1. Introduction

Vertical cavity surface emitting lasers (VCSELs) are a type of semi-conductor laser invented by K. Iga and co-workers in 1979 [1]. These devices have several advantages compared to conventional edge emitters. i) They emit perpendicularly to the surface, which makes cost effective batch production possible. ii) The very short optical cavity only supports one longitudinal mode producing a very stable wavelength. iii) Thanks to the small dimensions low threshold currents are obtained, which makes them very energy efficient. Despite these promising assets, it took nearly 20 years before the first commercial devices appeared on the market. Today, applications include multi-mode data communication, computer cluster interconnects and computer navigation devices such as laser mice. As a result, the yearly production has grown to over a 100 million units worldwide [2].

The small diode lasers discussed in this paper are fabricated by epitaxial vapor deposition of semiconductor layers on a GaAs substrate. The structure consists of one or more InGaAs quantum wells incorporated in between high-reflectivity Bragg reflectors consisting of λ/4 layers of AlGaAs with alternating composition and refractive index. In order to produce efficient devices, current confinement to small active areas is created by selective oxidation or proton implantation [2].

Compared to light emitting diodes (LEDs), the reflectors that define the optical cavity create a high power density inside the laser structure, which may result in faster degradation when small defects are present. For this reason, good quality control and detailed knowledge of failure modes is essential. Many different techniques have been employed to analyze failures, such as transmission electron microscopy (TEM), scanning electron microscopy (SEM), X-ray crystallography and chemical analysis [3,4]. For example, electron microscopy can be used to identify small defects in the mirror stack or the oxidized areas. X-ray- or chemical analysis yields information on crystal orientation and chemical composition, respectively. Unfortunately, most of these methods require extensive sample preparation and are therefore time-consuming and destructive.

Modern VCSELs have good lifetime characteristics, which explains why they are the devices of choice in many data communication applications. In order to further improve the reliability and resistance to environmental influences, a better understanding of the degradation mechanism and possible failure modes is needed. Problems related to excessive strain or contamination in the epitaxial layers can be detected using conventional tools as described above. However, as spatial localization of defects at an early stage of degradation is not easy, self-propagation dislocations make it difficult to locate the origin of a problem [5].

The highly accelerated lifetime test (HALT) is a standard method to follow device degradation in a relatively short time frame. In such a test, devices are exposed to high temperatures and high humidity conditions while their performance is checked regularly. For VCSELs the conventional detection of early failure is based on changes in the light output - current - voltage (L/I/V or LIV) characteristics. Despite its obvious value, this does not provide information about the origin of the failure mode in a particular device because LIV changes are normally observed only after a significant degree of degradation.

To solve this problem and to speed up the analysis without losing detail, alternative methods have been developed. Recently it was found that signs of degradation can be obtained by optical spectrum analysis (OSA) before significant changes are visible in the LIV data [6,7]. Analyzing the spectral output of lasers is particularly sensitive as the cavity effect strongly amplifies the influence of damage on the optical output.

The dimensions of most VCSELs produced allow multiple transverse modes to be supported. Very small defects, for example due to strain, can cause the transverse mode pattern to change. Kim et al. [6] and Weidberg et al. [7] showed that there is a clear relation between the spectral output and the device’s health. Both groups found that the width of the emission spectrum was reduced by prolonged exposure to harsh conditions such as high humidity as well as by fast degradation due to electro static discharge (ESD). Unfortunately, the OSA does not provide any spatial information concerning the initial defect location in the devices. In other words, even though it is capable of detecting degradation before it is visible in the LIV data, it is still an integrated signal. That means that a defect needs to be big enough to be detectable in the overall VCSEL output, while localizing the origin of a propagating defect at an early stage is not possible.

To be able to find the location of a problem it is also necessary to map the spectral output in order to visualize the transverse electromagnetic modes. This optical information can then be used to select a specific area for further investigation with, for example, TEM. A mapping of the spectral output can be made using a scanning stage under a conventional microscope with a high-magnification objective. However, the maximum achievable resolution is diffraction limited, which at an emission wavelength of 850 nm precludes looking at details smaller than about 500 nm. In addition, reflections between the objective and the mirror stack cause an overlay of several images and will make the results more difficult to interpret.

One way to circumvent these problems is the use of scanning near-field optical microscopy (SNOM or NSOM). This technique has been used to map the mode profile of VCSELs, but so far this has been limited to showing the actual mode output in pristine devices [8–13]. In most SNOM setups a drawn or etched optical fiber, with a 100-200 nm aperture at the tip, is glued to a tuning fork. This tuning fork is made to oscillate at its resonance frequency near the surface of interest. The feedback mechanism that retains a constant tip-sample distance during scanning relies on the changes in the resonance frequency of the tuning fork. With this approach the optical resolution of the image is limited by the diameter of the aperture and not by the wavelength of the radiation.

Recently, we have demonstrated a new type of near-field probe [14] that is capable of obtaining topography and near-field optical information simultaneously, and has some advantages compared to conventional SNOM systems. Ferrule-top probes are all-optical devices with an integrated optical readout. Therefore, they are easy-to-use and can be mounted in a variety of atomic force microscopy (AFM) systems. In contrast to most SNOM fibers, this probe makes use of a tiny cantilever that holds a tipped fiber at its end for transmitting light to or from a sample. These fiber-based devices are designed in such a way that they can be used at high temperatures and in high humidity conditions, which is required during HALT testing. This makes them very suitable as a tool in failure mode analysis, providing fast detection of emission deviations with high spatial resolution.

In this paper we present a new method, using this technology, that adds high-resolution spatial information to the spectral emission data. This allows one to detect small deviations in the transverse mode structure of the VCSEL output already before complete modes disappear from the spectrum. Conventional analysis techniques such as LIV analysis and far-field spectroscopy will be used in comparison to demonstrate the strength of our new technology. An important feature of the new method is the use of a fast high-resolution spectrometer allowing integration times of only 5 μs. As a result, the resolution and speed of imaging is not limited by the response time of the optical characterization system, but by the maximum achievable AFM speed.

2. Materials and methods

The most important part in the experimental setup is the ferrule-top probe, which is an all-optical device made from boro-silicate glass. The monolithic design and the low expansion coefficient of this material makes the probe insensitive to temperature variations. The fabrication procedure starts with machining a small ridge on the end face of a ferrule with a single bore hole in the center (VitroCom Inc.) (Fig. 1(a)). Then a small glass ribbon (VitroCom inc.) is glued on top of the ridge and is cut to the desired dimensions using a pico-second laser ablation system (Optec System with Lumera Laser source). A metal-coated single-mode fiber with a sharp tip at its core is anchored to the end of the cantilever and the main part of the ferrule as shown in Fig. 1(d). Using focused ion beam (FIB) milling (FEI Helios) a 5 μm cut is made to release the cantilever and to ensure good light transmission from the tip to the photo-detector. Finally a second fiber is inserted into the bore-hole of the ferrule and connected to an interferometer (Optics11 B.V., OP1550) to monitor the cantilever deflection. A detailed description of the fabrication procedure can be found in Ref. [14]. The probe is mounted in a home-built atomic force microscope (Fig. 2) that is placed on an optical table in a temperature controlled lab environment at 20 °C.

 

Fig. 1 Ferrule-top fabrication procedure. (a) boro-silicate ferrule with ridge (3 × 3 × 7 mm³), (b) ribbon glued to the ferrule, (c) ribbon cut into the right dimensions (300 × 2700 × 30 μm³), d) SNOM fiber anchored to the cantilever and cut using focused ion beam (FIB), (e) enlargement of (d), (f) scanning electron microscope image of the end of the cantilever with sharp tip and FIB cut; the insert shows the SNOM tip enlarged. Adapted from Ref. [14].

Download Full Size | PDF

 

Fig. 2 Schematic view of the experimental setup.

Download Full Size | PDF

Light from the VCSEL is collected by the near-field tip of the ferrule-top probe, divided in two beams by a fiber beam splitter (Thorlabs GHB) and directed to a photodiode (90%) and a fast spectrometer (10%). Both are triggered from the AFM software to synchronize topography and spectral data. The spectrometer is equipped with a 1200 lines/mm grating and a Basler spl4096kc camera, resulting in a spectral resolution of better than 0.1 nm. The setup can handle very short integration times, which allows for fast AFM scanning as the emission intensity is not a limiting factor. However, since AFMs are designed for flat surfaces, AFM scanning was hindered by the height of the VCSEL connector ring (>1 μm). To circumvent this issue the scan area was chosen to be as small as possible while enclosing the optical output area (Fig. 3).

 

Fig. 3 Topography image of a VCSEL (12 × 12 μm², 256 × 256 px.). The square represents the area that is scanned during optical mapping.

Download Full Size | PDF

Unless mentioned otherwise, for all experiments we used commercial 850-nm oxide confined GaAs/AlGaAs VCSELs packaged in TO46 cans obtained from Optek Technology through Farnell Inc. These devices have an oxide aperture of about 8 μm and a threshold current (I_th) of approximately 0.8 mA. The advised operating current is 10 mA (20 kA/cm²).

To explore the capabilities of our instrument, after removing the TO46 cap of a VCSEL to make the active area accessible for AFM investigation, the VCSEL was mounted in the experimental setup and optical and topographic images were recorded. The AFM was operated in contact mode with a scan speed of 3 μm/sec.

To show the potential of our technique in failure mode analysis, a HALT test was performed using a high humidity and temperature (relative humidity (RH) = 90%, T = 120^◦C, I = 12 mA) over a period of several weeks to determine changes in the LIV characteristics as a function of time. A drop of more than 5% in light output at a well-defined operating current is considered a failure. With this method, no degradation of the tested devices was observed in the first 7 weeks of the test. During these 7 weeks some of these devices were also tested with far-field microscopy to determine the capability of OSA and that approach showed degradation at an earlier stage. Far-field images of the optical output of the VCSEL were taken using a commercial Renishaw InVia Raman microscope equipped with a 100X Olympus objective (LMPlanfl, NA = 0.8) and a 1800 lines/mm diffraction grating. The stage of this instrument could be scanned laterally in order to make an optical mapping of the spectral output. The smallest achievable step size and integration time were 0.5 μm and 200 ms respectively. Due to the long minimum integration time, mapping using this spectrometer is slow.

To investigate the capability of our method, some VCSELs were also exposed to higher currents (up to 30 mA) for shorter periods of time, while taking measurements at regular intervals. The LIV curve was recorded every 10 minutes to ensure that no indication of degradation was observable yet. With the same intervals, AFM+SNOM images were taken at a current of 10 mA at room temperature to determine whether this approach would show changes in the mode structure at an earlier stage.

For most of our experiments the integration time of the spectrometer was set to 5 μs. When the VCSEL was operated at a low current - below lasing threshold - the integration time was set to 500 μs to obtain enough signal.

3. Results and Discussion

The spatial distribution of the spectral output of a VCSEL can be visualized as shown in Fig. 4 and depends on the operating current. The tested VCSELs are multi-transverse mode (single longitudinal mode) devices at normal operating current that emit a spectrum with a width of several nanometers rather than a single mode. Generating a mapping of the intensity at a particular wavelength provides transverse mode information (e.g. Fig. 5). Figure 4 shows images obtained at different operating currents while Fig. 5 shows mappings of different wavelengths obtained at 10 mA operating current to demonstrate the device behavior. In these images the resolution is insufficient to localize sub-wavelength defects.

 

Fig. 4 Far-field total emission maps of a VCSEL operated at different currents. Images (a), (b) and (c) show the emission integrated over the whole bandwidth at operating currents of 0.1, 1 and 10 mA respectively. Frames (d), (e) and (f) show the corresponding emission spectra at that particular operating current.

Download Full Size | PDF

 

Fig. 5 Far-field emission maps of the mode profiles of a VCSEL operated at 10 mA. The intensity of a particular wavelength representing a transverse emission mode was mapped: (a) 849.50 nm, (b) 849.01 nm, (c) 848.67 nm.

Download Full Size | PDF

Because of the limited lateral resolution of this method, degradation can only be observed when complete modes disappear from the emission spectrum, which can also be observed using OSA without spatial information as shown in Fig. 6. In this experiment the width of the emission spectrum was followed during a 7 week HALT test (RH = 90%, T = 120 °C, I = 12 mA). After 4 weeks of exposure to high humidity and temperature the width of the emission spectrum started to decrease significantly while no sign of degradation was observed in the LIV characteristics (not shown).

 

Fig. 6 Graph showing the decrease in width of the emission spectrum measured at 10 mA during a standard HALT test (RH = 90%, T = 120 °C, I = 12 mA during test); error bars represent the standard deviation (n = 5). ^*The distance between the 20dB (1% of the maximum intensity) values to the left and right of the peak is taken as the spectral width.

Download Full Size | PDF

Far-field mapping of the mode structure of a laser device showed that the narrowing of the emission spectrum is linked to the disappearance of certain modes. However, the onset of such changes is difficult to detect and localize in the far-field; the SNOM approach should be able to shed more light on the early stages of failure.

Using our system spatial information is added, which makes it possible to detect small changes in the mode structure long before complete modes disappear from the spectrum. When operating the VCSEL below laser threshold, the optical feedback does not play a role because there is no cavity amplification. As shown in Fig. 7 circular Laguerre-Gaussian modes can be identified. Far-field methods are not able to resolve these small structures. When operating a VCSEL above lasing threshold, only a few modes, depending on the cavity dimensions and mirror stack, are amplified. The cavity effect disturbs the formation of neatly symmetric transverse modes. In addition, since the fiber tip makes contact with the surface, at currents close to the lasing threshold, reflections may cause some optical feedback and force the output into a particular mode. However, this was found to be limited at currents around 10 mA. Figure 8 shows the transverse modes at an operating current of 10 mA observed using a ferrule-top AFM+SNOM probe. Compared to the far-field images (e.g. Fig. 5) the contrast between the different modes is much bigger. When scans were repeated multiple times at a current of 10 mA, very similar images were obtained. Such reproducibility is very important to be able to identify changes during degradation experiments.

 

Fig. 7 Near-field emission maps of the mode profiles of a VCSEL operated at 0.4 mA (not lasing). Image size: 10 × 10 μm², 256 ×256 px. (a) total emission mapping. Images (b) to (e) represent mappings of a selected wavelength representing a particular transverse emission mode: (b) 848.40 nm, (c) 848.14 nm, (d) 847.74 nm, (e) 847.34 nm, (f) 846.94 nm.

Download Full Size | PDF

 

Fig. 8 Near-field emission maps of the mode profiles of a VCSEL operated at 10 mA, scan area: 10 × 10 μm², 256 × 256 px. (a) total emission mapping. Images (b) to (e) represent mappings of a selected wavelength representing a particular transverse emission mode: (b) 850.73 nm, (c) 850.40 nm, (d) 849.67 nm, (e) 849.44 nm, (f) cross-section at the line in (b).

Download Full Size | PDF

A common method to determine the image resolution is to measure the distance between the 10% and 90% of the maximum intensity points. The line in Fig. 8(f) represents a cross-section at the location of the line in Fig. 8(b). From the steepness of the line it can be concluded that the resolution is better than 200 nm. In a previous paper [14] we showed a similar resolution of ferrule-top optical probes using a standard SNOM test grating.

The unique combination of topographic and optical information makes it possible to differentiate between various types of errors. Physical defects observed at the surface can be identified using analysis at different drive currents. The VCSEL shown in Fig. 9 (different type, I_th = 2.8 mA, aperture = 15 μm) has a small defect (depth ca. 100 nm) at the surface, which was also observed in the optical signal, but only when operated below lasing threshold. It is clear that the emission spectrum at the defective part is different from the rest of the device. When the current was increased to above the lasing threshold, this defect could no longer be observed optically because the feedback is sufficient to cause lasing.

 

Fig. 9 Topography image (a) and optical (c) mapping of a damaged VCSEL operated at 0.1 mA. Graphs (b) and (d) show the emission spectra at the locations indicated by the arrows. The spectral range between the lines was integrated to generate the optical mapping.

Download Full Size | PDF

Damage to the active region induces changes to the mode structure and can be detected using spectral imaging. Excessive current spikes can be the cause for such damage. As a proof-of- principle experiment, repeated SNOM analyses were performed on a VCSEL that was operated at high currents until damage started to occur. After 30 minutes at 15 mA a small defect was observed (Figs. 10(a)) and 10(b)). By increasing the current and exposure time this damage grew bigger (Figs. 10(c)) and 10(d)). During the experiment the VCSEL could stay in the instrument, only the tip was moved out of contact when the current was increased. Measurements were taken at a normal operating current of 10 mA at room temperature. Before every SNOM measurement, the IV characteristic was also recorded but no changes were observed other than the reversible shifts that are linked to the operating temperature of the device. This heat did not influence the all-glass ferrule-top AFM+SNOM measurement. The reason for this is that the probes are made from the low expansion coefficient material boro-silicate glass and that measurements are performed in contact-mode, which is a much more rigid method than the conventional tuning fork tapping-mode or shear force distance control.

 

Fig. 10 Near-field mappings of two emission modes; the arrows point at a location showing increasing laser damage from 0.06 μm² to about 0.5 μm². During the measurement the device was operated at 10 mA. Images (a) and (b) represent mappings recorded after operating the VCSEL 30 min. at 15 mA, images (c) and (d) represent mappings recorded after operating the VCSEL an additional 30 min. at 30 mA. Left images represent the intensity of the 849.67 nm emission; right images represent the intensity of the 849.20 nm emission (size: 10 × 10 μm², 256 × 256 px.). The inset shows an enlargement of the damage (scale bar: 200 nm).

Download Full Size | PDF

In Fig. 10 two different laser modes, representing 849.67 nm and 849.20 nm, are shown next to each other. The arrows indicate where changes start to occur. In Figs. 10(c) and 10(d) the defect has grown from 0.06 μm² to about 0.5 μm². Interestingly, the intensity moves from one mode to the other. If some surface effect or reflection would have disturbed the images one would expect the effect to be visible in all modes. The fact that only certain modes are affected and that the effect is opposite between two modes can indicate internal laser damage. The actual defect is likely to be much smaller but cannot be determined without destructive analysis. More importantly, however, this result shows that very small optical changes can be detected using a ferrule-top AFM+SNOM probe that are not measurable with other methods.

4. Conclusion

For failure mode analysis it is very important to obtain early information on defect propagation. Our results show that ferrule-top technology can be used to obtain faster and more detailed information about device degradation than OSA and LIV analysis. The unique combination of high-resolution spatial and spectral information makes it possible to visualize the transverse electromagnetic laser modes that are very sensitive to device degradation. Ferrule-top AFM+SNOM analysis is i) non-destructive, ii) can be used to investigate an area of 10 × 10 μm² with iii) a resolution better than 200 nm in iv) less than one hour. Because a large area can be investigated in a short amount of time and early failure information is provided, this technique can be used to quickly find the location where damage starts to occur. Other techniques, such as SEM or TEM analysis may then be used to further characterize the affected location. In contrast to conventional SNOM, in our experiment heat generated by the device did not influence the measurement, which would even allow one to use this technique during a standard HALT test. This paper is meant to show the suitability of ferrule-top technology for faster and more efficient failure mode analysis of VCSELs. Actual use in HALT testing is at this point beyond the scope of this paper.