The effectiveness of laser diode bar stacks in many applications, such as pumping of solid-state lasers and direct use in material processing, is limited by the poor beam quality associated with current commercial stacks. In this paper, we report the development of a technique for improving beam quality by characterising beam imperfections of a stack, and based on that information, fabricating a custom optical corrective plate that removes the effects of a series of opto-mechanical errors, which were introduced during the original fabrication of the stacks. Diode bars are typically fabricated with 25 emitters of 200 μm width along a 10 mm bar of semiconductor, which is then solder-bonded to a micro-channel water-cooled heat sink. Commercially produced units, emitting 50–100W per bar, are stacked on typically 1.8 mm pitch to build up a total laser power of 500–1000 W. For many applications, an individual fast-axis collimator (FAC) lens is attached for each bar at the manufacturing stage. Currently, plano-acylindrical lenses are used to provide low aberration collimation for the high numerical aperture fast-axis beam. Brauch et al. [1] provide a good review of the optical issues of bar stacks and their applications.

The poor effective beam quality of diode bar stacks is a consequence of the practical imperfections in manufacturing and mechanical mounting for the bars and FACs at the micron level, rather than the intrinsic properties of the beams from individual lasers emitters in a bar. The cylindrical lens for each bar introduces angular aberrations that produce a local radiance loss of a factor of 2 to 3, although this may be improved with modern lens technology. The “smile” effect, whereby the semiconductor bar is bent by differential expansion during solder bonding, prevents the collimation lens being correctly positioned for all points along the bar, resulting in beams with variable pointing direction. Also, errors in attaching the FAC to the heat sink with the required positional accuracy further degrade the angular spectrum of the emitted light. There is also the problem of stacking the individual diode bars accurately enough to keep a common radiation direction for all bars. In many applications of the laser diode stacks, subsequent aperture filling, beam shaping and beam combining optics are required. The spread into a much larger volume of phase space than the theoretical minimum, mainly through errors in ray angles from the FAC, increases the design difficulty and reduces the efficiency for the subsequent beam conditioning optics, and fails to transfer the average radiance at the facet to the output application end.

Once a diode stack is built, the beam errors are fixed for a given average power output and operating temperature. The fixed nature of the errors allows for correction if a suitable optical element can be made, placed close to the output plane where beam errors are predominantly angular. To demonstrate this, we report the use of a commercial diode bar stack made by OptoPower Corp., which consists of 11 bars vertically stacked with a 1.82 mm bar pitch, emitting an overall power of 440W cw. Each bar is fitted by the manufacturer with a plano-acylindrical FAC which we measured to be 1.5 mm aperture and 0.9 mm effective-focal-length, similar to lenses available from several manufacturers.

Using the set-up in Fig. 1, we have assessed the severity of the various sources of beam error for each bar with the stack operating at full power. A slit formed by two mirrors, which are mounted on a micrometer stage (not shown), selects a single bar at a time, allowing the beam to propagate towards the screen at the left whilst deflecting the rest of the beams onto beam dumps. A cylindrical lens projects a slow-axis image of the selected bar on to a diffusing screen located 2 metres away. In the fast-axis direction, the beam is a good approximation to the far-field at this distance. A CCD camera connected to a computer is used to record the image from the screen.

 

Fig. 1. Experimental setup used to measure beam properties of individual bars.

Download Full Size | PDF

Figure 2(a) shows images for each bar in the stack for the condition as delivered by the manufacturer and Fig. 2(b) show images for the same bars after a correction has been applied, which is explained later. Each beam image has dots corresponding to the slow-axis imaging of the 24 broad area emitters on the bar. In the fast-axis, the bar images are all significantly different due to differing combinations of opto-mechanical errors. The fast-axis M² factor measured for the collective beam from each bar ranges from 5.5 to 15.5, with a mean value of 10.4 for the 11 bars.

 

Fig. 2. Slow-axis image, fast-axis far-field at 2 metres for each diode bar in the stack. (a) As delivered by manufacturer. (b) After a correction has been applied.

Download Full Size | PDF

The varying beam shapes result from different combinations of errors for each bar. The lens attachment errors – defocus, transverse offset and twist – lead to beam pointing and beam size variations and image rotation. Imperfections in the fabrication of the cylindrical collimator lens or poor aberration correction in the lens design lead to break up into side-lobes or fringes, as clearly seen in Fig. 2(a) for bars 1, 2, 5, 9, 10 and 11. The departure of the semiconductor substrate from flatness, or microlens bending, gives rise to the bent far-field known as “smile”. As a result, the intensity distribution is modified from the expected gaussian-like profile and straight line image in the beam shapes shown in Fig. 2(a), with corresponding M² values much larger than expected for the near diffraction-limited beams available directly at the emitting facet.

 

Fig. 3. Corrective surface for bar 1. Calculated from the wavefront data.

Download Full Size | PDF

 

Fig. 4. Example of the fast-axis wavefront correction for one emitter in bar 1.

Download Full Size | PDF

Since the beam errors are both pointing errors and local aberrations, we have used a wavefront sensing method based on the Hartmann technique [2] to measure the fast-axis wavefront variation at a plane close to the FAC output surface. This measurement was performed using a sensing device built in-house, which has a spatial resolution of 40 μm, an angular sensitivity of 19 μrad and a dynamic range of ~2600. By using this method, we obtain the local wavefront tilt, which is then numerically integrated as explained in Ref. [2] to obtain the Optical Path Difference (OPD) in the fast-axis. With this information we calculate the surface depth required for a refractive correction surface fabricated in silica, which will eliminate the OPD across the fast-axis. This OPD correction is applied where significant laser intensity is detectable. The required corrective surface for bar 1 is shown in Fig. 3 as an example. This surface shape data is translated by computer into cutting instructions for a specialised, high performance CO₂ laser machining/polishing workstation. As described in Ref. [3], the optical surfaces are fabricated by raster scanned, pulsed CO₂ laser ablation [4] followed by laser polishing [5], using a 50μm spot diameter and 10μm raster pitch. Since publishing [4], we have significantly enhanced the machining process by developing a complex control system to keep the CO₂ laser stabilised on a single line and to control the dispensed energy in each laser pulse. As a result, we are now capable of fabricating smooth surfaces of arbitrary shape up to 40 μm deep with a depth resolution of 100 nm. After laser polishing [5], the measured loss outside a scatter cone angle of 10mrad is 0.5% at 940nm. A similar optic fabrication process, but on a much smaller scale, has been reported [6] to correct a single laser emitter, using excimer laser ablation of a polymer surface on a collimator lens.

The laser micro-machining produces 11 correction zones, one for each of the bars, on a single substrate. Each correction zone is a generally smoothly-varying surface over an area of 11 mm × 1.6 mm, with a surface height range of typically 10 to 15 μm. The prescription of the zones is designed to ensure all the bars have beams aligned along a common axis. The corrective plate is aligned using a micro-positioning stage at the plane of the wavefront measurement, and can be fixed satisfactorily using UV-cured glue. Relative misalignment of the stack and correction plate of ±5 μm in the fast-axis, ±15μm in the slow axis and ±250 μm in the proximity to the measurement plane are observed to produce noticeable degradation in the quality of the correction. It should be noted that such tolerances are far more relaxed than for the FAC lenses.

The effectiveness of the correction process for the whole stack is evident from Fig. 2(b), which shows the correction of the bar images in Fig. 2(a). The bars images are now straight, the side lobes are effectively removed and the central lobe is far brighter. Figure 4 shows an example of the uncorrected and corrected wavefront for one emitter in bar 1. The wavefront OPD range has been reduced from ~4λ to <λ/2 by the correction at 940nm. The correction is effected on single emitters by reducing aberrations and on bars by eliminating the smile. The fast-axis beam profile for a single emitter is shown in Fig. 5(a). The removal of the errors has transferred light from the tails into the main beam yielding a narrower beam with twice the peak intensity. Results for a whole bar demonstrate both aberration correction at emitter level and removal of overall bar smile. Figure 5(b) shows the transformation of the beam profile for bar 6 from a broad distribution with a step into a single-lobe with less than 3 times the divergence of the uncorrected beam and about 3 times the peak intensity.

 

Fig. 5. Fast-axis intensity profiles of uncorrected and corrected beams, taken after propagation of 2 metres. (a) Profile of a single emitter from bar 6. (b) Bar-averaged profile of bar 6.

Download Full Size | PDF

In terms of the M² factor, the bars are improved by more than a factor of 5. For example the beam in Fig. 5(b), which had an uncorrected M²=13.5 is now improved to 2.5. In terms of the smile effect (departure from straightness) the worst affected (bar 6) has been reduced from 3.3 μm to <0.4μm. The residual smile effect for all bars in the stack is found to be ≤ 0.4μm, while the bar average pointing errors have been reduced to 0.7 mrad rms.

As a figure of merit for the achieved correction we present the Power-In-The-Bucket (PITB) curves for the uncorrected and corrected version for one bar in Fig. 6(a) and for the whole stack in Fig. 6(b). For one bar, the angle that contains 80% of the power has been reduced by 4.3 times and for the whole stack it was reduced by 3.2 times.

 

Fig. 6. Power in the bucket curves for (a) Bar 6 uncorrected and corrected. (b) The combined beams from the whole stack uncorrected and corrected.

Download Full Size | PDF

A significant part of the improvement comes from the near-elimination of the smile effect, which was once considered not compensatable [1]. In the past, limited smile compensation has been reported using mechanical deflection of the cylindrical lens [7] or rotation of a separate slow-axis cylindrical lens after fast-axis collimation [8], but neither is very practical for large diode stacks. We believe that the correction plate technique solves a problem that has existed since the first development of diode bars and which manufacturers still struggle to eliminate.

The diode stack beam correction reported in this paper was achieved with an OptoPower diode stack, bought in 1998, which we believe exhibits most of the problems that could be present, making it a good test stack to demonstrate this technique. Since then, FAC lenses with near-diffraction-limited performance have appeared in the market; also manufacturers of stacks have greatly improved the alignment accuracy of the FAC lenses. However, new stacks may still suffer from significant smile effect, an error that degrades the high beam quality once available at the output facet of the bars. We have observed this in a range of newer commercial stacks purchased in the last few years, which we have also corrected.

Conclusion

In conclusion, we report a correction process for laser diode stacks that greatly reduces smile, lens aberrations and beam pointing error, to the point where these no longer limit optical design for beam transfer and brightness delivered to an application. We believe that this technique can significantly increase the brightness of the current generation of commercial stacks. Corrected stacks will enable improved beam handling and combining, such as beam interleaving, wavelength combining and beam reformatting.