I've a simple buffer amp (AD8044 video amp, 150MHz, high slew rate) set up to buffer a signal (NTSC monochrome video) as a basic follower.

The circuit is the most basic one and of course includes large and small (>10uF, 0.1, 0.01) bypass caps.

With a lower slew-rate amp (MCP6294) the circuit works as expected, but when the AD8044 is inserted in its place, the output goes wild by swinging between rails with great instability.
Adding a cap (0.05uF) to the output terminal almost eliminates the instability, but there's still a little bit of noise on the signal.

1) Why is the '8044 camp unstable without the output cap? I can see how it "slows down" the frequency response of the feedback loop (essentially filtering), but shouldn't the amp be stable without it? The capacitive load is just a 10x scope probe.

2) Why is there a little noise injected with the faster amp, and how can I remove it?

 

1) You likely have a problem in your layout or decoupling. It isn't enough to cause the lower frequency op amp to oscillate but the higher response of the AD8044 allows the oscillations to form. The output cap is a brute force way to reduce the frequency response and thus the oscillations.

How is the circuit physically built? Is it breadboard or circuit board? Do you have a ground-plane? For high frequency you should have a ground-plane with very short connections for the bypass caps (preferably surface-mount types) between the power pins and ground-plane.

2) A faster op amp has a wider bandwidth and thus a larger noise bandwidth. Random circuit noise is proportional to the square-root of the bandwidth. A small series resistor and parallel cap at the output can reduce the bandwidth and noise.

 

1) I'm working to test out my pcb-based design. The board is already laid out (4-layer, with a ground plane) and will be sent to a boardhouse soon, but I wanted to check the circuit. So right now this was put onto a solderless breadboard.
I expected the extra capacitance from this setup to cause some variations from the pcb design, but I didn't think it'd be this significant. It's possible, then, that this circuit will work on the pcb, since that indeed does have a ground plane and proper bypass caps; I might leave some open pads for adding a resistor and cap if needed.

2) Thanks. Is clean high speed output always a result only of proper filtering, then? But those filters can also inject noise.

 

AD8044 is almost guaranteed to oscillate on a solderless BB. Follow Carl's suggestions for supply decoupling.
The output of the AD8044 has push-pull collectors, so it has a high open-loop output impedance. Any small load capacitance, like your BB or a scope probe, adds a pole in the feedback loop which will reduce the phase margin, and possibly push it into oscillation. A large capacitor like you added will probably become a dominant pole, which will stabilize the loop but kill your bandwidth.
Read the datasheet about driving capacitive loads. It recommends adding a resistor in series with the output. With a 20pF scope probe, you could go as high as several hundred ohms and still get flat response to 5MHz (NTSC monochrome BW).
Why are you using such a high bandwidth op amp for a 5MHz signal? It is overkill, and makes your design unnecessarily "touchy".

 

1) I'm working to test out my pcb-based design. The board is already laid out (4-layer, with a ground plane) and will be sent to a boardhouse soon, but I wanted to check the circuit. So right now this was put onto a solderless breadboard.
I expected the extra capacitance from this setup to cause some variations from the pcb design, but I didn't think it'd be this significant. It's possible, then, that this circuit will work on the pcb, since that indeed does have a ground plane and proper bypass caps; I might leave some open pads for adding a resistor and cap if needed.

2) Thanks. Is clean high speed output always a result only of proper filtering, then? But those filters can also inject noise.

You are supposed to breadboard the circuit before you build the PCB.

If you need to build a test circuit for such a high frequency design you can not use a solderless breadboard. They have high capacitance and significant coupling between nodes. You should use a blank breadboard PCB with a solid copper ground with as short interconnect leads as possible.

How can a filter inject noise?

I agree with Ron. You shouldn't use an amp that has a significantly higher bandwidth then you need. It creates a greater chance for instability/oscillations and increases the wide-band noise.