The basic idea is to add a couple of good low-noise FETs in front of an existing op amp. Most op amps don't operate the front-end transistors as rich as the output. Yet in a case like this, there's no reason at all not to run more current through the front end than in the rest of the op amp. My first pick is the 2N5486, which has less than 1 pf of CRSS, but has a lot of gms (4 millimhos) and low voltage noise (at IS = 3 mA). So for my first design, I'll just put a matched pair* of 2N5486s in front of a decent wideband op amp, such as the LM6171 (Fig. 5). What's the voltage noise of this amplifier? We may be able to get an average of 3 nV/√Hz, out to 10 kHz.
When you're designing an op amp, remember this: adding gain is one of the cheapest things you can add. You on-ly need to be careful about how to give that gain away—to roll it off.
In this case, it's easy. The R1-C1 network in Figure 5 just rolls off the gain for a fairly smooth frequency response. To achieve 2 MHz of bandwidth and a fairly good, smooth 6-dB/octave rolloff, I suggest R1 = 75 Ω, and C1 = 100 pF as a good place to start your design.
But now, look at the refinements in Figure 6. We can roll off the amplifier's gain simply in two swoops. The low-frequency gain is rolled off by RX and CX. Then after the gain rolls off flatly, we roll it off some more by RY and CY. When we are finished, it should look something like curve X in Figure 7. This isn't exactly rocket science. We just want to make it a practical design. But this is a whole system design. You can't very well design and optimize the op amp alone. It's the op amp, the feedback system, the noise filters, and the post-amplifiers that have to be considered and optimized all together. My first-hack proposals for these damping/stabilization components are:
RX = 5.1 kΩ, CX = 50 pF
RY = 330 Ω, CY = 7.5 pF
The whole point behind making your own op amp is that you do not have to just build an op amp with a smooth 6-dB/octave rolloff, all the way out to a few megahertz. You can roll off the gain at a 6 dB/octave out to some intermediate frequency, and then flatten out the gain. Then, at a higher frequency, let it roll off some more in some vaguely controlled way. This would make a lousy general-purpose op amp, but it might be ideal for a case where the noise gain is rising, such as in a transimpedance amplifier. (Look at the old LM709. When you choose the correct damping networks, it can provide a gain of 1000 out to some high frequency like 1 MHz.)
Also note that I added a second pair of 2N5486s to improve the voltage noise. Yes, this will approximately double the input capacitance. But if your CS is already large, this may easily improve the signal-to-noise ratio. If it's good to have two, will three be better? I'll let you figure that out! But, yes, four or five may provide definite improvements... or that might not be the case.
I won't recommend that you design your own op amp if you can buy one that does the job. But if the best one you can buy isn't good enough, then there's some hope here. Designing your own composite op amp is not that hard, and not that expensive, even if you are going to build one or 10 or 1000. The post-amplifier can be inexpensive. Of course, all of the basic designs will be somewhat different if you are running on ±5-V supplies, or ±15-V supplies.
Either way, it's not that difficult, but the design compromises are slightly different. Here, I just showed a couple of ±15-V applications. (The ±5-V designs differ mostly by using a low-voltage, rail-to-rail-output op amp.)
In future columns on this topic, I will comment on other aspects of design and optimization for transimpedance am-plifiers.
Meanwhile, try to avoid Tee networks in the feedback network. They often cause poor signal-to-noise ratios. Next time, I'll explain that completely. Yes, a Tee network might help you avoid buying 1000-MΩ resistors, but that's only okay when you have proven that the noise is okay.
All for now. / Comments invited!
RAP / Robert A. Pease / Engineer
rap@galaxy.nsc.com—or:
Mail Stop D2597A
National Semiconductor
P.O. Box 58090
Santa Clara, CA 95052-8090
*For this case, grade a good number of 2N5486s into 20-mV bins of VS, with VGD = 7 V, and IS = 3.8 mA. Take units out of the same bin for good matched pairs.
P.S. If you design in an op amp, try to avoid relying on nonguaranteed characteristics, such as noise, which is rarely guaranteed.
P.P.S. I neglected to mention that any resistor may have a built-in capacitance of 0.3 to 0.8 pF. If you add that to any imperfect layout, the capacitance could be so big that you wish it were smaller. Good layout and good engineering can easily cut the C to less than 0.2 pf. For example, make the feedback resistance out of three or four resistors in series, and install a shield land between the ends of the resistor. More later. /rap
Reprints
)
