Photodiode Transimpedance Amplifier Oscillation
A photodiode-based optical receiver uses a transimpedance amplifier with a 1 MΩ feedback resistor. The photodiode has 12 pF junction capacitance and is biased at zero volts; its cable adds 8 pF, and the op-amp input capacitance is 3 pF. The op-amp has a 10 MHz gain-bandwidth product, 5 pA input bias current, and rail-to-rail output. There is no feedback capacitor; the layout uses a short guard-ring input. At low light the output is quiet, but when a 50 kHz modulated signal is applied the output shows a 1.2 MHz burst oscillation that grows with photodiode cable length. The required signal bandwidth is 100 kHz. The production test limits output offset to ±20 mV and requires at least 45° phase margin. A junior engineer proposes adding a 100 pF feedback capacitor to 'filter the noise'.
1. Diagnose the oscillation using the input and feedback poles. Explain why increasing cable length makes it worse.
Read the worked answer
The total capacitance at the inverting input is C_in = 12 pF + 8 pF + 3 pF = 23 pF. With R_f = 1 MΩ, the input pole is at f_p = 1/(2π R_f C_in) ≈ 1/(2π × 1 MΩ × 23 pF) ≈ 6.9 kHz. That pole adds phase lag inside the transimpedance loop, while the op-amp's 10 MHz gain-bandwidth product supplies gain. With no feedback capacitor, the feedback network has no zero to cancel the lag, so the loop can cross 0 dB with very little phase margin and ring near the gain crossover. A longer cable raises C_in, lowers f_p, adds more phase lag, and reduces the margin further, which matches the observed growth of the 1.2 MHz burst. The 5 pA bias current gives only about 5 µV across 1 MΩ, so offset is not the cause. The fix is a feedback capacitor, not shielding alone.
2. Calculate a feedback capacitor that gives roughly 45° phase margin, and check whether the resulting bandwidth still meets the 100 kHz requirement.
Read the worked answer
A standard design estimate is C_f ≈ sqrt(C_in/(2π R_f GBW)). Substituting C_in = 23 pF, R_f = 1 MΩ, and GBW = 10 MHz gives C_f ≈ sqrt(23 pF/(2π × 1 MΩ × 10 MHz)) ≈ 0.6 pF. A 0.5–1 pF capacitor is appropriate; 0.6 pF is not a standard value, so use 0.5 pF or 1 pF and verify by simulation or measurement. The feedback zero is at f_z = 1/(2π R_f C_f) ≈ 265 kHz for 0.6 pF, and the closed-loop bandwidth is on the order of 250 kHz, so the 100 kHz signal passes. The phase margin should be checked over the cable capacitance range, not just at the nominal 8 pF. A variable or socketed photodiode cable changes C_in, so specify the maximum cable length and recheck margin at that value. The 45° target is a minimum, not a comfortable design margin.
3. Evaluate the proposed 100 pF feedback capacitor. What happens to bandwidth, noise, and offset?
Read the worked answer
A 100 pF feedback capacitor creates a feedback zero at f_z = 1/(2π × 1 MΩ × 100 pF) ≈ 1.6 kHz. Above that frequency the transimpedance rolls off, so a 100 kHz signal is attenuated by roughly 36 dB relative to the low-frequency gain. That destroys the required signal bandwidth and is not an acceptable 'noise filter'. The capacitor does improve phase margin and reduces high-frequency noise gain, but it does so by over-damping the signal path. Offset is unaffected: the 5 pA bias current still produces about 5 µV across 1 MΩ, well inside ±20 mV, unless dark current or leakage is much larger. The correct approach is a small capacitor in the 0.5–1 pF range, selected for the actual C_in and verified for 45° phase margin, plus a guard ring and clean cable handling. If more noise filtering is needed, place it after the first stage, not directly across the 1 MΩ resistor.