Photodiode design and characterization for component makers and OEMs

I design and characterize photodiodes and the circuits that read them, from the first device model to the board, the 3D model and the datasheet. Hire me for one piece or for the whole chain.

Discuss a photodiode project Try the live datasheet

Reverse bias V_R10.0 V

Depletion width: 36.8 µm
Capacitance: 14.1 pF
Light: 850 nm

Section view of the device from the live datasheet, not to scale. Move the bias to widen the depletion region.

Interactive photodiode datasheet

Change the reverse bias, temperature, wavelength or load, and the curves and the spec table recalculate from device physics. You can also drag along the dark current, capacitance and responsivity charts to move the operating point.

Si PIN5.0 mm² active areaInGaAs PIN1 mm diameter

Device model

Reverse bias V_R10.0 V

Temperature T25 °C

Wavelength λ850 nm

Load resistance R_L1.00 kΩ

Responsivity R: 0.641A/W
Dark current I_D: 148pA
Capacitance C_j: 14.1pF
Bandwidth f_{−3 dB}: 8.09MHz

Noise current at the operating point

Shot noise, dark current6.9 × 10⁻¹⁵ A/√Hz
Johnson noise, shunt4.0 × 10⁻¹⁵ A/√Hz
Johnson noise, 1.00 kΩ load4.1 × 10⁻¹² A/√Hz

Dark current vs reverse bias

Junction capacitance vs reverse bias

Spectral responsivity

Frequency response

Electrical and optical characteristics, Si PIN, 5.0 mm² active area
Parameter	Conditions	Value
Responsivity Rλ = 850 nm, V_R = 10.0 V	λ = 850 nm, V_R = 10.0 V	0.641 A/W
Quantum efficiency ηλ = 850 nm, V_R = 10.0 V	λ = 850 nm, V_R = 10.0 V	93.5 %
Dark current I_DV_R = 10.0 V, T = 25 °C	V_R = 10.0 V, T = 25 °C	148 pA
Shunt resistance R_shV = ±10 mV, T = 25 °C	V = ±10 mV, T = 25 °C	1.02 GΩ
Junction capacitance C_jV_R = 10.0 V, f = 1 MHz	V_R = 10.0 V, f = 1 MHz	14.1 pF
Depletion width WV_R = 10.0 V	V_R = 10.0 V	36.8 µm
Bandwidth f_{−3 dB}V_R = 10.0 V, R_L = 1.00 kΩ, λ = 850 nm	V_R = 10.0 V, R_L = 1.00 kΩ, λ = 850 nm	8.09 MHz
Rise time, 10 % to 90 % t_rV_R = 10.0 V, R_L = 1.00 kΩ, λ = 850 nm	V_R = 10.0 V, R_L = 1.00 kΩ, λ = 850 nm	60.7 ns
RC limit f_RCR_L + R_s = 1.01 kΩ	R_L + R_s = 1.01 kΩ	11.2 MHz
Transit limit f_trV_R = 10.0 V	V_R = 10.0 V	138 MHz
Diffusion tail, share of signal p_diffλ = 850 nm, V_R = 10.0 V, below 1.25 MHz	λ = 850 nm, V_R = 10.0 V, below 1.25 MHz	13.0 %
NEP, diode alone NEPλ = 850 nm, V_R = 10.0 V, T = 25 °C	λ = 850 nm, V_R = 10.0 V, T = 25 °C	1.24 × 10⁻¹⁴ W/√Hz
NEP with the load NEPR_L = 1.00 kΩ at 25 °C	R_L = 1.00 kΩ at 25 °C	6.33 × 10⁻¹² W/√Hz
Specific detectivity D*Diode alone	Diode alone	1.80 × 10¹³ cm·√Hz/W
Linear range, load-line limit P_maxR_L = 1.00 kΩ, V_R = 10.0 V, λ = 850 nm	R_L = 1.00 kΩ, V_R = 10.0 V, λ = 850 nm	> 10 mW
Breakdown voltage V_BRI_R = 10 µA, 25 °C	I_R = 10 µA, 25 °C	120 V
Active area A5.0 mm² active area	5.0 mm² active area	5.0 mm²

Equation: R = η λ / 1239.84. = 0.935 × 850 / 1239.84. Result 0.641 A/W. 1239.84 is hc/q in eV nm, so λ goes in nanometers and R comes out in amps per watt.Equation: η = (1 − r) (ηfront + ηdrift + ηdiff). = (1 − 0.05) × (0.002 + 0.855 + 0.128). ηdrift ∝ 1 − e to the power (−α W), α = 533 cm⁻¹ at 850 nm, W = 36.8 µm. Result 93.5 %. Light absorbed inside the depletion width is collected at once. Carriers made below it diffuse in, and some recombine on the way.Equation: ID = M I0 2 to the power ((T − 25)/Td) (W/W0) (1 − e to the power (−VR/VT)) + VR/Rleak. = 1.0006 × 25.0 pA × 2 to the power ((25 − 25)/9) × 3.91 × 1.000 + 10.0 V / 200 GΩ. VT = kT/q = 25.7 mV, W = 36.8 µm, W0 = 9.42 µm. Result 148 pA. Generation current in the depleted volume, which doubles every 9 °C, times the gain M that creeps in before breakdown, plus surface leakage through Rleak.Equation: Rsh = 1 / (I0(T)/VT + 1/Rleak). = 1 / (25.0 pA / 25.7 mV + 1 / 200 GΩ). Result 1.02 GΩ. The slope of the dark current at 0 V: generation current over the thermal voltage, in parallel with the leakage resistor. It falls as the diode warms.Equation: Cj = C0 / (1 + VR/Vbi) to the power m. = 55.0 pF / (1 + 10.0 / 0.70) to the power (0.5). Result 14.1 pF. Abrupt junction, m = 0.5. The 300.0 µm layer is still not fully depleted at the 30 V rating, so Cj keeps falling with bias.Equation: W = εr ε0 A / Cj. = 11.7 × 8.854 pF/m × 5.00 mm² / 14.1 pF. Result 36.8 µm. The parallel-plate relation, solved for the plate gap.Equation: |H(f)| squared = D(f) / ((1 + (f/fRC) squared) (1 + (f/ftr) squared)) = 1/2. fRC = 11.2 MHz, ftr = 138 MHz, fD = 1.25 MHz, p = 13.0 %. D(f) = (1 + (1 − p) squared (f/fD) squared) / (1 + (f/fD) squared). Result 8.09 MHz. Solved numerically for the frequency where the response is down 3 dB. D(f) is the slow diffusion share rolling off first.Equation: tr = t(90 %) − t(10 %) of the step response. τRC = 14.2 ns, τtr = 1.15 ns, τD = 127 ns, p = 13.0 %. each τ = 1/(2π f) of its limit, poles in cascade. Result 60.7 ns. The single-pole rule, 0.35 over the bandwidth, would give 43.3 ns. The diffusion tail stretches the edge well past it.Equation: fRC = 1 / (2π (Rs + RL in parallel with Rsh) Cj). = 1 / (2π × 1.01 kΩ × 14.1 pF). Rs + RL in parallel with Rsh = 10.0 Ω + 1.00 kΩ in parallel with 1.02 GΩ. Result 11.2 MHz. The junction capacitance charging through the series and load resistance. The shunt resistance is so large that it barely counts.Equation: ftr ≈ 0.44 v / W. = 0.44 × 1.16 × 10⁶ cm/s / 36.8 µm. v = μE / (1 + μE/vsat), E = (VR + Vbi)/W = 2.91 kV/cm. Result 138 MHz. Holes are the slower carrier: μ = 450 cm²/V s, vsat = 1.0 × 10⁷ cm/s. The 0.44 assumes light absorbed evenly across W.Equation: pdiff = ηdiff / (ηfront + ηdrift + ηdiff). = 0.128 / (0.002 + 0.855 + 0.128). τD = x squared / (2D) = 127 ns, x = 17.1 µm, D = μ kT/q = 11.6 cm²/s. Result 13.0 %. Carriers made below the depletion region diffuse up slowly. This share of the signal arrives late and rolls off above fD = 1.25 MHz.Equation: NEP = square root of (2q ID + 4kT/Rsh) / R. = square root of (2q × 148 pA + 4kT / 1.02 GΩ) / 0.641 A/W. = 7.97 × 10⁻¹⁵ A/square root of Hz / 0.641 A/W. Result 1.24 × 10⁻¹⁴ W/√Hz. Shot noise of the dark current and Johnson noise of the shunt resistance, with the diode at 25 °C. The shot noise is the larger here.Equation: NEP = square root of (2q ID + 4kT/Rsh + 4kT0/RL) / R. = 4.06 × 10⁻¹² A/square root of Hz / 0.641 A/W. shot 6.9 × 10⁻¹⁵, shunt 4.0 × 10⁻¹⁵, load 4.1 × 10⁻¹² A/square root of Hz; T0 = 25 °C. Result 6.33 × 10⁻¹² W/√Hz. The load sits on the board at 25 °C. At 1.00 kΩ its Johnson noise is the largest term by far.Equation: D* = square root of A / NEP. = square root of (0.0500 cm²) / 1.24 × 10⁻¹⁴ W/square root of Hz. Result 1.80 × 10¹³ cm·√Hz/W. NEP scaled by the square root of the area, so parts of different sizes compare on one scale. Diode alone, without the load.Equation: Pmax = (VR + Vbi) / ((RL + Rs) R). = (10.0 V + 0.70 V) / ((1.00 kΩ + 10.0 Ω) × 0.641 A/W). Result 16.5 mW. Above this power the voltage across the load eats into the bias and the response compresses. The table lists anything over 10 mW as > 10 mW.Equation: M = 1 / (1 − (VR/VBR) to the power n). = 1 / (1 − (10.0 / 120) to the power (3)). Result 1.0006. VBR is a property of the device. The model uses it for the gain M that multiplies the dark current near breakdown; the bias stops at 30 V.Equation: Cj = εr ε0 A / W. = 11.7 × 8.854 pF/m × 5.00 mm² / 36.8 µm. Result 14.1 pF. Area is a design choice. Capacitance grows with it, so a part with twice the area has half the RC bandwidth.

Photodiode design process

I can design the photodiode and its readout electronics together, from the first model to the documents your buyers read. Each step works from the outputs of the one before, so the simulation, the board and the datasheet agree with each other.

Photodiode device model
Measured responsivity, dark current, capacitance and shunt resistance become an equivalent circuit I can simulate.
TIA and readout circuit
Feedback resistor and capacitor, op-amp choice, bias, bandwidth, noise gain and stability, worked out by hand first.
MATLAB and SPICE simulation
The circuit runs with the device model across bias, temperature and part tolerances.
Verification against requirements
Every requirement gets a pass or a fail, with the simulation or measurement that backs it.
KiCad schematic and PCB layout
Schematic and board in KiCad, with short input traces and a guard ring around the summing node.
3D model
A STEP model of the board and the photodiode package for your mechanical team.
Datasheet and app note
Datasheet curves and an application note built from the same model and test data as the hardware.

Transimpedance amplifier design

Most photodiodes end up driving a transimpedance amplifier. The feedback resistor sets the gain. The diode capacitance and the op amp's gain-bandwidth decide how fast the stage can run and how much feedback capacitance keeps it stable. This panel uses the diode from the live datasheet at its present bias.

Si PIN, 5.0 mm² active area, at V_R = 10.0 V, 25 °C and 850 nm: C_j = 14.1 pF, R = 0.641 A/W

Feedback resistor R_f100 kΩ

Feedback capacitor C_f521 fF

Optical power P10.0 µW

Output swing V_out: 641 mV
Bandwidth f_{−3 dB}: 3.80 MHz
Phase margin φ_m: 53 °1.2 dB of peaking
Output noise v_n: 431 µVrms, whole band
Input noise i_n: 1.5 × 10⁻¹² A/√Hzat low frequency
Signal to noise SNR: 63 dBI_p = 6.41 µA

Bandwidth estimate

\displaystyle \sqrt{\frac{\mathrm{GBW}}{2\pi R_{\mathrm f}C_{\mathrm{in}}}} = 3.05\,\mathrm{MHz}

Transimpedance gain relative to R_f

FET-input op amp: 100 MHz gain-bandwidth, 4 nV/√Hz, 3 fA/√Hz, 3 pF input capacitance, ±4.8 V output swing.

CAD render of a photodiode readout board with photodiode connections, adjustment controls, and power and signal connectors — Photodiode readout board — CAD layout.

Photodiode pulse response

A short pulse of light hits the diode from the live datasheet, and the scope shows the voltage across its load resistor. Three things set the edge: the junction capacitance charging through the load, the time carriers take to cross the depletion region, and slow carriers that diffuse in from below it. More bias lowers the capacitance and sharpens the edge, while a bigger load slows it.

Si PIN5.0 mm² active areaInGaAs PIN1 mm diameter

Shares the live datasheet's operating point

X1X2100 % 641 µV0 V

Illustrative sample · device driving a resistor load with a 1 µW light pulse. Probe and board capacitance are left out, and the rise time comes from the same model as the datasheet.

Reverse bias V_R10.0 V

Load resistance R_L1.00 kΩ

Wavelength λ850 nm

Pulse width t_p300 ns

What sets the edge

RC charging τ_RC: 14.2 ns; (R_s + R_L ∥ R_sh)·C_j = 1.01 kΩ × 14.1 pF, a 31.2 ns edge on its own
Carrier transit t_tr: 3.18 ns; W/v across the depletion region, a 2.53 ns edge on its own
Diffusion tail τ_D (sets the edge): 127 ns; 13.0 % of the signal diffuses in from below the depletion region and arrives this late

Photodiode specifications

The main electrical characteristics, with the textbook relations the live model uses.

Responsivity and quantum efficiency
$\displaystyle R = \frac{\eta\lambda}{1239.84}$
R in A/W; wavelength λ in nm.
Responsivity is photocurrent per watt of light. Quantum efficiency η is the share of photons that end up as collected carriers. The photon energy ties the two together, so a perfect silicon diode would give 0.73 A/W at 900 nm. Reflection, a thin dead layer at the surface and light that passes too deep keep real parts below that line.
Dark current
$\displaystyle I_{\mathrm D}(T) \approx I_{\mathrm D}(25\,{}^\circ\mathrm C)\,2^{\frac{T-25\,{}^\circ\mathrm C}{T_{\mathrm d}}}$
The current that flows with no light: thermal generation in the depleted volume plus surface leakage. It grows with reverse bias as the depletion region widens, and with temperature. In silicon it typically doubles every 8 °C to 10 °C, so a hot enclosure can raise it more than tenfold.
Junction capacitance
$\displaystyle C(V) = \frac{C_0}{\left(1+\frac{V}{V_{\mathrm{bi}}}\right)^m}$
m ≈ 0.5 for an abrupt junction; ≈ 0.33 for a linearly graded junction.
The depletion region behaves like a parallel-plate capacitor, C = εA/W. Reverse bias widens W, so capacitance falls. In a PIN the curve flattens once the intrinsic layer is fully depleted, which you can see on the InGaAs device above.
Shunt resistance
$\displaystyle R_{\mathrm{sh}} = \left.\left(\frac{\mathrm dI}{\mathrm dV}\right)^{-1}\right|_{V=0}$
The slope resistance of the diode near 0 V, measured at about ±10 mV. It sets the Johnson noise of an unbiased diode and falls as temperature rises, at roughly the rate the dark current grows.
Bandwidth and rise time
$\displaystyle f_{\mathrm{RC}} = \frac{1}{2\pi(R_{\mathrm L}+R_{\mathrm s})C_{\mathrm j}}$ $\displaystyle f_{\mathrm{tr}} \approx \frac{0.44\,v}{W}$ $\displaystyle t_{\mathrm r} \approx \frac{0.35}{f_{-3\,\mathrm{dB}}}$
Three limits stack up. The junction capacitance charges through the load and series resistance, carriers take time to cross the depletion width, and carriers made below the depletion region diffuse in slowly and leave a tail. The tail matters most at long wavelengths and low bias, and it breaks the usual single-pole link between rise time and bandwidth.
NEP and noise
$\displaystyle \mathrm{NEP} = \frac{\sqrt{2qI_{\mathrm D}+\frac{4kT}{R_{\mathrm{sh}}}+\frac{4kT}{R_{\mathrm L}}}}{R}$
Noise-equivalent power in W/√Hz.
Noise-equivalent power is the light level that gives a signal equal to the noise in a 1 Hz bandwidth. The diode adds shot noise from its dark current and Johnson noise from its shunt resistance. The load or feedback resistor adds its own Johnson noise, and with a small load that term is usually the largest by far.
Linearity and saturation
$\displaystyle P_{\max} \approx \frac{V_{\mathrm R}+V_{\mathrm{bi}}}{(R_{\mathrm L}+R_{\mathrm s})R}$
Photocurrent follows optical power over many decades. With a resistor load, the output stops being linear as the voltage across the load approaches the reverse bias plus the built-in voltage, so a bigger load buys signal and gives up range. A transimpedance amplifier holds the diode at a fixed voltage and stays linear much further.
Breakdown voltage
$\displaystyle M \approx \frac{1}{1-\left(\frac{V}{V_{\mathrm{BR}}}\right)^n}$
The reverse voltage where avalanche multiplication makes the current climb steeply. Datasheets state it at a set reverse current, and the rated maximum bias sits well below it. Some gain creeps in before breakdown, which shows as an upturn at the top of a dark current curve.

Services available

Component makers bring me a new part to model and document. OEMs bring me the light level, wavelength and speed they need, and I find a part and design the front end around it.

Selection and design-in

A shortlist of parts that meet your wavelength, area, speed and noise targets
Bias, load or TIA values for each, with the calculation file
A side-by-side table your engineers can check

Request a quote

Device and front-end design

Active area, intrinsic-layer thickness and the capacitance each choice costs
Transimpedance amplifier design: feedback values, op-amp choice, stability and noise
Bias and bandwidth budget across temperature

Request a quote

Characterization plans

Test conditions for responsivity, dark current, capacitance, shunt resistance, speed and noise
Fixtures, instruments and sample sizes
Data templates and the script that reduces the measurements

Request a quote

Datasheets from measured data

Spec tables with conditions, typical values and limits taken from your measurements
Curves plotted from your data, with the script that draws them
Modeled and measured values kept apart and labeled

Request a quote Datasheet service

Application notes

Biasing, front-end and noise notes for the engineers who design your part in
Worked examples with figures drawn from code

Request a quote Application notes

Geometry, package and window

Custom active-area shapes and sizes, with the capacitance and speed each one costs
Package choice, and window or filter material with its transmission and angle effects
STEP models and drawings through the CAD service

Request a quote 3D models and drawings

Deliverables

A report or datasheet draft in Word and PDF, in your template or a clean layout of my own
MATLAB, Python or Excel files that recompute every number
SPICE models and KiCad files when the job includes the circuit or the board
Figures drawn from code and your data
Review rounds agreed before work starts

Common questions

Do you work from our measured data?

Yes. Datasheet values and curves come from your measurements, with the test conditions on every line. Anything modeled instead of measured is labeled that way.

Can you help before the part is final?

Yes, and that is when device and front-end trade-offs are cheapest to change. I'm happy to sign an NDA first.

Which photodiodes do you cover?

Silicon and InGaAs PIN and PN photodiodes, from the visible to about 1.7 µm, as single elements or in custom shapes.

Do you lay out the board as well?

Yes. I draw the schematic and PCB in KiCad and can hand over a STEP model of the assembly. You keep all the source files.

Discuss your photodiode project

Tell me about the part and what it has to do. I'll reply with a scope and a fixed price.

Discuss a photodiode project Get a free content review

Photodiode design and characterization for component makers and OEMs

Interactive photodiode datasheet

Photodiode design process

Photodiode device model

TIA and readout circuit

MATLAB and SPICE simulation

Verification against requirements

KiCad schematic and PCB layout

3D model

Datasheet and app note

Transimpedance amplifier design

Photodiode pulse response

Photodiode specifications

Responsivity and quantum efficiency

Dark current

Junction capacitance

Shunt resistance

Bandwidth and rise time

NEP and noise

Linearity and saturation

Breakdown voltage

Services available

Selection and design-in

Device and front-end design

Characterization plans

Datasheets from measured data

Application notes

Geometry, package and window

Deliverables

Common questions

Do you work from our measured data?

Can you help before the part is final?

Which photodiodes do you cover?

Do you lay out the board as well?

Discuss your photodiode project