Back to back FETs

[MacGyverS2000]

Can anyone think of issues I might encounter by putting MOSFETs back to back (D to S, and S to D, G and G tied together), like an AC switch? When the gate is enabled, I want a signal to pass through this "switch", regardless of which end has the higher potential. I didn't think a single MOSFET would accomplish the goal due to the varying potentials.

As always, replacement circuit suggestions welcome (I'm using SOT-23 packages, so any replacement must be compact).

 

[VEBill]

Your description reminded me of those 4066 and similar 'CMOS Switches', but when I checked apparently not quite the same. They are using P- and N-Channel switches in parallel to perform the same function.

See:

http://www.maxim-ic.com/appnotes.cfm/appnote_number/638

 

[MacGyverS2000]

Hmmmm, devices like the MAX4626 look useful, but pricey. Two MOSFETs would be about 1/10th the price, even though they would take twice the board real estate. Ron is surprisingly high on those buggers, must be the lack of silicon real estate for the FETS due to it being taken up by the control circuitry.

Good reading. Never thought about Ron being in parallel, so that can only help things. At $0.90 a pop, though, I may stick with the dual NFETs back to back... unless someone can give me a reason it won't work.

 

[VEBill]

I seem to recall that there isn't very much significant difference between the S and D ends of a MOSFET. It is just a bulk semiconductor between those pins.

Are you sure that you'll need two?

 

[IRstuff]

Actually, there is a difference. A MOSFET is actually a 4-terminal device, with the substrate being the 4th terminal. The substrate is usually connected to the source terminal on a 3-terminal package.

Which, will cause a problem if the source is raised higher than the drain during operation. The substrate diode to the drain will be forward biased in that condition. That's why the conventional MOSFET switches leave the substrate connection tied to the lowest circuit potential to ensure that the substrate is never forward biased. Obviously, the p-channel substrate will need to be tied to the highest circuit potential.

This is why the Ron on MOSFET switches are so bad. If you recall the MOSFET body effect, it's directly impacted by the reverse bias of the substrate relative to the channel, and acts to reduce the effective gain of the transistor.

TTFN

 

[VEBill]

Thanks for that correction.

I guess that those substrate diodes are going cause problems for MacGyverS2000's planned back-to-back MOSFETs ?

 

[sreid]

Take a look at the Maxim/Dallas Semiconductor analog switches. Pretty cheap and easy to use.

 

[IRstuff]

I think there are 4-terminal varieties of MOSFETS, although I suspect they're probably relatively low performance ones. Those would work if the substrates are connected correctly.

But, you find that the Ron performance will be degraded.

TTFN

 

[sreid]

A N Channel J-FET can be used. A resistor connected gate to source gives zero volts and the device is on. Pull the gate to, say, -15 V with an open collector driver and the device is off for a wide range of drain and source voltages.

 

[MacGyverS2000]

Kinda forgot about the whole reverse bias deal... it sounded good while it lasted though,

I suppose I could waste an extra pin and control a P and an N separately, but it defeats my purpose of saving board space AND a pin. Not sure yet what I'd do for the negative voltage to control the PFET. This is to selectively allow 2-way communication over a 0-5V bus line, so finding a bi-way switch that would also reamplify the signal back to 0-5V as it degraded along the line would be ideal.

 

[nbucska]

What is the data of your signal and of the input circuit ?

<nbucska@pcperipherals DOT com> subj: eng-tips

 

[MacGyverS2000]

Short and sweet of it: a 0-5V binary data bus, two-way communication between micro #1 and #2. A third micro somewhere along the line should be able to interrupt the communications by shutting off the line. So, the switch should be two-way in nature, and ideally should have a single control line.

At the moment, I don't believe I'll need directional-specific control (disable one way, but not the other), but if it's possible with little to no extra, I may find a use for it.

 

[felixc]

What about the bus-switches? Aren't they doing what you're looking for?

http://www.fairchildsemi.com/products/interface/switch.html

 

[MacGyverS2000]

felix, Now THOSE are some worthy components to check out. Looks like the NC7SZ384 is going to be a potential winner. I also have a new keyword to look for in the hopes I'm lucky enough to find something even better and/or cheaper.

At first I was only slightly worried about the on resistance of 5 ohms, but for longer runs I'll be using a driver periodically along the line anyway. Package size is where I want it and price is within reason at $0.145 (SOT-23) and $0.119 (SC70, never heard of this package, seems Fairchild is still waiting for a JEDEC registration number on it, but the dimensions on a 6 pin LEADLESS are 1 x 1.45 mm!!! My eyesight isn't that good.).

I Never knew this type of product existed, which seems to be the growing trend these days (at least for me). I only know what my problem is and a vague idea how to solve it, but ignorant of any prepackaged solution... there's only so many product announcements one person can view. It has become painfully clear the last several months, however, that I need to renew my subscription to EDN, EE Times, etc. I let the subscriptions pass years ago as I just didn't have the time to read them and still have a life outside of work. I guess I need to reevaluate my time management again, as I've spent days looking for a solution I should have known existed.

felix, you deserve a star for that one

 

[felixc]

MacGyver, from the questions and answers I was not able to understand what you were trying to achieve, and the stuff was getting pretty complex until nbucska asked the basic question again.
Other manufacturers are also making similar switches. Try to get a second-sourced part, as they haven't reached the same perennial status as logic gates.
The bus-switches are getting very popular as they do not contribute to delays, other than series resistance, and give an no-hassle bidirectionality.
With Internet, until something very special really interests me, I don't read the magazines anymore. However I subscribe to the email news from the big manufacturers like National, Fairchild, and TI, and also from the Woodstock Wire. There's more than enough for me in all these, and no paper to waste.

 

[nbucska]

I don't understand what you are trying to do. If this is
a communication line between two micros, you can just command them to neglect it ( it costs one pin ) unless
the communication is too fast.

What are the types of the micros? What is the comm. speed?
Why do you have to disconnect ?

<nbucska@pcperipherals DOT com> subj: eng-tips

 

[MacGyverS2000]

I'm preventing micros from further down the line from receiving messages at all. Without going into unecessary detail, processors further down the line should not be able to listen in on the communications conversation until processors further up the line have OK'd it.

This has stemmed from one of my usual hair-brained schemes, but it is actually a quite elegant solution (I think) to a nagging problem. Hardware is minimal, comm protocols are reduced in size, which also reduces software and bandwidth.

 

[MacGyverS2000]

I'm bringing this one back to the top as I'm looking for a related component. I would like to periodically "rejuvenate" the comm line strength with a driver. However, I'm not sure what keywords I should be looking for, or even if this item exists.

Essentially, it would be a two-way repeater. Two-way communication lines 'A' and 'B' are connected together and sending messages back and forth to multiple processors along the line. My protocol won't allow both buses to have the line at the same time, so no need to worry about collisions. All I need is a driver that keeps the current level up, resharpens the waveform edge, and a really nice thing would be an output enable pin.

Should I be looking for a single chip solution, or can anyone suggest a minimal parts count discrete component solution? The bus-switches mentioned before are great (and are being designed into the project), but they don't add any drive to the line, they just connect them together.

 

[MacGyverS2000]

I like using things they weren't really intended for. In this case, a voltage interface transceiver may be useful. A piece such as Fairchild's 74LVXC4245 (

http://fairchildsemi.com/ds/74/74LVXC4245.pdf

) translates the differing voltages on two different buses (for example, one's at 3.3V while the other is at 5.0V). The idea would work by simply saying both bus voltages are the same, but at 8 bits, it's way too large... I need something similar, but with only 1 bit (like in an SOT-23-5 package).

The search continues...

 

[felixc]

Are your bus signals at the same voltage? Which one? Have you looked at single-gate packages, usually SOT23. You could use two xx125 or 126-type of tri-state gate, connected back to back. (would be an equivalent to a lot of fets) I'm not sure if they're available in 5V technology though.