Is it possible to parallel two op amps to get twice the output current?
We get this question periodically on our E2E forums. Though we may answer with a qualified “yes,” it tends to make us shudder just a bit. It can be done… but with great care. So let me come quickly to a key point. Don’t use the simple circuit on the left. Directly paralleling inputs and output of two op amps is sure to start a serious argument between the two. Differing offset voltages will cause them to fight over the correct output voltage. They may burn all their output current capability in the battle with one output current (sourcing) flowing into the other (sinking current).
Figure 1b has a chance. Op amp A1 is the “master” and A2 is the so-called “slave,” replicating the output voltage of the master. R3 and R4 promote reasonably equal sharing of the load current, even though A2’s output may be slightly different. Feedback is connected on the load-side of R3 and R4 so their voltage drop is corrected. You lose some output voltage swing capability in the I∙R drop on these resistors so you will be tempted to make them low in value. But the offset voltage of A2 will cause extra quiescent current equal to Vos/(R3+R4). It’s a tricky tradeoff.
Be very cautious with high speed signals. You want A2 to accurately replicate the output of A1. If the signal moves too fast, the phase shift of A2 will cause differing output voltages and wasted current. It’s important to avoid slewing. If necessary, add an R-C filter at the input so the fastest rate of change on the output of A1 is well below slewing speeds. The dynamic behavior of two amplifiers may not match well during slewing.
Don’t use older generation op amps that have output inversion (phase reversal) behaviors. If A1 can overdrive the input common-mode range of A2 and its output inverts the result is ugly.
Above all, check the behavior of your circuit thoroughly. SPICE may tell you whether you have a basic working circuit, but op amp macro-models may not accurately predict the quirks that could befall this circuit. Build a breadboard and check all signals and conditions carefully. If your op amp is multi-sourced, consider that not all manufacturers’ devices behave exactly the same. (But, of course, you have only one source for op amps, right?)
Do you think I’m a bit leery of paralleling op amps? Well, yes… call me leery. It can be successful but proceed with caution. I recommend that you consider an easier path—using an op amp with more output current. Here are a few possibilities:
Have you successfully paralleled op amps? Or do you have scars from trying? Comments welcome.
Thanks for reading,
Bruce email: email@example.com
Table of Contents… with 50+ other interesting technical topics.
This reminds me of a discussion about paralleling two batteries for double capacitance: If they don't have exactly the same output voltage and discharge curve, it will end up with wasting quite some share of the total capacity between the two.
And putting two (rechargeable) batteries of not exactly the same capacity in a row for double voltage may result in desctruction of the one with smaller capacitance - The main reason why the advertised '1000 recharge cycles' shrink to a few dozends in real-world applications.
Even if two OpAmps are of the same type, or even on the same die, they won't produce exactly the same result. Should be obvious (and so should be the implications) but is often forgotten by todays 'digital' thinking.
Thanks for bringing this to our attention.
In the Oct 2012 issue of Elektor magazine, pg 14, Doug Self parallels sixteen 5532 op amps, using two parallel sets in bridge-mode per channel in his "5532 OpAmplifier". He claims 16 W/channel at 0.01% THD. The output current sharing resistors are only 1 ohm per op amp.
I have done this with no ill effect at modest gain of 2 t0 4 to increase output drive current in Figure 5 of datasheets.maximintegrated.com/.../MX7543.pdf. I do the same thing with logic gates to increase fan-out. I will keep your cautions in mind in the future.
We are using a op-amp based relay driver in a electronic starter where 2 stages of LM324 quad op-amps are paralleled without any current sharing resistances. The relay consumes around 35mA which is a bit too much for a single stage. We have sold thousands of these without any complaints whatsoever.
Of course this is not an amplifier application.....more of a high current buffer which operates on only two voltage levels HI & LO.
In 2010 I used 2 * PA93 in parallel to drive a 500 nF load, 110 Volts pk-pk, DC to 10kHz, using the isolation resistors, required for opamp stability.
The application sections of most operational amplifier data books show emitter followers. I think this is a good solution. One disadavantage is that rail to rail output operation will be compromised.
Sam, paralleling gates to increase fan-out is not a problem at all. Gates will either try to pull to GND or up to VCC, not to a specific voltage in-between. So as long as all gates are pulling into th esame direction, their output can be easily added with no side-effects.
Multiple OpAmps, however, might have a different opinion about the output voltage and therefore fight each other if you do not allow a small output voltage difference between them by coupling resistors.
Sam Green - Did you link to the wrong datasheet? No paralleled op-amps here.
shantanu dasgupta - your coil driver example isn't the same as it is only Hi/Lo, like logic paralleling.
Percy Williams - don't forget the problem of cross-over distortion, unless things are optimally biased.
Bruce Trump - I'd like to see further discussion, possibly with more elegant approaches, to address the problem of achieving good slew rates driving highly capacitive loads while ensuring stability.
Back 40+ years ago I was in a seminar with a leader who knew what he was talking about. He mentioned Gumpert's Law: "When contemplating the design of a direct coupled amplifier, the First thing to consider is DON'T!” I rather think paralleling opamps would also fall into that category. I did put a complementary pair of transistors on the end of an opamp. Bases tied to the opamp output, emitters tied together, and the feedback loop tied to the emitters. If was “interesting” during the zero crossing, but the output was powerful enough for the load and the opamp took care of the zero crossing. It worked and it was stable. A number of engineers looked at it and shook their heads. But it worked and in those dark days before Digikey, I had the parts on hand. It may have been a 709 (yes, we used them) or a 741.
Clearly using a power Opamp is the answer. Just drop it in and forgetaboutit. Engineering is making what you need from what you can get. Today, ELSI devices are on development boards and it is not unheard of to have several boards on the bench, working together. Or Not Working together! Pile of torn hair on the floor, walls turned blue from the frustrated profanity, reading the manual through again...
Remember the CK722 ?
dox: If you take the feedback from the emiiters rather than the bases, then the overall circuit including the op-amp and the two transistors acts like an op-amp and cross-over distorsion is eliminated.
Sam Green and Jens-Michael Gross: I have just done a Google search about connecting gates in parallel to increase the fan out capability, and have come up with nothing for or against. I would guess that there might be a slight switching delay from one gate to another, which at high frequency may cause overheating, due to current passing from VCC through one output into the other and to ground. This effect could be minimised by only connecting gates in the same package. The effect of "Shoot-through" could be reduced by connecting the outputs together via small value resitors in each output. Does anyone have anything to say about this?
Percy, your gues sis right, if there are different propagation times, then the two gates might have complementary output for a short time. Using gates from the same package should indeed minimize the effect. Output series resistors will reduce this, but also reduce the maximum current then and/or increase the output voltage drop.
On the MSP430 processors, often two output port pins are used in parallel, not so much to increase the fan-out, (the MSPs can drive up to 40mA per pin, depending on supply voltage and tolerable output voltage drop) but rather to reduce the voltage drop at medium currents.
However, for standard gates, it might be an idea to change the technology rather than paralleling gates. (e.g. HCT instead of LS)
@ Bill Grenoble : I liked the 709, nice op amp , not the same common mode voltage range as the K2-W , but a bit faster.
re buffering: If you used a AC126 / AC127 pair as emitter followers, the cross over distortion would be negligible.
An older thread but ... @Charles Hansen1 - Doug Self's design parallels 4 x NE5532 with 10R isolation resistors. However, note that his design goal is noise reduction not output drive capability. The rationale behind parallel op amps to reduce noise is explained in his book "Small Signal AudioDesign."
NwAvGuy (nwavguy.blogspot.jp/.../o2-details.html) parallels 2 x NJM4556 in his O2 headphone amp design with 1R isolation resistors. NwAvGuy's design goal is increased output drive
My original design for parallel op amps was published in1984 in a Burr-Brown data sheet for the OPA111, a then- new dielectrically- isolated op amp. It is shown as Figure 16, 'An N-Stage Parallel-Input Amplifier for Reduced Relative Amplifier Noise at the Output." I called this approach "The Lunatic Fringe Amplifier"- but not for publication. Jim Williams and apparently a few audio folks have later promoted this paralleling idea. At the time I came up with this, my friend Mark Stitt doubted that it would actually result in lower noise but was intrigued enough with this unusual architecture to breadboard the circuit and test it. It did work.
The noise reduction takes advantage of the fact that the op amps' uncorrelated noise adds as the square root of the sums while the signal, which is correlated, adds directly. This statistical trick also applied to noise and drift reduction of summed voltage references- another idea that Mark (and maybe Bruce) breadboarded and tested long-term with success. That test PCB was stuffed full of voltage references, all summed together into one low-noise, very stable output.
Speaking of Mark Stitt, a brilliant Mechanical Engineer who became an equally brilliant electronics engineer, someone should write an article about the now- disbanded Burr- Brown "cast of characters".
All content and materials on this site are provided "as is". TI and its respective suppliers and providers of content make no representations about the suitability of these materials for any purpose and disclaim all warranties and conditions with regard to these materials, including but not limited to all implied warranties and conditions of merchantability, fitness for a particular purpose, title and non-infringement of any third party intellectual property right. No license, either express or implied, by estoppel or otherwise, is granted by TI. Use of the information on this site may require a license from a third party, or a license from TI.
TI is a global semiconductor design and manufacturing company. Innovate with 100,000+ analog ICs andembedded processors, along with software, tools and the industry’s largest sales/support staff.