3.2 Solids and Nanostructures: Electrons, Spins, and Phonons
Project coordinator(s): C. von Korff Schmising, M. Wörner

T3: Optoelectronic devices


High-power semiconductor lasers are the most efficient man-made light sources, and can convert more than 80% electric energy into light. Currently emission powers of one kW continuous-wave powers are extracted from a single monolithic semiconductor array. Such operation, however, is connected with very high internal power densities of more than 1010 W/cm3. We are interested in the intrinsic limitations of such optoelectronic devices in terms of output power and beam quality (brightness). For this purpose we analyze devices, but also their components such as surfaces and interfaces or gain materials such as quantum wells, superlattices and quantum dots.
For our experiments, we use optical tools, in particular transient spectroscopy that represents a generic competence of MBI. Such work is naturally carried out as collaborative work with device vendors, who provide us with high-quality industry-grade devices and structures. In BMBF-projects such as BlauLas, we work together with Osram OS (Regensburg), Dilas GmbH (Maiz) and Laserline GmbH (Mülheim) or in the frame of bilateral research contracts with Lumentum (Santa Clara) and 3S-Photonics (Nozay).
Beginning in 2014, the material basis of the investigated devices underwent a transition from infrared-emitting GaAs-based devices towards GaN-based wide-bandgap devices emitting in the ultraviolet to green spectral regions. This new field involves a lot of new challenges and opportunities and we are looking forward this work.

Limits of high-power laser operation in GaN-based diode lasers

GaN-based diode lasers are main sources of light in the ultraviolet and blue spectral regions. Intense research and developing work following their invention by Nakamura has resulted in increasing emission powers without, however, having met the benchmarks in terms of absolute power set by GaAs-based high-power diode lasers. This may be one of the reasons, why the catastrophic optical damage (COD) has not received the same attention as for GaAs-based diode lasers. Moreover, GaN is a harder material than GaAs and its surfaces withstand higher power densities. Special facet coating technologies, waveguide architectures, and window structures have been developed for GaN devices, which make them less susceptible to COD.
Why has COD in GaN-devices nonetheless grown increasingly important? COD is a generic sudden degradation mechanism, which originates from the interplay of localized light absorption and temperature rise eventually resulting in a positive feedback loop. The energy for this process is exclusively provided by light, and the process can start at any absorbing site. This includes the facets of diode lasers, internal defects within the waveguide, as well as absorbing sites in passive waveguides. In view of the advent of further increasing emission powers, COD will become an issue in GaN-based devices as well. Thus, it is important to clarify to what extent the details of the process are different from GaAs-based devices.
We present first results on nanosecond kinetics of COD events in GaN-based devices. The COD was artificially provoked in a single-pulse step test scheme. This method has been introduced for infrared devices because of two motivations: First it is relevant to determine device limits in terms of maximum pulsed single mode16 or absolute operation powers independently on the repetition rate. Second, it turned out for GaAs-based broad area lasers that single-pulse testing allows for provoking the COD even in such devices, where the COD is definitely not the primary degradation mechanism. Here, we analyze the COD behavior of a batch of standard GaN-based devices. Moreover, time- and spatially-resolved emission monitoring as well as time-integrated thermal imaging allows for detailed analysis of the COD process on a nanosecond timescale. COD in GaN-based devices is (i) a ‘hot’ process, (ii) can be re-ignited in subsequent pulses, and (iii) can lead to pronounced material ejection out of the front facet. The latter observation makes the COD process in GaN-based devices unique compared to infrared lasers.
The 12 R&D test devices analyzed in this study (named A-L) stem from the same batch. The epitaxial structure is based on InGaN multi quantum well active layers, GaN waveguides and AlGaN cladding layers. The chips are 1.2 mm in length, the emitter stripe width is 15 µm. Laser threshold and slope efficiency are of the order of 0.2 A and 1.6 W/A. We investigate chips which are not optimized for high power levels to enable studies of facet stabilities. In contrast, optimized devices enable continuous-wave (cw) power levels of several Watt. The chips are mounted by hard solder in a p-side up configuration as shown in Fig. 1 (a).

Figure 1 (a) Schematic diagram of the experimental setup including a photomicrograph of a packaged chip. (b,c) Scanning electron micrographs of the front facets of devices B and C, which experienced COD in 5 pulses of 4.6 A and one single pulse of 5.2 A, respectively. The total operation times of the devices B and C after the beginning of the COD process (‘burning times’) were 3.6 µs (in 5 pulses) and 730 ns (in 1 pulse), respectively.

We use single current pulses with durations of ~800 ns starting at 200 mA. For each successive pulse in the step test, the current is increased by 350-400 mA until the devices fail. The time between the single pulses is on the order of minutes resulting in full thermal decoupling. Figure 1 (a) depicts the main layout of the experimental setup. The emission near field of the device (photomicrograph of a packaged chip in center) is imaged to the entrance slit of a Hamamatsu C5680 streak camera equipped with the M5677 single sweep module. The sweep time of the streak camera was 1 µs. The temporal resolution is better than 10 ns. A dichroic mirror tilted by 45° directs the emission to a fast Si-photodiode (ThorLabs DET10A/M, <5 ns rise time), which provides spatially integrated information about the momentary emission power. A thermocamera Thermosensorik CMT384 is used to image either the front facet (in this case replacing the streak camera) or the side of the device, i.e., along the cavity (z-axis). The spectral sensitivity is confined to the 3.5-6.0 µm wavelength range. Primary emission from the device (spontaneous emission and scattered laser light) is monitored along the z-axis by a Si-CCD camera, Basler Ace acA1300-30um with an integration time of 16 µs; see CCD image in Fig. 1 (a). Imaging optics and filters are not included in the scheme of Fig. 1 (a). Figures 1 (b,c) show typical damage pattern at front facets of devices B and C as measured by scanning electron microscopy in secondary electron mode.

Figure 2 Current pulses (a) and photodiode traces (b) representing the emission power as measured during the test with device A. Degradation took place after 700 ns within the 13th pulse; see thick red line. (c) Light-current characteristic for single-pulse operation of device A. (d) Primary emission (spontaneous emission and lasing) along z-axis as extracted from cuts through emission images; see dashed line in the CCD image (side view) in Fig. 1 (a). The degradation event is indicated by arrows in all subfigures.

Figure 3 (a) Streak-camera images as monitored during the first 15 pulses of device A. An arrow at the 13th pulse marks the COD event. (b,c) Difference images highlighting the changes from pulse # 11 to 12 (b) and # 12 to13 (c).

Figure 2 shows a data set from device A from a step test including current (a) and emitted laser pulses (b). Figure 2 (c) shows the light-current characteristic in single pulse operation, while (d) gives the primary emission along z-axis during all current pulses. This data is extracted from defined cuts through the CCD-camera images [cf. dashed line in the CCD image (side view) in Fig.1 (a)]. Figures 2 (b-d) point to sudden degradation in the 13th pulse (4.7 A) at t=700 ns. The current pulses in (a) do not reflect this event. The increased level of scattered emission from the region 0<z<200 µm in (d) even prefigures the degradation site at the front facet. In successive pulses (after the initial degradation event in the 13th pulse) the entire cavity 0<z< 1200 µm shows an increased level of scattered primary emission, most probably spontaneous emission.
Figure 3 (a) shows the time evolution of the emission near field of device A as taken by the streak camera during 15 pulses. This data confirms the spatially averaged information from Fig. 2 (b), in particular about the sudden degradation close to the end of the 13th pulse; see arrow. Details of the spatial filamentation are visible in Fig. 3 (b) and (c), showing the differences of successive images. While (b) shows the situation just before the degradation event, (c) is the difference between the streak images taken in the 13th and 12th pulse. Sudden degradation sets in at x~1 µm and t=600 ns.

Figure 4 Selected thermocamera images taken during two separate step tests in front (a-d) and side view (e-g) geometry. (a-d) Thermocamera images as taken from device B in the 11th to 14th pulse at current amplitudes of 4.2, 4.6, 4.6, and 4.6 A, respectively. (e-g) Thermocamera images as taken from device A in the12th to 14th pulses at current amplitudes of 4.3, 4.3, and 4.7 A, respectively. In (e-g) the same color code as in (a-d) has been used with a total range from 8-29 counts only. The dimension is given by the chips cavity length of 1.2 mm with the bright bond wires on top. The material ejection in the 14th pulse is highlighted by a yellow ellipse.

Figures 4 (a-g) present two sequences of thermocamera images as taken from the front facet (device B) and along the z-axis (device A). Each panel is compiled of 3 thermal images taken one after the other. The emissivity contrast represents the geometry (gray scale) taken with 1 ms integration time. The overlayed color code points to the Planck’s thermal radiation (blackbody radiation) and is generated as the difference between the image taken during the current pulse and a reference image (without current), both recorded with 10 µs integration time20 to reduce the influence of long-term drifts and pixel-nonuniformity. Pixels with values below the upper noise level (i.e. <7 counts at a noise level in the -7 to 7 count range) are made transparent. We should note that the noise-equivalent temperature in this particular regime (for camera view towards the front facet) amounts to ~400 K as caused by spatial and temporal averaging.18,21 The image sequence clearly shows a step increase in thermal radiation when proceeding from 4.2 to 4.6 A (from <8 to ~60 counts) This allows for a rough temperature estimate of ~2000 °C and shows clearly that the observed sudden degradation is a hot process, which likely involves local melting processes.22 Moreover, the following images (c) and (d) taken with 4.6 A current pulses, demonstrate that the degradation process restarts. Here, the initial degradation site serves as the starting point for the continuation of the process and the positive feedback loop gets closed rather via macroscopic defect generation (extrinsic loop) than by thermal bandgap shrinkage (intrinsic loop).7 The thermocamera image sequence from device A, taken in side view, reveals matching details, namely a step increase in thermal radiation from <8 to ~30 counts when increasing the current amplitude from 4.3 to 4.7 A. Here, the side imaging geometry reduces the total thermal signal by a factor of ~2. It allows, however, to observe another rather unexpected effect, namely a beam of material being ejected from the front facet; see Fig. 4 (g).

The data presented so far allow for identifying the observed sudden degradation at 4.6, 4.7 A and 5.2 A (devices B, A, and C) with the COD, as well know from GaAs-based devices. This assignment is based on the
• appearance of damage pattern at the front facet,
• emission power decay kinetics,
• near field dynamics, and
• the step temperature increase.
Therefore, we consider the current values above the individual COD thresholds of the particular devices. The COD thresholds of all investigated devices range between 4.5 and 5.5 A with two lapses at ~3 A. The observed rather narrow spread of individual COD thresholds, suggests that we indeed address a relevant degradation mechanism of the batch. We define a value of ~5 A as the COD threshold of the batch. In terms of emission power and power density per aperture, this threshold corresponds to ~7 W and 460 mW/µm, respectively. The latter figure excels those known for GaAs-based devices in continuous wave (cw) operation,23 but needs to be compared to values determined under similar pulsed operation conditions. For 980 nm emitting GaAs-based devices, values beyond 1000 mW/µm have been observed.24 Nevertheless, the present technology is suitable to make the GaN-based devices highly resistive against COD during cw operation, the operation mode these specific chips are designed for.
An exponential fit of the power decay of device A in the 13th pulse in the 700-800 ns range in Fig. 2 (b) provides a 1/e-decay time of 130 ns. For device C, we find almost the same figure, while for device B a value 90 ns is obtained. Consequently, a value of 100 ns at 450 mW/µm is representative for the batch. In an earlier study of AlGaAs-based devices for the 650-1000 nm emission range, we identified the power decay time to be governed by heat removal from the COD site, which involves the quantum well and vicinal partitions of the waveguide. This process is controlled by the thermal resistance of the waveguide material, which is the lowest towards the GaAs (long wavelength) side and larger for AlAs-richer materials (short wavelength), at least for AlAs mole fractions smaller than 0.5.26 The data for the actual GaN-based batch fit well into the GaAs-side. This agrees with the fact that GaN has a lower thermal resistance than GaAs or AlGaAs of any composition.
Comparing Figs. 2 (b) and 3 (c), there is an apparent inconsistency regarding COD kinetics. While the integrated power starts to decay at about 700 ns after the beginning of the pulse, there is an obvious power loss at the COD site (x=1 µm) at 600 ns already. Fig. 3 (c) shows, parallel to the power drop at 600 ns at x=1 µm, a power increase in the 5 < x < 8 µm range, i.e., within the first 100 ns after the beginning of the COD process there is no (external) power loss, but only a redistribution of power between the filaments. A related behavior has been observed for 980 nm emitting broad-area lasers,28 and resulted in the creation of a second COD-starting site at the location of the newly formed filament. Figure 3 (b), the snapshot before COD, reveals another feature of COD kinetics. Precisely at the position of the initial COD site, there is a distinct filament. This could be either incidental or a result of a local heating process already attracting light by thermal lensing. In the latter case, this heating would have predetermined the initial COD site in the 12th pulse, even before the actual ‘degradation process’ in terms of power loss. All in all, the filamentation kinetics in our GaN-based devices at ultrahigh powers is very similar to what is known from GaAs-based broad-area devices.
In contrast, the material ejection from device A in the 14th pulse [Fig. 4 (g)] is a completely unexpected event. The presence of a ‘beam’ of heated material, however, qualitatively confirms the estimate of a very high temperature during COD. Taking into account the length of the beam [Fig. 4 (g)] and the (remaining) integration time of the camera, we find a velocity of the ejected material of not smaller 1000 µm/6 µs~170 m/s, corresponding to about half of the velocity of sound (in air). The beam-like geometry of the material ejection is also rather unexpected. Its formation in the 14th pulse, i.e., one pulse after the initial COD pulse becomes plausible, when assuming that the initial pulse created a small macroscopic defect (~µm) near the front facet only. The following pulse probably causes a melting process in the interior, starting at the interface between this defect and the healthy inside of the cavity. Therefore, the by now recrystallized initial defect might have acted like a plug, which becomes released with a certain delay, which is caused by the duration of the melting process through this defect towards the facet. The beam-like geometry of the ejected material reflects probably the flow of molten material, e.g., Gallium, from the interior.
The damage patterns being created within 5 successive pulses [see Fig. 1 (b)] and a single pulse [see Fig. 1 (c)] look quite similar. We conclude that the facet modification occurs during the initial pulse only, while further pulses rather cause melting processes in the interior. The observed ‘holes’ point to effective material loss, and have been observed in GaN-based devices during cw operation as well.5 Holes are a sign of material loss, in line with the material ejection phenomenon reported above. In addition, the typical facet damage during the step test in this device batch involves split lines and even a delamination of the emitter stripe in the region close to the facet. Whether the latter observation is typical for single pulse COD in GaN-based devices in general, or for the investigated batch, or for the p-up packaging only needs clarification in future experiments.

In summary, COD in 450 nm emitting InGaN/GaN diode lasers has been investigated. The spatio-temporal evolution of the near field of the emitter stripe, the spontaneous emission, as well as the thermal kinetics are monitored. COD in this type of devices is a hot process taking place at ~2000°C. In a few (1-4) subsequent pulses this high temperature is reached again. Within the first 100 ns after the onset of the COD process, the filamentation undergoes serious changes resulting in additional local intensity maxima in the near field. Thereafter, a decay of emission power takes place on a 100 ns time scale (1/e-decay) marking the ultimate power degradation process of the devices. The outer appearance of the damage pattern at the front facet is predominantly created within the initial COD pulse. Moreover, the COD process can involve material ejection out of the front facet.