The Gold Standard of Photonics | Transimpedance Gain, Photodiode Noise and the Bandwidth Trade-Off

The light reaching a photodiode is optical power. Its responsivity at a given wavelength converts part of that power into photocurrent, under appropriate operating conditions. A transimpedance amplifier turns the current into a voltage that an ADC, counter or subsequent filter may process. Each stage has its own calibration and failure modes.

If an optical source emits 5 mW and only 2 mW reaches the detector, an electronic reading cannot recover the lost 3 mW without a validated optical transmission model. Likewise, a voltage drift might be caused by detector responsivity or amplifier gain rather than by a true change in the source.

Before designing the amplifier, state wavelength, anticipated photocurrent, dark current, required time response, maximum light exposure, temperature, detector junction capacitance and the supply rails. The useful specification is a traceable measurement from incident photons to reported signal, not ‘largest possible resistor’.

[Contents](#pt-tia-contents) · [Next chapter](#pt-tia-chapter-2) · Photonics subject guide

In the ideal inverting topology, an operational amplifier holds its summing input near a fixed reference potential while feedback current flows through resistor Rf. If the photodiode delivers current I into that node under the chosen orientation, Vout − Vref ≈ −I Rf. The sign depends on photodiode polarity and wiring; its magnitude is I Rf.

A 10 μA current through 100 kΩ produces a nominal 1.00 V magnitude because 10×10⁻⁶ A × 100,000 Ω = 1.00 V. The resistor is not an optical gain coefficient. The gain has units volts per ampere, or ohms, and is commonly called transimpedance.

Actual current can include photocurrent, leakage and dark current. Feedback capacitance adds frequency dependence, and the op amp introduces finite open-loop gain, offsets and noise. Treating I Rf as a complete high-speed design formula overlooks the physical origin of the photocurrent and the dynamics of the feedback loop.

[Contents](#pt-tia-contents) · [Previous chapter](#pt-tia-chapter-1) · [Next chapter](#pt-tia-chapter-3) · Photonics subject guide

For an ideal linear photodiode at a specified wavelength, Iphoto = ℛ(λ) Pincident, where ℛ(λ) is responsivity in A/W. Suppose a fictional responsivity of 0.50 A/W and a detector optical power of 20 μW. The nominal photocurrent is 0.50 × 20×10⁻⁶ = 10 μA.

The ideal low-frequency output for 100 kΩ feedback is then 1 V in magnitude. If illumination doubles and the detector remains linear, the current and ideal voltage double. That scaling breaks when the amplifier saturates, photodiode collection changes or gain and noise depend on operating point.

A spectral change matters even at unchanged total incident watts because ℛ depends on wavelength. The earlier responsivity drift guide explains why subtracting dark output is not the same as calibrating detector gain.

Always distinguish incident power at the sensitive area from total source power. Beam clipping, angular acceptance and window losses are upstream optical questions; the transimpedance design begins with the current the junction actually produces.

[Contents](#pt-tia-contents) · [Previous chapter](#pt-tia-chapter-2) · [Next chapter](#pt-tia-chapter-4) · Photonics subject guide

Under a simple Poisson-current model, root-mean-square shot noise across an ideal equivalent noise bandwidth B is ish = √(2q I B), where q ≈ 1.602×10⁻¹⁹ C. For I = 10 μA and B = 100,000 Hz, ish ≈ 0.566 nA rms.

Multiplying by 100 kΩ gives about 56.6 μV rms at the output in an ideal flat transimpedance band. This is a constructed noise estimate, not a universal amplifier specification. It describes only shot noise associated with the specified current and bandwidth.

The formula changes in practice with total shot-noise-generating current including relevant dark contributions, detector multiplication, excess noise, optical correlations and the actual frequency response. Equivalent noise bandwidth is not always equal to a circuit’s quoted −3 dB bandwidth.

Reducing B by a factor of four reduces this white-noise rms contribution by a factor of two, while it also reduces the bandwidth available to follow rapid light changes. The trade-off must be made against the required event duration.

[Contents](#pt-tia-contents) · [Previous chapter](#pt-tia-chapter-3) · [Next chapter](#pt-tia-chapter-5) · Photonics subject guide

An ideal resistor at absolute temperature T produces thermal current-noise spectral density referred to the input of √(4kBT/Rf), where kB is Boltzmann’s constant, under the simplified feedback model. Over an ideal rectangular noise bandwidth B, irms = √(4kBT B/Rf).

At T = 300 K, B = 100 kHz and Rf = 100 kΩ, the input-equivalent resistor contribution is about 0.129 nA rms. The corresponding output-equivalent value is about 12.9 μV rms. This is smaller than the 0.566 nA fictional signal shot noise, but should not be omitted when signals become weaker.

If independent white-noise sources are well modeled as uncorrelated, their rms values combine in quadrature. Here √(0.566² + 0.129²) ≈ 0.580 nA input-referred, ignoring op-amp and other detector noise. A measured result that disagrees with this simplified budget does not automatically indicate an optical-source problem.

[Contents](#pt-tia-contents) · [Previous chapter](#pt-tia-chapter-4) · [Next chapter](#pt-tia-chapter-6) · Photonics subject guide

A capacitor Cf placed in parallel with feedback resistor Rf gives a simple network impedance Zf(s) = Rf/(1+sRfCf). For an ideal amplifier in this restricted first-order feedback model, the magnitude rolls off around fRC = 1/(2πRfCf). The capacitor shunts some feedback at higher frequencies.

With Rf = 100 kΩ and Cf = 1 pF, fRC ≈ 1.59 MHz. Doubling Cf to 2 pF gives about 796 kHz. Raising Rf to 1 MΩ while keeping Cf at 1 pF gives about 159 kHz. These are calculations of the feedback impedance pole, not guarantees about actual closed-loop bandwidth or adequate phase margin.

A physical photodiode adds junction capacitance. The amplifier’s finite gain–bandwidth product, input capacitances, stray board capacitance and noise gain all affect the transfer function. Some design limits may appear at frequencies below or above the nominal RC pole.

Report which bandwidth definition is used—signal −3 dB, noise-equivalent, settling-time or a full-system communications criterion—because each answers a different measurement question.

[Contents](#pt-tia-contents) · [Previous chapter](#pt-tia-chapter-5) · [Next chapter](#pt-tia-chapter-7) · Photonics subject guide

The ideal 10 μA × 1 MΩ product is 10 V in magnitude. A circuit powered from modest rails may not be able to swing that far, even at DC. Saturation distorts both amplitude and settling. The amplifier can produce a convincingly flat but wrong high-light reading.

Large gain can also amplify leakage, offset-related effects and low-frequency drift in the measurement chain. A gain setting suitable for detecting nanowatt-scale signals may be completely unsuitable for a photodiode exposed to a bright flash.

A good design provides gain range or sufficient headroom, defines worst-case photocurrent and checks recovery from overload. A stable 1-V result with adequate noise performance may be more useful than a nominal 10-V sensitivity that clips the actual signal.

Do not interpret a clipped output as evidence that optical power has reached a physical plateau. Test linearity by varying known illumination over the desired dynamic range and measuring the observed current-to-voltage response.

[Contents](#pt-tia-contents) · [Previous chapter](#pt-tia-chapter-6) · [Next chapter](#pt-tia-chapter-8) · Photonics subject guide

The photodiode junction and amplifier inputs present capacitance at the inverting node. Coupled to the feedback network and finite amplifier open-loop gain, they modify noise gain and feedback phase. The circuit may ring, peak or oscillate even when the resistor and capacitor look reasonable on a simple schematic.

Some feedback capacitance can compensate the circuit, but the required amount depends on the amplifier, junction capacitance, resistor and layout parasitics. An ideal RC calculation that ignores these terms is a teaching limit, not an adequate compensation design.

Use a relevant manufacturer transimpedance design procedure with the actual amplifier gain–bandwidth and capacitance specifications. Simulate loop response with a realistic model and verify the built circuit through controlled optical or electrical step testing.

Very small capacitors may be comparable with the board’s stray capacitance. If the assumed 1 pF is effectively 2 pF after layout and packaging, the calculated feedback pole changes accordingly, even before the more complicated stability limits enter.

[Contents](#pt-tia-contents) · [Previous chapter](#pt-tia-chapter-7) · [Next chapter](#pt-tia-chapter-9) · Photonics subject guide

Characterize the detector in darkness, under steady known illumination and under a modulated test source. Record reference current or optical power, output mean, rms noise in a stated bandwidth, frequency response and evidence of ringing or saturation.

Noise may come from the detector’s dark current, shot statistics, resistor thermal agitation, amplifier input current and voltage noise, power supplies or electromagnetic pickup. The scaling of noise with illumination, gain and bandwidth can help distinguish them, but each test must control the other parameters.

An output-noise voltage alone is not a full noise metric. To compare two gains, refer noise to the detector input under appropriate transfer functions, or express the minimum detectable optical signal using calibrated responsivity and bandwidth.

The table separates observable checks from conclusions that need more evidence. Traceability includes the photodiode, electronics and optical stimulus, not just an oscilloscope screenshot.

[Contents](#pt-tia-contents) · [Previous chapter](#pt-tia-chapter-8) · [Next chapter](#pt-tia-chapter-10) · Photonics subject guide

Does doubling the feedback resistor always double measurement quality? No. It doubles ideal voltage per ampere, but it can change output headroom, resistor noise expressed in volts and system bandwidth and stability.

Can shot noise be removed by subtracting a dark reading? No. Subtraction can correct an estimate of an offset, but stochastic shot noise remains in a measurement and must be addressed statistically.

Does the feedback RC pole equal the real photodiode amplifier bandwidth? Not necessarily. The op amp’s loop dynamics and input capacitance can impose stricter limits or peaking; the simple pole describes one part of the network.

Does a high gain imply a lower optical detection limit in every case? No. It depends on input-referred noise, detector responsivity, integration bandwidth, environmental interference and whether subsequent electronics are limiting.

[Contents](#pt-tia-contents) · [Previous chapter](#pt-tia-chapter-9) · [Next chapter](#pt-tia-chapter-11) · Photonics subject guide

Reduce photocurrent to 1 μA while holding Rf = 100 kΩ and ideal B = 100 kHz. The signal becomes 0.10 V. Shot noise falls by √10 from the 10-μA case to about 0.179 nA rms; feedback resistor thermal input noise stays near 0.129 nA rms at unchanged temperature and bandwidth.

The simplified quadrature noise becomes √(0.179² + 0.129²) ≈ 0.221 nA input-referred. At the output that is around 22.1 μV rms. The useful voltage is lower than before, but judging quality from voltage amplitude alone misses the changing signal-to-noise ratio.

If you raise Rf to 1 MΩ to restore a nominal 1-V output for 1 μA, reconsider bandwidth, compensation, thermal noise and headroom. The feedback-network pole with 1 pF would be about 159 kHz; the actual closed-loop design must still be independently validated.

[Contents](#pt-tia-contents) · [Previous chapter](#pt-tia-chapter-10) · [Next chapter](#pt-tia-chapter-12) · Photonics subject guide

The gold standard is to report optical power at the detector, responsivity, expected current range, feedback values, input capacitance, modeled noise and measured frequency response. The feedback resistor’s voltage gain is only one element of that chain.

The Photonics subject root connects this work to source power, waveguides, detector response and calibration. Cascaded optical losses explains why uncertainty in readings at successive optical planes can be correlated.

For a different optical bottleneck, étendue and ray acceptance studies how much light geometry can deliver before the photodiode sees any photons. Each link answers one narrower question without confusing optical collection and electronic amplification.

Sources and further reading

External sources support the technical principles and their scope. All calculated values are original fictional teaching examples and should not be mistaken for field, product or laboratory measurements.

• Texas Instruments: What You Need to Know About Transimpedance Amplifiers, Part 1 — Current-to-voltage conversion, capacitance and loop compensation
• Analog Devices: Transimpedance Amplifier Noise Considerations — Feedback noise, capacitance, gain–bandwidth and stability limitations
• NIST: Photonic Radiometry — Optical power and photodetector metrology context
• MIT OpenCourseWare: Fundamentals of Photonics — Optical detectors and signal foundations

Gold Standard of Photonics · How X Works · Gold Standard of Education

[Contents](#pt-tia-contents) · [Previous chapter](#pt-tia-chapter-11) · Photonics subject guide