If a datasheet (like AD828) says that an opamp is stable at Gain >2 (or reccomends to work with G>2, hence it is clearly not unity gain stable), what can we deduct about its stability in the inverting configuration at G=-1; G=-2 or G<<-2 (like in any transimpedance amplifier configuration)? Is it always instable in the three above cases if not compensated?
Loop gain is the stability determining factor.
Loop Gain = Beta * Ao where Beta = feedback fraction = R1/(R1+R2) and Ao = open loop gain.
1/Beta = Noise Gain.
So a non inverting amplifier with a closed loop gain of 2 (R1=R2, Beta = 0.5 and Noise Gain=2) has the same Beta and therefore the same noise gain as an inverting amplifier with a closed loop gain of -1 (R1=R2, Beta = 0.5 and Noise Gain = 2).
This means that an inverting amplifier with a gain of -1 is as stable as a non-inverting amplifier with a gain of 2.
In addition to Noise Gain being the stability determining factor, Noise Gain also determines the bandwidth of an amplifier.
Bandwidth = GBW/Noise Gain.
So a non-inverting amplifier with a gain of 2 (R1=R2) has the same bandwidth as an inverting amplifier with a gain of -1 (R1=R2). If you make the closed loop gains of the two amplifiers both equal to 2 then the inverting amplifier will have a bandwidth equal to 2/3 the bandwidth of the non-inverting amplifier.
Non-Inverting amplifier with a closed loop gain of 2 has R1=R2 and a noise gain of 2. Inverting amplifier with a closed loop gain of 2 has R2=2*R1 and a noise gain of 3.
Stabilty is a function of the total feedback phaseshift.
1) Rout + Cload: 100 ohms and 100pf are 10,000 picosecond time constant, producing 45 degrees phaseshift at 100 MegaRadians/second of 16MHz. Many opamps have Rout (internal output resistance) near 100 ohms; some have Rout >>> 1Kohms.
2) phase margin beyond 90 degrees: a 60 degree phase margin opamp (Unity Gain phase margin) has 90+30 = 120 degrees phase shift
3) phase shift at the virtual_ground node: assume 10pF on that node, and resistive equivalent (Rin || Rfb, or Rg || Rfb) of 1,000 ohms; this produces 10,000 picosecond tme constant, or 45 degrees at 16MHz.
What rescues a feedback network? Usually the parasitic feedback capacitance in parallel with the feedback resistor. IMHO