Settling time of Analog Switches

[Mike Edwards]

Hello,

I am designing a circuit that must be able to rapidly select between two
precision voltage sources, but I am having trouble reaching an acceptable
compromise between precision and settling time. My requirements are as
follows:

- able to select the output of one of two steady-state DACs with
12-bit resolution
- output must settle to 0.02% within 200 ns
- input voltages may swing +-2.5v
- when selecting between two identical input voltages, the output of
the switch must differ by less than 1 LSB (~1mV)

To meet the first and last requirement, I need a dual analog switch whose
on-resistance is closely matched between channels. This, coupled with the
bipolar requirement, makes it difficult to meet the settling time
requirement. As far as I can tell, settling time is not very well defined in
analog switch datasheets, and I have little understanding of the variables
that affect it. The switch I initially selected, the ADG623, seems to meet
all my requirements. AD claims that it will turn on in 75ns, but after
prototyping and testing the circuit, I find that "turning on" does not mean
"settling to 0.02%". I get a nearly immediate charge injection spike,
followed by about 350ns of settling.

Does anyone know of any faster switches that will meet my needs, things I
may be doing wrong that could cause this to be so slow, any alternate
solutions that could work, or any reasons why this just isn't going to work?

by the way, the drains of the switches are tied together and fed into an
OPA681 (2pC||100k) in non-inverting configuration. The DACs are buffered by
OPA2227s, which are fed to the sources of the switches.

Thanks in advance.

 

[Mike Edwards]

Tough one. For CMOS switches, there is a tradeoff between switching

speed/on resistance/charge injection; the parts with 1 pC-range
charge have high on resistances and switch slow. The charge injection
can screw up the input *or* output of an opamp, causing long recovery
tails. You can buffer the DACs with c-load-tolerant opamps with output
caps to ground, which will help the drive side at least.

Yes, that was the first problem I ran into; the drive-side op-amp would ring
for several microseconds every time the switch opened/closed. A few pF cap
across its feedback path fixed that.

On the output
side, jfet opamps are a lot more tolerant of spikes than bipolars.

The output-side op-amp seems fine. Actually, when I get rid of the op-amp
and replace it with a resistor and cap to ground, I still get approximately
the same settling effect.

Why not use one killer-fast DAC and switch/clock the digital inputs?
There are some wonderful 12-16 bit communications DACs around these
days.

I originally ruled this out because it would be difficult to fit a DAC
requiring 12-16 data lines into the system design. However, it is starting
to look more and more attractive. Do you know of any good DACs that settle
in under 200ns and achieve monotonicity with 12-bit precision?

Thanks for your help...
Mike.

 

[Jim Meyer]

by the way, the drains of the switches are tied together and fed into an
OPA681 (2pC||100k) in non-inverting configuration. The DACs are buffered by
OPA2227s, which are fed to the sources of the switches.

Thanks in advance.

Try an inverting configuration for the amp connected to the
switches. The switches work much better into a virtual zero Ohm load.

Jim

 

[John Larkin]

Yes, that was the first problem I ran into; the drive-side op-amp would ring
for several microseconds every time the switch opened/closed. A few pF cap
across its feedback path fixed that.

The output-side op-amp seems fine. Actually, when I get rid of the op-amp
and replace it with a resistor and cap to ground, I still get approximately
the same settling effect.


I originally ruled this out because it would be difficult to fit a DAC
requiring 12-16 data lines into the system design. However, it is starting
to look more and more attractive. Do you know of any good DACs that settle
in under 200ns and achieve monotonicity with 12-bit precision?

Thanks for your help...
Mike.

Hi, Mike

The LTC1668 is a 16-bit, 50 MS/s dac, rated to settle to 0.1% in 30 ns
typ. I mention that one because I've used it recently (in an arbitrary
waveform generator) and it seems to behave very well, with very low
glitching. I think Analog Devices has some *very* fast (like, hundreds
of MHz) 12 and 14-bit parts... take a look at their "TxDac" parts.

It's gonna be hard to *measure* this stuff. A regular scope would be
useless here. A Tek 7A13 differential comparator might be good enough.

What are you building?

John

 

[Fred Bloggs]

I am designing a circuit that must be able to rapidly select between two
precision voltage sources, but I am having trouble reaching an acceptable
compromise between precision and settling time. My requirements are as
follows:

- able to select the output of one of two steady-state DACs with
12-bit resolution
- output must settle to 0.02% within 200 ns
- input voltages may swing +-2.5v
- when selecting between two identical input voltages, the output of
the switch must differ by less than 1 LSB (~1mV)

To meet the first and last requirement, I need a dual analog switch whose
on-resistance is closely matched between channels. This, coupled with the
bipolar requirement, makes it difficult to meet the settling time
requirement. As far as I can tell, settling time is not very well defined in
analog switch datasheets, and I have little understanding of the variables
that affect it. The switch I initially selected, the ADG623, seems to meet
all my requirements. AD claims that it will turn on in 75ns, but after
prototyping and testing the circuit, I find that "turning on" does not mean
"settling to 0.02%". I get a nearly immediate charge injection spike,
followed by about 350ns of settling.

Does anyone know of any faster switches that will meet my needs, things I
may be doing wrong that could cause this to be so slow, any alternate
solutions that could work, or any reasons why this just isn't going to work?

by the way, the drains of the switches are tied together and fed into an
OPA681 (2pC||100k) in non-inverting configuration. The DACs are buffered by
OPA2227s, which are fed to the sources of the switches.

Something along the lines of a MAX4313 video muxamp should meet the
requirements, although your circuit parameters need to be rearranged a bit:
http://pdfserv.maxim-ic.com/en/ds/MAX4310-MAX4315.pdf

 

[Bill Sloman]

Tough one. For CMOS switches, there is a tradeoff between switching
speed/on resistance/charge injection; the parts with 1 pC-range
charge have high on resistances and switch slow. The charge injection
can screw up the input *or* output of an opamp, causing long recovery
tails. You can buffer the DACs with c-load-tolerant opamps with output
caps to ground, which will help the drive side at least. On the output
side, jfet opamps are a lot more tolerant of spikes than bipolars.

I'd be tempted to do this with discrete DMOS or GaAs Fets, but the
gate drivers would not be a pretty sight. There are some nice
integrated GaAs RF switches, but they typically won't swing this much.

<snipped sensible suggestion>

I'd also look into a diode-bridge switch. Even with carefully matched
diodes, the 1mV offset might be difficult to reach, but diode bridges
can have quite low on-resistances, quite low off-capacitances, and
very little feed-through.

The 1999 Farnell catalogue lists the HP(now Agilent) 5082-2080 as
batch matched from the 5082-2835 Schottky diode, and the 5082-2805 as
matched quad of the 5082-2800.

When I worked on a diode bridge, we were using the National
Semiconductor LM3039 six-diode array in a single package, which
typically offered 0.5mV matching between the diodes, 5mV worst case,
but Farnell doesn't seem to stock it any more, and it could well be
obsolete.

If you wanted to try the discrete transistor switch approach, we found
that the SD214 lateral DMOS FET was pretty good - the gate-to-drain
(reverse-transfer) capacitance is particularly low at around 0.5pF -
but as John Larkin says, the gate driver can be a bit of a pain.

http://www.linearsystems.com/datasheets/SD214.pdf

The SD214 is a four lead device, and you have work out what you want
to do with the sustrate connection - you can tie to the source, but
IIRR we tended to tie it to the negative rail, which cost us a bit of
gate drive, as the substrate then acts as a second gate.

 

[John Jardine]

Hello,

I am designing a circuit that must be able to rapidly select between two
precision voltage sources, but I am having trouble reaching an acceptable
compromise between precision and settling time. My requirements are as
follows:

- able to select the output of one of two steady-state DACs with
12-bit resolution
- output must settle to 0.02% within 200 ns
- input voltages may swing +-2.5v
- when selecting between two identical input voltages, the output of
the switch must differ by less than 1 LSB (~1mV)

To meet the first and last requirement, I need a dual analog switch whose
on-resistance is closely matched between channels. This, coupled with the
bipolar requirement, makes it difficult to meet the settling time
requirement. As far as I can tell, settling time is not very well defined in
analog switch datasheets, and I have little understanding of the variables
that affect it. The switch I initially selected, the ADG623, seems to meet
all my requirements. AD claims that it will turn on in 75ns, but after
prototyping and testing the circuit, I find that "turning on" does not mean
"settling to 0.02%". I get a nearly immediate charge injection spike,
followed by about 350ns of settling.

Does anyone know of any faster switches that will meet my needs, things I
may be doing wrong that could cause this to be so slow, any alternate
solutions that could work, or any reasons why this just isn't going to work?

by the way, the drains of the switches are tied together and fed into an
OPA681 (2pC||100k) in non-inverting configuration. The DACs are buffered by
OPA2227s, which are fed to the sources of the switches.

Thanks in advance.

Maybe level shift the voltages up by +2.5V so they will sit inside a 5.5V
rail and then feed the signals via a 4 bit 'bus switch' such as FST3125,
arranged as 2 pairs.
5nS, 5pF, 4ohms, could help a lot with the dynamics.
regards
john

 

[Vladimir Vassilevsky]

Hello,

I am designing a circuit that must be able to rapidly select between two
precision voltage sources, but I am having trouble reaching an acceptable
compromise between precision and settling time. My requirements are as
follows:

- able to select the output of one of two steady-state DACs with
12-bit resolution
- output must settle to 0.02% within 200 ns

The modern CMOS switches have the equvalent RC about 5e-9 sec. The RC
will settle to 0.02% if the delay is at least 10xRC, or 50nsec. So it
looks possible to meet 200ns.

- input voltages may swing +-2.5v
- when selecting between two identical input voltages, the output of
the switch must differ by less than 1 LSB (~1mV)

To meet the first and last requirement, I need a dual analog switch whose
on-resistance is closely matched between channels. This, coupled with the
bipolar requirement, makes it difficult to meet the settling time
requirement. As far as I can tell, settling time is not very well defined in
analog switch datasheets, and I have little understanding of the variables
that affect it. The switch I initially selected, the ADG623, seems to meet
all my requirements. AD claims that it will turn on in 75ns, but after
prototyping and testing the circuit, I find that "turning on" does not mean
"settling to 0.02%". I get a nearly immediate charge injection spike,
followed by about 350ns of settling.

To my knowledge, the best CMOS switches are built by MAXIM.
My advice is to use a multiplexer driven from the low impedance sources.
This way the spikes will be minimal.

Vladimir Vassilevsky

DSP and Mixed Signal Design Consultant

http://www.abvolt.com