It is expensive to manufacture a GaN laser, and its peak efficiency is not that impressive. So why is this device, rather than the LED, being touted as the future of solid-state lighting? RICHARD STEVENSON investigates.

Combining the light from four lasers emitting at different wavelengths can produce a white light source. Very high efficiencies are promised, but are not possible today due to the low efficiency of sources emitting in the green and red. Credit: Randy Montoya.

If I asked you what you wanted from the bulb of tomorrow, incredibly high efficiency would probably top your list. Not far behind would be a low purchase price and a long lifetime, and you might also value a product delivering directional light and a manufacturing process that treads lightly on the environment.

One bulb ticking quite a few of these boxes is that based on the LED. This form of lighting is the leader in the efficiency stakes, it is free of mercury, and it excels in reliability, lasting 25 times as long as an incandescent and more than twice that of a compact fluorescent. But this bulb is not fault-free. Its light is challenging to direct, its retail price puts many off investing in solid-state technology, and its efficiency is not head-and-shoulders above that of some sources.

The high price of the LED bulb and an efficacy that is good, rather than great, are both consequences of LED droop − the mysterious malady that causes a decline in efficiency at higher drive currents. If droop did not exist, bulbs could operate at more than 200 lumens-per-Watt while being driven from a handful of tiny LED chips running at incredibly high drive currents. Such bulbs would be cheap to make, not just because of the minimal amount of semiconductor material in them, but because high LED efficiencies simplify heat sinking.

Unfortunately, it’s hard to see this vision of cheap, ultra-efficient LED bulbs becoming a reality. Droop is an intrinsic weakness of the LED, and while it is possible to push out its impact to higher current densities, it seems that it will always be sapping efficiencies at really high drive currents. So, what is needed is an alternative way forward, possibly a device that provides a similar level of reliability to the LED, but doesn’t suffer from droop. And, in an ideal world, it delivers directional light.

Come to think of it, doesn’t this device already exist? Researchers Jonathan Wierer and Jeffrey Tsao certainly think so − they are even arguing that laser diodes are the logical choice for the future of lighting. Along with Dmitry Sizov from Corning, they have recently published a paper in Laser & Photonics Reviewsdetailing calculations that show the tremendous promise of this form of lighting.

Using lasers for lighting will raise a few eyebrows: After all, aren’t they too expensive? Aren’t they too inefficient? And how can these monochromatic sources produce a high-quality white light source? Well, as we are about to see, such concerns are not showstoppers: Lasers actually have the potential to be cheap enough and sufficiently efficient to produce very affordable, high quality lighting.

How many lasers?

The scientists from Sandia have been considering novel approaches to lighting for many years. Back in 2007, their calculations revealed that efficacies of 408 lm/W, combined with a colour-rendering index of more than 90, are possible by combining four narrowband sources. These calculations were for line-widths of 1 nm and emission at 463 nm, 530 nm, 573 nm and 614 nm.

Unfortunately, building an efficient lighting system based on this approach is a long way off. Although it has been possible for some time to deliver a reasonably efficient output from a narrowband-emitting blue laser, efficiency plummets at longer wavelengths, similar to the LED. This weakness, known as the green gap, occurs because, as more indium is added to the InGaN quantum well to push its emission to longer wavelengths, two unwanted effects occur: material quality diminishes; and there is an increase in the strength of the internal electric fields in the LED, which pull apart electrons and holes and impair efficient radiative recombination.

The answer is not to turn to the AlGaInP material system, even for the red source. Although this quaternary enables the manufacture of efficient lasers at around 650 nm − the spectral region used for DVD players and recorders − efficiency rapidly falls at the shorter wavelengths necessary for solid-state lighting, due to plummeting carrier confinement in the quantum wells. So if efficient green, yellow and red lasers are to exist, progress needs to be made with III-nitride materials. This should happen, but it will take time, so Tsao and Wierer are proposing an interim step: pumping a phosphor with a blue laser.

Constructing a white-light source in this manner is not a new idea. Back in 1962, Nick Holonyak, co-inventor of the laser diode, speculated on the development of the laser as a practical light source. Fast-forward to 2007, and at the Photonics West meeting in San Jose Nichia’s engineers demonstrated such a product, a white light source formed by coupling a GaN laser into a fibre that had a phosphor coated to the other end. And, more recently, BMW announced that it is developing headlights based on GaN lasers, which can deliver directional beams that would be hard to create with LEDs.

The discouraging news is that Nichia no longer markets the fibre laser white-light product. Tsao argues that it does not follow, however, that this implies that laser-based lighting is fundamentally flawed: “I think Nichia was just a little bit early. Back then lasers were not as efficient as they are now, and even now they are not efficient enough.”

Today, blue-emitting lasers have a power conversion efficiency of up to 30 percent, while state-of-the-art LEDs can hit 70 percent. So it appears that lasers are well behind at the moment. But these figures don’t tell the full story. A fundamental weakness of the LED is that its peak efficiency occurs at a very low current density, typically around 1 Acm-2. If LED light bulbs were to operate in that regime numerous chips would be required to produce enough light, and total chip costs would be exorbitant.  Thus, the current density through the devices that are deployed commercially is cranked up to ten-to-twenty times this value, where droop kicks in.

The origin of droop

Although the cause of this efficiency-sapping malady is highly controversial, Wierer is adamant that it is the result of Auger recombination. “[Auger] has been measured in five or six different ways, and it seems like the values that people are measuring for that recombination are converging on a certain number.” Alternative theories for droop, such as electron leakage and carrier leakage, just don’t cut it with Wierer: “Every time I read a paper that suggests a mechanism that is not Auger, I can shoot holes in it.”

However, even if it turns out that Wierer and all those in the Auger camp are wrong about the cause of droop, it would not alter the key message from this study – that lasers, rather than LEDs, are the most promising devices for solid-state lighting. That’s because whatever causes droop drives down efficiency at very high current densities and ultimately allows lasers to operate more efficiently in this regime.

Although lasers are impaired by Auger recombination, carrier-density-related parasitic recombination losses are ‘clamped’ once the current density is high enough to induce lasing.  That’s because, beyond threshold, additional current density is determined not by the clamped radiative (due to spontaneous emission) and non-radiative recombination. Instead, it depends on recombination due to stimulated emission and an increase in the density of cavity photons.

Nonetheless, by assigning droop to Auger processes, a set of self-consistent equations can be employed to describe the behaviour of both classes of device. “I’ve always been a big believer that you can look at LEDs and learn something about lasers, and vice-versa,” says Wierer. “The devices are related, because some of their operations are similar. The laser diode before threshold is just a spontaneous emitter, so it has got to behave like an LED, even though it is in a cavity.”

Although lasers are not plagued by droop, they are impaired by another significant loss mechanism at high drive currents – resistive heating, which is proportional to both resistance and the square of the current. “[This loss] is normally more severe than it is in LEDs because of geometry– the laser diode normally has a smaller area, so you are trying to force current through a smaller area,” says Wierer.

Predicting the future

To gauge how profound an impact the laser can make in lighting, it’s important to not just consider the performance that the laser and the LED deliver today, but what they will be capable of tomorrow. Wierer and his co-workers have tried to anticipate similar degrees of improvement to both classes of device by using a set of self-consistent equations and considering probable refinements to today’s state-of-the-art emitters.

For the LED, the baseline device chosen by the team was a Philips-Lumileds Luxeon Rebel royal-blue LED. This high-performance, thin-film chip features a silver mirror contact to the p-region and a roughened n-type surface that enables an extraction efficiency of 80 percent. The active region of this device is unknown, so the theorists took an ‘educated guess’ – three quantum wells with a thickness of 2.5 nm.

Calculations for an ‘improved’ LED indicate what might be possible in the future. With the improved LED, Auger loss is trimmed by reducing carrier densities in a non-polar LED

with 20 wells that exhibits one-quarter of the resistance and one-tenth of the mirror loss of today’s device.

Calculations for the efficiency of the standard laser are based on a design with the same active region as the standard LED. Characterisation of this laser indicates an internal loss of 6 cm-1, a modal gain coefficient of 23.9 cm-1 and an inhomogeneous line broadening of 30 meV. Meanwhile, the ‘fully improved’ laser is non-polar; its mirror loss and internal loss are ten times and four times lower, respectively; the inhomogeneous line width is just 20 meV; and optical confinement is four times higher. 

Plotting power conversion efficiency for the improved LED and laser as a function of input power density shows that, at really high input powers, the laser is more efficient (see Figure 1a). However, if the performance of the LED improves while that of the laser stands still, this gap could shrink fast.

Figure 1. (a) The power conversion efficiency of LEDs peaks at a far lower current density than that of a laser. Today, the peak efficiency of an LED is much higher than that of a laser, but if lasers improve, this performance gap will shrink. (b) The higher the current density in lasers and LEDs, the smaller the chip must be to prevent overheating. (c)The heat-sink limited flux is determined by the current density running through the device, its power conversion efficiency and its maximum size. (d) The maximum acceptable manufacturing cost for a laser chip for solid-state lighting is far higher than that for an LED, thanks to the far higher current density of operation

Thermal limitations

One significant downside of driving devices harder is that it leads to greater chip heating. In a light bulb, the temperature of the light-emitting chip cannot exceed a certain value, so this places an upper limit on the input power density.

Wierer and his co-theorists have considered the case where chips are attached to a simple (not ‘super-expensive) heat sink, the maximum temperature rise is limited to 55 °C and the ambient temperature is 25 °C. Modelling shows that the dimensions of the LED can be orders of magnitude larger than that of the laser at peak efficiency (see Figure 1b).

Armed with details of the maximum size of the device, plus values for the input power density and the power conversion efficiency, the team went on to calculate the maximum light output at peak efficiency from the chip.

This is just 130 lumens for a state of-the-art laser, but 4200 lumens for an LED (see Figure 1c). “[With an LED], you are driving it at lower powers compared to the laser, so you are getting much less light per square centimetre – but you can make the chip much bigger, so you win,” says Wierer.

However, he is quick to point out that you don’t win by that much. “It’s not like the laser is so tiny you can’t get any light out of it.” This difference in maximum lumen output will decrease as devices improve. With an LED, the improvements in device design will lead to higher power density (at peak efficiency), leading to a modest decrease in heat-sink-limited chip area

and a modest decrease in maximum output (or heat-sink-limited flux) to 4600 lumens. But for the laser, the heat-sink-limited chip flux will increase, spurring white light output to 1050 lumens at peak efficiency.

Although the LED can deliver more light than the laser, that finding on its own could lead to erroneous conclusions regarding the future of solid-state lighting. For what really matters is this: The acceptable chip cost for economical solid-state lighting. And judged against that metric, the laser is the clear winner, because it has a much higher allowable areal cost than the LED (see Figure 1d).

“Because you are getting so much light per square centimetre [with a laser], you can make the chip really expensive,” explains Tsao. In his view, cheap solid-state lighting is far more feasible with a laser than with really low-cost LEDs made on silicon. “GaN-on-silicon was touted to be the thing because it is so cheap, but it’s not that cheap on a per-lumen-output basis.”

Although no one is currently making laser-based light bulbs, Soraa of California has made what Wierer describes as “the next logical step” by producing GaN LEDs on native substrates and driving them at higher current densities. “ [Soraa’s] LEDs operate at around 400 nm, not in the blue, because droop is less there”, says Weirer, adding that this allows them to run at higher current densities.

Where are we today?

When the scientists from Sandia calculated the acceptable chip-cost-per-unit-area, they included a figure for the ratio between the capital cost of light and its operating cost. With established forms of lighting, this ratio is one-sixth, and that is the value that the theorists have employed.“There is nothing magic about it, and maybe it will not be one-sixth in future, but one-sixth is already pretty cheap,” says Tsao.

How do today’s solid-state lamps compare to that one-to-six ratio? Well, the Cree 60 W bulb that puts out 800 lumens and retails for just under $13 is not that far away. It draws 9.5 W, and if it were used for 20,000 hours, it would run up an electricity bill of $19. That means that if its retail price could fall to just over $3, it would hit the traditional ratio.

Reducing the price tag to this level will require trimming the costs of all the bulb’s components. Opportunities for cost reductions vary. There is more potential to slash costs for chips than there is for more mature parts. Given this state of affairs, the scientists have assumed that when LEDs make a big impact in lighting, they will account for just 10 percent of total bulb costs – a far lower proportion than they do today. Although it seems that LED bulbs can be competitive, laser-based variants promises to be even more attractive, because they can not only cut chip costs, but also boost efficiency. Does that mean that the consumer would actually be willing to pay more for a laser-based bulb, rather than an LED-based one, due to the energy savings? Ironically, no, at least in the long term, explains Tsao: “You are going to be willing to pay less, because you want the purchase price to be smaller than the price of the fuel.” So, a switch to laser based lighting should lead to a fall in the price of the bulb, which is actually what one would expect, based on the one-to-six ratio.

This situation sounds great for the general public. However, it is not necessarily going to happen, and it will hinge on a substantial increase in the efficiency of the laser, so that it can displace the LED within the bulb. “And it’s not just efficiency for electricity consumption,” says Tsao,“but efficiency that is so high that it makes heat-sinking easier. That is a big potential factor in how to get your package costs down.”

If this happens, the producers of lasers that can hit these high efficiencies should see rocketing sales, as will suppliers of GaN substrates. But not everyone in the III-V industry will be a winner. “This would be a disaster for equipment manufacturers,” admits Tsao. “There might be some yield problems at the beginning, but you will not need very many tools.”


Above: Jonathan Wierer and Jeffrey Tsao from Sandia National Laboratories have modelled the performance of lasers and LEDs and determined that the former holds more promise for low-cost, solid-state lighting. (Credit: Mike Coltrin)

Dmitry Sizov from Corning is the third member of the team that is revealing the promise of laser-based solid-state lighting