Pulsed versus Continuous-Wave LIV Measurement: Self-Heating and Methodology

Scope

This article describes the choice between pulsed and continuous-wave (CW) modes for semiconductor laser LIV measurement, with focus on the self-heating bias introduced by CW operation and its impact on extracted parameters. Coverage includes duty cycle requirements, equipment configuration for pulsed measurements, and the quantitative magnitude of self-heating across common device classes. Pulsed measurements for high-speed characterization (small-signal modulation response, eye-diagram tests) are outside scope.

The physical issue

A semiconductor laser dissipates a fraction of its input electrical power as heat. The dissipated power is $P_{\text{diss}}=IV-P_{\text{out}}$, where $IV$ is the input electrical power and $P_{\text{out}}$ is the emitted optical power. For typical edge-emitter operation at ${\eta }_{\text{WPE}}\sim 30\mathrm{\%}$, approximately 70% of the input electrical power becomes heat.

This dissipated heat raises the active region temperature above the heatsink temperature according to the thermal impedance $R_{\text{th}}$ of the device-and-mount assembly:

$$\mathrm{\Delta }{T}_{\text{junction}}=R_{\text{th}}\cdot P_{\text{diss}}\mathrm{.}$$

For a typical telecom DFB on a CuW submount with active TEC cooling, $R_{\text{th}}\sim 30-60$ K/W. For a high-power 980 nm pump diode on a copper heatsink, $R_{\text{th}}\sim 5-15$ K/W. For an unmounted edge-emitter die clamped between thermal contacts, $R_{\text{th}}$ may exceed 100 K/W.

The consequences for LIV measurement are systematic. As drive current is swept upward during a CW LIV measurement, dissipated power increases, the active region temperature rises, and the LIV curve responds to the elevated temperature rather than to the heatsink setpoint. Three observable effects result:

1. Threshold current increases with the junction temperature. The threshold itself shifts only by the heating at threshold, which is small, but a line fitted where the slope is falling (effect 2) is tilted, so its intercept can move either way; in the worked comparison below, fits that extend to higher currents move it down
2. Slope efficiency decreases along the sweep, biasing extracted ${\eta }_s$ low and producing apparent curvature in the LIV
3. Optical output rolls over at high current, producing the characteristic "thermal rollover" shape

Pulsed measurement at low duty cycle largely removes the steady-state temperature rise and comes close to the device's isothermal behavior.

Quantitative impact on extracted parameters

For a typical 1550 nm DFB operating at 60 mA, 1.4 V forward, and 10 mW optical output:

$$\begin{array}{rl}{\displaystyle P_{\text{diss}}}& {\displaystyle =IV-P_{\text{out}}}\\ {\displaystyle }& {\displaystyle =(0.060)(1.4)-0.010=74\text{ mW}\mathrm{.}}\end{array}$$

With $R_{\text{th}}=50$ K/W:

$$\mathrm{\Delta }{T}_{\text{junction}}=50\cdot 0.074=3.7\text{ K}\mathrm{.}$$

For a typical $T_0=60$ K, even the full 3.7 K would raise the threshold current by

$$\begin{array}{rl}{\displaystyle \frac{I_{\text{th}}(T+\mathrm{\Delta }T)}{I_{\text{th}}(T)}}& {\displaystyle =\exp (\mathrm{\Delta }T\mathrm{/}{T}_0)}\\ {\displaystyle }& {\displaystyle =\exp (3.7\mathrm{/}60)=1.064,}\end{array}$$

a 6.4% increase, but the threshold sees only the junction rise at threshold: a threshold near 8 mA at about 1 V dissipates about 8 mW, a rise of 0.4 K and an increase of 0.7%. In a $T_0$ extraction that rise grows with the heatsink temperature, because $I_{\text{th}}$ does, which tilts the $\ln I_{\text{th}}$ versus $T$ line and lowers the extracted $T_0$ by about 1% (Extracting T₀ from LIV measurements). The extracted $T_0$ is nearly unaffected by self-heating.

For a high-power pump diode operating at 1 A, 1.8 V forward, and 700 mW optical output:

$$P_{\text{diss}}=1.0\cdot 1.8-0.7=1.1\text{ W}\mathrm{.}$$

With $R_{\text{th}}=8$ K/W:

$$\mathrm{\Delta }{T}_{\text{junction}}=8\cdot 1.1=8.8\text{ K}\mathrm{.}$$

Self-heating becomes significant. For a $T_0=130$ K InGaAs pump diode, the full 8.8 K would raise $I_{\text{th}}$ by $\exp (8.8\mathrm{/}130)-1=7\mathrm{\%}$, and the threshold itself sees only the smaller rise at threshold dissipation. More critically, the elevated temperature reduces slope efficiency, so the LIV exhibits visible concave-down curvature beyond $\sim 0.5$ A. The extracted ${\eta }_s$ then depends strongly on the fitting range.

The takeaway: self-heating bias scales with $R_{\text{th}}\cdot P_{\text{diss}}$. For low-power devices ($P_{\text{diss}}\lesssim 100$ mW) with good thermal mounting, CW measurement is acceptable for most extracted parameters. For high-power devices ($P_{\text{diss}}\gtrsim 500$ mW), pulsed measurement is required for accurate slope efficiency, $T_0$, and absolute threshold values.

Pulsed measurement principle

In pulsed mode, the drive current is applied as a train of short rectangular pulses of duration ${\tau }_p$ (pulse width) and repetition period $T_r$ (or equivalently repetition rate $f=1\mathrm{/}{T}_r$). Optical output is measured only during the pulses.

The duty cycle is

$$\text{DC}={\tau }_p\mathrm{/}{T}_r={\tau }_p\cdot f\mathrm{.}$$

During each pulse the active region heats up, and between pulses it cools back toward the heatsink temperature. Heat leaves the active region through the chip, the submount and the heatsink in turn, and each stage responds on its own time scale: in InP the thermal diffusion length is about 7 µm after 1 µs, and heat takes a few hundred microseconds to cross a 100 µm-thick chip, while the submount and heatsink respond over milliseconds. The temperature rise during a pulse is fastest at the start and slows as the heat spreads into larger volumes, so it does not grow in proportion to the pulse width. A 1 µs pulse sees only part of the CW rise, and how large a part depends on the device and its mount, so it is checked by measurement rather than assumed: the extracted parameters should not change when the pulse width is halved or doubled (see Validation).

Between pulses the device cools. Averaged over the repetition period, the temperature rise is about the duty cycle times the CW rise, so at a duty cycle of 0.1% the heat left over from earlier pulses is negligible. A common configuration uses ${\tau }_p=1\mu$s and $f=1$ kHz, giving DC = 0.1%.

Equipment

Pulsed LIV measurement requires equipment substantially different from a CW setup.

Microsecond pulses need a dedicated pulsed laser-diode current source or a pulse generator with a current driver; check its minimum pulse width and rise time against the pulse the measurement needs. Pulsed measurements at currents above $\sim 1$ A typically require dedicated high-current pulse generators.

The detector must have bandwidth sufficient to resolve the pulse shape. For a 1 μs pulse, the detector bandwidth should be $\ge 10$ MHz to faithfully reproduce the pulse top. Slower detectors integrate over rising and falling edges and introduce systematic errors in extracted optical power per pulse.

Procedure

1. Configure the pulse parameters

Set the pulse width and repetition rate to give DC $\le 0.1\mathrm{\%}$. For most semiconductor lasers, ${\tau }_p=1\mu$s and $f=1$ kHz are standard. Higher repetition rates allow faster sweeps but increase DC; lower rates reduce noise but extend the measurement duration.

Verify that the chosen pulse width exceeds the device turn-on time (typically nanoseconds to tens of nanoseconds) and is long enough to allow the optical output to reach its quasi-steady state during the pulse. A scope trace of the optical output should show a flat-top pulse with no visible rise or droop during the integration window.

2. Set the sampling window

Configure the boxcar or oscilloscope to sample the flat-top region of the optical pulse, excluding the leading and trailing edges. A typical sampling window is 60% of the pulse width, centered on the pulse.

3. Verify zero-bias background

With the laser turned off (or biased below threshold), verify that the optical signal level is at or below the detector noise floor. Stray light from room illumination, photodetector dark current, or pulsed pickup from the current driver can contaminate the measurement.

4. Sweep the current

Step the pulse amplitude through the desired current range, recording the integrated optical power per pulse and the voltage during the pulse at each step. The resulting LIV represents the device behavior at the heatsink temperature, largely free of CW self-heating bias.

5. Compare to CW (optional)

A useful verification step is to repeat the LIV sweep in CW mode at the same heatsink temperature. The CW curve will show a slightly higher actual threshold (although a line fitted over a wide current range can place it lower, as in the worked comparison below), lower slope efficiency, and eventual thermal rollover; the pulsed curve will be cleaner and extend higher in current before any rollover (since the rollover in pulsed mode is due to intrinsic carrier and barrier effects rather than self-heating).

When to use which mode

Use CW mode when:

• The device is low-power and well-thermally-mounted ($\mathrm{\Delta }{T}_{\text{CW}}<2$ K at the maximum operating current)
• The application of interest is itself CW (datacenter optical modules, telecom transmitters, sensing applications)
• The measurement is for in-process screening rather than parameter extraction
• Pulsed equipment is not available

Use pulsed mode when:

• The device is high-power ($P_{\text{diss}}>500$ mW at the maximum operating current)
• The device is unmounted (bare die on a clip, edge-coupled to fiber on a probe station)
• Accurate slope efficiency is required at currents well above threshold
• $T_0$ extraction is performed at high current operating points
• Thermal rollover is being characterized as a function of mount quality, and the underlying isothermal LIV must be separated from the thermal effect
• The application is itself pulsed (high-energy pulsed lasers, gain-switched sources, mode-locked oscillators)

Worked comparison

The comparison below is computed for the 1310 nm Fabry–Pérot laser used in the other LIV articles, at a 25 °C heatsink. The pulsed curve (1 μs, 1 kHz, DC = 0.1%) is taken as isothermal, with a threshold of 8.0 mA and a slope efficiency of 0.55 W/A. The CW curve adds self-heating, with $R_{\text{th}}=50$ K/W, within the range given above for a mounted telecom laser, a forward voltage of $0.95\text{ V}+5\mathrm{\Omega }\times I$, and the temperature dependence of the T₀ extraction article: $T_0=59.6$ K for the threshold, and $T_1=143$ K for the slope efficiency, fitted to the same data. At each current the junction temperature and the output power are solved together.

The CW curve falls below the pulsed one by 1–3% up to 75 mA and by 7% at 150 mA, where the junction runs 9 K above the heatsink. The deficit at 10 mA is larger than at 30 mA because near threshold a small shift in threshold is a large fraction of the output: at threshold the junction is already 0.4 K above the heatsink, which moves the CW threshold to 8.05 mA.

Extracted parameters, from fits at 1 mA steps:

The CW fits understate the slope efficiency by 1.8% over 10–50 mA and by 3.9% over 10–100 mA. They also place the threshold lower than the pulsed fit does, although the actual CW threshold, 8.05 mA, is higher: the slope falls along the sweep as the junction warms, so the fitted line is tilted and its intercept moves to lower current, by more the further the fit extends. A CW fit confined to 10–20 mA gives 8.04 mA, close to the actual CW threshold. For this low-power device on a good mount the differences are small, as the rule of thumb above predicts; for higher-power devices they are substantially larger.

Sources of error specific to pulsed measurement

Detector bandwidth too low. A detector with insufficient bandwidth integrates over the rising and falling edges of the pulse, producing measured power lower than the true flat-top power. Verify on a scope trace that the detected pulse shape is rectangular with $\ge 60\mathrm{\%}$ flat-top duration.

Pulse-to-pulse jitter in current amplitude. Inexpensive pulse drivers may have $\sim 1\mathrm{\%}$ amplitude jitter between pulses, producing apparent noise in the LIV. Averaging over many pulses (typical: 100–1000 pulses per current step) suppresses this.

Ground-loop pickup in the boxcar gate. The gate signal for boxcar integration shares a return path with the current pulse; ground-loop coupling can produce spurious counts at the boxcar output that are not from optical signal. Verify that the boxcar reading equals the noise floor when the current pulse is below threshold.

Inductive pulse distortion. Long cables between the pulse driver and the laser introduce inductance that distorts the pulse shape. The current pulse at the device may have a different shape than the programmed pulse. Verify pulse shape with a current probe at the device leads.

Insufficient pulse-off time for thermal recovery. If the pulses come too close together, heat from earlier pulses accumulates and the average junction temperature rises above the heatsink. Keep the duty cycle at or below 0.1%, so the average rise is at most about 0.1% of the CW rise, and confirm that halving the repetition rate does not change the result.

Edge-resolved detection of relaxation oscillations. For pulse widths approaching the relaxation oscillation period of the laser (typically 0.1 to 0.3 ns for telecom DFBs, which relax at 3 to 10 GHz), the leading edge of the optical pulse exhibits relaxation oscillations that bias the integrated power measurement. Use pulse widths $\gtrsim 100{\tau }_{\text{RO}}$ (tens of nanoseconds) to ensure the integration window is in the steady-state region.

Validation

The pulsed LIV at low DC ($\le 0.1\mathrm{\%}$) should be independent of pulse width and repetition rate within stated tolerances. Halve and double ${\tau }_p$ (0.5 and 2 μs around 1 μs) and verify that extracted parameters do not change by more than the measurement uncertainty.

For devices with verified low thermal impedance and low operating power, the pulsed and CW LIVs should agree closely at low currents, to within a few percent for the device in the worked comparison. Significant disagreement in the low-power regime indicates either pulse-shape distortion or detector calibration error.

References

For the textbook treatment of device thermal impedance and self-heating effects in semiconductor lasers, see Piprek (2003), Semiconductor Optoelectronic Devices: Introduction to Physics and Simulation, chapter 7. For thermal impedance measurement methodology and reference values for common device packages, see the JEDEC JESD51 series of standards.