February 28, 2006 AT 8:02 pm

Specifying an op-amp

So you need an op amp…and you can’t use just any generic op amp because of certain requirements. For example, a low supply voltage (

Finding the right op amp isn't hard, here is an example of how to go about your search. For this example, I will specify the op amp to be used for the circuit from last week, a tape-head preamplifier. The schematic used an OP37. However, this op amp requires a 8V powersupply, and I'm going to be using only a 3V battery. Also, its good to see if there are any cheaper alternatives.

Specifically, the op amp is going to amplify a 10uV peak-to-peak signal by 1000 (up to 10mVpp). The op amp should work from 20Hz to 20KHz (optimal audio range), & be not too tough to solder.
First, write down what the constraints are:

  • Has to run on a 3V “single supply” and we’ll have our DC offset at 1.5V
  • Input offset voltage (Vos) must be
  • Since we need to handle such a large swing, it should be rail-to-rail to within .2V of the rails to give us space.
  • 1000 gain at 20KHz means the gain bandwidth (GBW) must be > 20MHz (in reality it can be much lower because the tape probably isnt good enough to record past 12KHz)
  • Since our input signal is 10uV then (at worst) we want to have less than 5% noise (which isn’t so good but we’re willing to have poor quality audio). 5% of 10uV = 500nV. 500nV/sqrt(20000) = 3.5nVrtHz noise figure.
  • Package should be SOIC for easy soldering
  • Low power is nice but not -that- necessary, maybe
  • Low price! Must be under $2 at quantity 100, the lower the better

Stuff we don’t care about:

  • Operating temperature
  • Slew rate (at 10mV and 20KHz, it can be really slow)

The OP37, for comparison, has 12MHz GBW, requires 8V power supply, not rail-to-rail, Vos = 0.03mV, and noise figure of 3.2nVrtHz. So: great offset, great noise, so-so bandwidth and incompatible power requirements.

Lets go to TI and see what they have to offer. Select >=16MHz GBW, rail-to-rail and 8-SOIC package. The only chip they’ve got is the OPA350.

Now lets try National. Their system is a little tougher to use: click on 1mA offset first, which will pare down the options to 5 items, none of which are SOIC. You’ll notice national doesnt have a very good selection of low noise, ~20MHz op amps.

Next, we go to Analog Devices. Enter in package = SOIC, Vcc-Vee = 3V, noise

Checking STMicro (click on “low noise”) they don’t have anything that runs on 3V.

Linear Technologies has a bunch of incredibly low noise op amps, at extraordinary bandwidths, but they’re rather expensive, at least $3 which is outside of our budget.

Maxim has a few good options (sort by noise, then compare all of the ones 20MHz GBW, that leaves the MAX4488.

Part Noise (nVrtHz) GBW (MHz) Vos (mV) Iq (mA) $/qty 100
OP37 3.2 12 0.03 5.5 $1.07
OPA350 5 38 0.5 7.5 $1.73
AD8655 4 29 0.05 4.5 $0.88
MAX4488 4.5 42 0.75 2.5 $0.72

There are a couple other manufacturers but we’ve covered the most common. So let’s stop here. The OPA350 is clearly not a good choice, it’s the most expensive, noisiest and most power-hungry. So we’ll just ignore that, leaving the AD8655 and MAX4488. Now its just a comparison between price, power and noise. Since our noise figure is already higher than we’d like, I’ll place priority on that: the price difference is pretty small and power isn’t a huge priority.

Finally, we have chosen the AD8655. Yay Analog Devices!



  1. This is a nice example and I suspect you’re aware of the following but I wanted to mention a couple things for the potential edification of other readers:

    1. The Art of Electronics also contains some good information on op-amp selection.
    2. In a somewhat demanding amplifier design such as the one in your example, it is sometimes also worth considering a two-stage design. Sometimes two op-amps of somewhat lesser specs (purchased as a dual monolithic device) cascaded appropriately work as well as and cost less than a single fancier one. You generally wouldn’t be able to get out of whatever noise requirement you have (for the first stage anyways) but the requirement for gain-bandwidth product of each op-amp is reduced appreciably.

  2. I would like to mention that the AD8655 actually has a thermal noise floor of 2.7nV/rtHz.

Sorry, the comment form is closed at this time.