rule of buying Maxim parts: don't.

[fcb]

Latching relays.  There are several types and they both need driving differently.

S/R Coil - have two coils, set and reset.  Easy to drive, tend to be the bigger relay type.

Bidirectional Coil - this type requires you momentarily apply current one way to set, and another way to reset. This sounds like the kind of parts your using.

The way I do it (I use these on a product), is with a ULN2803A driven by a 4094, and some resistors.  Basically build an H-Bridge with resistors in the top posistions, and two ULN channels in the lower positions.  It's pretty cost-effective, although you'll need to experiment with the right value resistors.  I also put a 470R and 220uF in the supply rail to this driver to avoid burning out the resistors in-case of fault, it only takes 0.2S to recharge, so my software takes this into account.  Relays: Axicom FX2 D3243.

[Zbig]

Thank you fcb, I thought about using Darlington pairs as well but I'm afraid the necessary passives needed for my setup would drive my parts count through the roof (I need 24 relays) and I want it to be as compact as possible. But I'll look on that option carefully once again. Yes, the relays I'm going to use are single coil (bipolar) ones (I've mentioned that in the topic  ).

Too bad TPIC6C595 suggested by c4757p requires 5V power and logic as I'll be using 3.3V everywhere else within the digital domain but throwing in one level shifter and a 5V LDO regulator is not an end of the world, I guess.

[ptricks]

MC33999 would be my choice. About $10 to drive all 24relays, no other parts required,each chip controls 16 relays , at a cost of about $5 a chip.
http://www.freescale.com/webapp/sps/site/prod_summary.jsp?code=MC33999

[Zbig]

That's 16 low-side outputs, ptricks, so for "bidirectional" driving of 24 single-coil latching relays I'd need 3 of those but it certainly deserves a closer look, thank you!

[fcb]

Thank you fcb, I thought about using Darlington pairs as well but I'm afraid the necessary passives needed for my setup would drive my parts count through the roof (I need 24 relays) and I want it to be as compact as possible. But I'll look on that option carefully once again. Yes, the relays I'm going to use are single coil (bipolar) ones (I've mentioned that in the topic  ).

24 relays = 48x resistors + 1x resistor + 1x electrolytic + 3x ULN2803A's + 3x 4094B/74HC595 = $2

Too bad TPIC6C595 suggested by c4757p requires 5V power and logic as I'll be using 3.3V everywhere else within the digital domain but throwing in one level shifter and a 5V LDO regulator is not an end of the world, I guess.

Unlikely you'll need a level shifter. Beware the TPC6C595 doesn't have back-EMF diodes.

EDIT: Just looked at your original post, the way I've suggested seems to be the same as the MAXIM datasheet (I beat them by 11 years...) with the exception that I supply the relays common rail from an RC to cope with faults and lower the 'thwack'.

[ptricks]

Is there any chance you can use pulse relays ? Then you only need to power on and off without the reversing stuff.

[jeroen74]

I think you could reduce the number of outputs to half+1 if you don't have the requirement to set all relays to their new position at once. Connect one side of each relay to a common driver, and the other side of the coil of each relay to its own driver; only 25 drivers instead of 48. The common one could be a pair of transistors with separate control lines for each transistor so you can tristate it.

[fcb]

I'm not sure this will save many parts (I really haven't thought this one through yet).

Could you build a 5x5 matrix of relay coils, then have 10 (simple) half-bridges driving the Rows and Columns. You'd probably have to get the voltage or timing right (as there will be other paths through the matrix to drive the relays).

[jeroen74]

It will save a lot on PCB traces, instead of 48 you only have 25. If you go with a 3x8 matrix, it's only 11 traces. Could well make the difference between a trivially to route single layer smaller PCB and bigger two layer with loads of via's

[Zbig]

I think it's a brilliant idea to matrix these relays, actually. Thanks, guys!

Thank you everyone for your great input so far, now I have several options to evaluate.