What’s this?
Data sheets are one thing. Real world performance is another. Graphics cards and CPUs have benchmark comparison sites, but we humble analog circuit people don’t have an equivalent.
I was recently in need of a vaguely sinusoidal oscillator with a simultaneous pulse/square output, of around 120 KHz. A huge number of op amps should be able to do this without a problem—including all of the quad opamp parts I had. Now, I quite like the pinout of op amps, and having an extra op amp to buffer the output of my simple oscillator circuit would be a definite advantage over a dual part. So, I made an inventory of the reasonably nonexpensive quad op amps I had in my shelves, and came up with:
- LM2902 in two versions (Texas Instruments and STMicroelectronics), 1MHz GBW bipolar op amp; output goes to negative rail
- TL074, JFET input, 3MHz GBW op amp.
- LF347, JFET input, 4MHz GBW op amp.
- TL974, Low noise high-speed, rail-to-rail output (not input) bipolar opamp, 18MHz GBW
- TLV2372, CMOS rail-to-rail input & output op amp, 3MHz GBW
Op amp-based oscillators are not high frequency; the phase shift across frequencies accumulate rather quickly; therefore,
The test setting
The test

The actual circuit was built on a piece of simple, tri-pad copper

The frequency is adjusted with two potentiometers (10K and 100K). The filter capacitors are all polyester film capacitors, whereas the charge pump uses tantalum and multilayer ceramic capacitors. The main power supply is itself bypassed by a .47µF ceramic capacitor, within close distance to both integrated circuits.
The test setup uses a Rigol DS1102 100MHz/1GSS oscilloscope with x10 500MHz probes from Tektronix.
Test results
The yellow trace is the TRIANGLE output; the blue trace is the SQUARE output in all of the tested devices.
The LM2902 from Texas Instruments

- Maximum frequency: 8.9 KHz
- Triangle waveshape distortion: >20%
- Square waveshape sloping: >50% of
cycle
The worst performer of them all. The LM2902 is very similar to the

The most likely cause of this is poor phase margin – as the frequency increases (as well as that of the harmonics), you eventually get to a phase shift that causes destructive interference. Therefore, a set maximum frequency will appear. If you decrease R or C above that, the frequency will again begin to drop.
The LM2902 from STMicroelectronics

- Maximum frequency: 11.5 KHz
- Triangle waveshape distortion: >20% (clipping)
- Square waveshape sloping: >50% of cycle
I’ll be damned! Just because a part has the same number, doesn’t

The STM part performs better than the TI part – but still with a maximum frequency an order of magnitude less than what the goal was. Also, note that the clipping artefacts on the “triangle” wave can’t be reduced with reduced gain in the circuit. It won’t slam against its output maxima, but the wave shape will still be clipped off.
All in all, no matter where you get the LM2902 – an old National part, the TI-branded part, or STMicroelectronics, it’s a piss-poor chip for oscillator purposes. As is its relatives (LM358, LMx24).
The TL074 from Thomson (STMicroelectronics)

- Maximum frequency undistorted: 88.7 KHz
- Triangle waveshape distortion: <3%
- Square waveshape sloping: <10% of cycle
A true classic. The TL074 is a nice, FET-input, reasonable-bandwidth

Now, we’re within striking distance of our target goal. In fact, I’d be happy with the TL074’s results – it’s cheap, and you can hit about 200KHz with it if you’re willing to accept a lot of distortion. But, this distortion can be filtered out with a few RC stages… Not bad! Not bad at all!
The LF347 from Texas Instruments

- Maximum frequency undistorted: 100 KHz
- Triangle waveshape distortion: <2%
- Square waveshape sloping: <7% of cycle
Very similar to the TL074, but the increased gain-bandwidth product reveals itself in an increased distortion-free frequency. And, it is, in fact, cheaper than the TL074—at least if you’re ordering new TI stock for both. However, it comes at the cost of increased harmonic distortion and noise density. This seems like a good all-rounder for the oscillator application, and earns a definite recommendation from here.

The TL974 from Texas Instruments

- Maximum frequency undistorted: 140 KHz
- Triangle waveshape distortion: <4%
- Square waveshape sloping: <25% of cycle
Impressive! An all-bipolar operational amplifier performs 40% better

However, there does seem to be a bit of an amplitude issue with
Editor’s note I actually managed to destroy a TL974 during testing. It seems to be less resistant to voltage transients as well as loading than the other opamps.
The TLV2372 from Texas Instruments

- Maximum frequency undistorted: 25 KHz
- Triangle waveshape distortion: <2%
- Square waveshape sloping: <2% of cycle
Well, well, well. We all thought that expensive, rail-to-rail input/output CMOS parts were here to save the world. And then, they perform miserably; worse than their gain-bandwidth product would indicate.
If you increase the frequency beyond the 25KHz mark in this test application, you end up with

Therefore, it seems that CMOS parts are exclusively for
However; these lackings of the “lesser parts” can be corrected for. The
Conclusion
So which would you pick? Honestly, the LF347, although it misses
The LF347 is about 0.2 euros (for 1 piece) from most suppliers,
But today – just for today – the humble LF347 wins!