91av

Imagine packing 140 lasers onto a single piece of semiconductor measuring
about 2 millimetres by 7 millimetres. Imagine that each laser emits light
at a different wavelength, making the surface a bright array of different-coloured
spots. Imagine, moreover, that the lasers are made so precisely that each
has a wavelength 0.3 nanometres longer than that of its neighbour on one
side, and 0.3 nanometres shorter than that of its neighbour on the other.
Such an array would have been a mere flight of fancy just a few years ago,
but not any more. Communications specialists can now demonstrate one that
does nearly all these things – the only exception being that the light it
produces is in the infrared range and so is not visible.

Dramatic advances are being made in laser technology, just as electronics
was making giant strides a generation ago. Once again, the latest developments
exploit improvements in semiconductor technology. To pack a single chip
with so many lasers, researchers are now taking advantage of manufacturing
processes, such as molecular beam epitaxy and ion beam implantation, that
can create layers only a few atoms thick and were developed to produce complex
integrated circuits. These techniques also enable single lasers to emit
more light at shorter wavelengths than before, giving technologists finer
tools to record, transmit and read more information.

The developers of the laser array see their work helping to improve
the performance of high-speed optical communications. Because the lasers
will operate at many different wavelengths, the array will be able to transmit
data at impressive rates through optical fibres or through free space between
computer boards or chips. Similar arrays could be used in optical computers
that will process data as beams of light instead of pulses of electric current.
Meanwhile, single lasers emitting light at visible (shorter) wavelengths
could improve the performance of laser printers by enabling the machines
to write faster, and increase the density of data storage on optical disks.
But the greatest possibilities may be still to come, sparked off by the
new capabilities of the emerging technology.

The earliest electronic devices relied on thermionic valves, and the
next generation on individual transistors. Today, thousands or even millions
of transistors are crammed onto the integrated circuits that are carried
on a single semiconductor chip. Laser technology is still in its early stage.
Gas lasers, the counterpart of valves, are still widely used, even when
only milliwatts of light are required. Though engineers prefer solid-state
technology, almost for its own sake, the older technology often remains
cheaper.

The most familiar gas laser is the helium-neon type, a tube about the
size of a rolled-up magazine that emits red light – a wavelength that has
only recently become available from semiconductor lasers. The counterpart
of the single transistor is the semiconductor laser with just one light-emitting
element on a chip of gallium arsenide. It was invented nearly 30 years ago
in the US and is now the most common laser, usually buried deep in the bowels
of equipment like fibre-optic communication networks, laser printers and
CD players. But it operates only at low powers in the infrared range. Larger
lasers exist only in research laboratories and industrial plants; they consist
of gas-filled tubes powered by an electric discharge, or crystalline rods
powered by bursts of light from powerful flash lamps.

The latest semiconductor lasers may make much of the older technology
obsolete – just as solid-state devices took over from transistors and low-power
valves – but they will not replace high-power lasers. For the time being,
semiconductor lasers are being seriously considered only for applications
requiring no more than about 10 watts of steady light. Semiconductor electronics
developed the same way, starting out at low powers and working up to higher
levels.

Like integrated electronics, arrays of many lasers on a single substrate
have important advantages. In some arrays each laser can be turned on and
off independently. At Bell Communications Research (Bellcore) in Red Bank,
New Jersey, scientists can do this up to 5 billion times a second in one
two-by-eight array, to generate a staggering 80 billion pulses per second.
Other arrays combine the outputs of many laser elements to generate a single
beam of higher power, which in some cases can be steered electronically,
with no moving parts. Researchers have barely started to explore the possibilities.

New ways of making small, thin structures of semiconductor materials
are improving the efficiency of single lasers. The smaller structures confine
the flow of current and light in the semiconductor more precisely. This
makes the devices more efficient at converting electrical energy into light,
and they last longer, too. These are critical improvements because, until
recently, the lifetime and output powers of some semiconductor lasers have
been disappointing, especially for devices generating short-wavelength light.

The choice of materials is crucial. Many semiconducting materials, notably
silicon, normally do not emit light. The first generation of commercial
lasers was made of gallium arsenide, which emits light only in the invisible
infrared region. Only in the mid-1980s did semiconductor lasers producing
red light go on sale, and they have acquired important uses in displays
and bar-code scanners. They are made from indium gallium phosphide and yield
less power than gallium arsenide lasers. In July 1991, the 3M company, an
American maker of plastics and electronics, announced that it had developed
the first semiconductor laser to emit green light, which has a shorter wavelength
than red and infrared light, and so can be focused to a much smaller spot.
This is an important develpment because it allows more data to be stored.

The latest semiconductor lasers rely on the same phenomenon as the pioneering
devices developed in the early 1960s. The semiconductor must contain two
regions ‘doped’ with different types of impurities. In one region, the impurities
add extra electrons to the crystal; in the other, they create electron vacancies,
or ‘holes’. Applying a voltage across the material, so that the negative
terminal is attached to the material with extra electrons, known as n-type,
and the positive terminal to the material with holes, known as p-type, causes
electrons and holes to move towards the junction of the two regions. Some
electrons fill holes at the junction, causing current to flow. Because these
devices have two electrical terminals, they are known as ‘diodes’.

The union of an electron and a hole releases the energy that had kept
the electron free to move about the crystal lattice. In many traditional
semiconductors, such as silicon and germanium, virtually all this energy
is released in the form of heat; with the newer generation of semiconductor
compounds, such as gallium arsenide and indium phosphide, some of the energy
takes the form of light. This is the basis of both light-emitting diodes
(LEDs), which are often used as indicator lights or displays, and semiconductor
diode lasers.

LEDs emit light in all directions as the electrons release energy spontaneously
at the junction. Lasers are more complex. The edges of the semiconductor
wafer are made to reflect some light back along the junction plane, stimulating
other electrons to release light energy at the same wavelength. The process
amplifies the intensity of the light and also directs it along the line
between these two mirror edges (one or both of which let some light escape).
In practice, semiconductors operate as lasers only when the current passing
through the junction exceeds a threshold value; at lower currents, they
operate as LEDs.

The first semiconductor lasers, which were made of two thick layers
of gallium arsenide with different dopings, produced so much more heat than
light that they burned out unless cooled to well below room temperature.
The key to improving them has been to confine more of the current and the
reflected light in the junction plane (see Figure 1a). Engineers found they
could do this by sandwiching the junction plane between two thin layers
in which aluminium replaced some of the gallium (see Figure 1b). Using the
same principle, they could confine light to a narrow strip within the junction
plane (see Figure 1c).

Tighter confinement of current flow and light improves efficiency by
concentrating energy. The higher the current, the more electrons are available
to emit light. The more light is concentrated, the more efficiently it extracts
energy from the electrons. By 1980, this technology had produced inexpensive
commercial lasers that could operate continuously for many thousands of
hours. The layers were 0.1 to 1 micrometres thick, and the strips in the
junction plane 1 to 10 micrometres wide. Then developers began to investigate
what might happen if structures were made even smaller.

THE SHRINKING CHIP SOLUTION

A micrometre is tiny on a human scale but large on an atomic one; a
1-micrometre layer is a ‘bulk’ material, in which electrons can have a broad
range of energies. That ceases to be the case when dimensions shrink below
about 0.02 micrometres (20 nanometres), or about 35 atomic layers in gallium
arsenide. In such minute structures, the laws of quantum mechanics limit
electrons and holes to certain energy levels. This changes the material’s
properties, which depend on electron energies. The development of molecular
beam epitaxy, which deposits atoms layer by layer, opened the possibility
of making such thin layers.

The first step was to make a ‘quantum well’, a layer typically several
nanometres thick, sandwiched between two thicker layers of a slightly different
composition. The compositions are chosen so that electrons that carry current
have slightly less energy in the quantum well layer than in the barrier
layers that sandwich it. Electrons become trapped in the quantum well if
they lack the energy to return to the barrier layers. The barrier layers
must be thicker than the quantum wells to keep the electrons from escaping
by the quantum-mechanical phenomenon of tunnelling, which lets electrons
through thin regions into which they would otherwise have too little energy
to pass.

The quantum well is formed at the junction between n-type and p-type
semiconductor materials where it can trap electrons and holes, which can
be stimulated to release laser light. Stacks of quantum wells can be made
by alternating quantum well and barrier layers. This increases the number
of trapped electrons and holes, without reducing their density. Constance
Chang-Hasnain’s team at Bellcore used three quantum wells in the 140-element
laser arrays described at the beginning of this article. The barrier layers
were gallium arsenide, while indium was substituted for 20 per cent of the
gallium in the quantum wells, giving a compound with composition In0.2Ga0.8As.
Because quantum wells confine electrons and holes more closely than bulk
semiconductors, they make more efficient lasers. They are already used in
some commercial semiconductor lasers.

Quantum wells confine electrons in only one dimension in a thin layer.
The next logical step is quantum wires, which confine electrons in two dimensions
– in a narrow strip that is part of a thin layer (see Figure 2). Developers
hope that the tighter confinement in quantum wire structures will improve
laser performance, further reducing the threshold for laser action and the
waste energy that must be dissipated as heat. They even speak of the potential
for ‘quantum dots’, which would confine electrons in three dimensions –
length, width and height.

The tighter confinement should reduce the current needed to reach the
threshold for laser action. Quantum well lasers have threshold currents
of about a milliampere, but thresholds could be about a microampere for
quantum wire lasers and about a nanoampere for quantum dots, says Eli Kapon
of Bellcore, a pioneer in the field. Though the smaller lasers are likely
to be more efficient, their individual outputs will be low. The quantum
well elements in the Bellcore array emit only about 2 milliwatts of light,
while the output of quantum wire elements will be of the order of microwatts.
Kapon says ‘that’s fine for the kind of applications people are envisioning’,
such as closely packed arrays of many tiny lasers, each one emitting only
a little light, for optical computing or communications.

The first quantum wire lasers were made two years ago, but so far the
lowest threshold, obtained by Kapon’s group, is only 0.6 milliamperes. The
major problem is that thin layers are much easier to make than the thin
lines needed for the second dimension of confinement. Using present technology,
layers a few nanometres thick can be formed fairly easily. ‘There is no
equivalent fabrication technique that will give you automatically quantum
wires and dots,’ says Kapon. Quantum confinement requires lines no more
than 20 nanometres wide. Though a scanning tunnelling microscope can work
at this scale, it cannot do so easily enough for mass production. No one
has yet reported operating a quantum dot laser.

Theorists predicted quantum wells, but they were surprised by another
property of very thin semiconductor layers – that they tolerate much more
internal strain than thicker layers. Researchers discovered this when they
were trying to make new types of semiconductor lasers. For a long time,
strain was believed to be bad for lasers: ‘Now it turns out that it seems
to be making everything better and better, as long as you don’t have too
much,’ says Anders Olsson, head of the solid-state and quantum optics research
department at AT&T Bell Laboratories in Murray Hill, New Jersey.

The initial impetus for making strained layers came from problems in
growing the semiconductors used in lasers. The thin laser layers are deposited
on a thicker substrate of either gallium arsenide or indium phosphide. (More
complex compounds of three or four elements are too difficult to grow in
large wafers.) Good-quality layers can be formed only if they have atomic
spacing close to that of the substrate. If the atomic spacings differ too
much, the strain causes the layers to develop flaws – imperfections in the
crystal where atoms are missing or extra atoms are inserted – that are likely
to make the laser fail. Atomic spacing depends on composition, so this limits
the choice of compounds for semiconductor lasers.

For layers a micrometre thick, the restriction is quite stringent –
atomic spacing must be within 0.1 per cent. Fortunately, substituting aluminium
for more than half of the gallium in gallium arsenide changes atomic spacing
by less than that, so lasers of those materials were the first type developed
commercially; they emit light at 0.75 to 0.9 micrometres in the near infrared
range. However, to match a lattice to indium phosphide – for longer-wavelength
lasers – requires the proportions of the four elements indium, gallium,
arsenic and phosphorus, to be balanced. Manufacturers do this to produce
lasers emitting light at 1.3 and 1.55 micrometres, the infrared wavelengths
used in fibre-optic communication systems. These lasers cost much more than
gallium arsenide lasers.

In the past few years, developers have grown indium-gallium phosphide
on gallium arsenide substrates for lasers at 0.67 micrometres in the red
range, and have produced shorter red wavelengths by replacing some of the
gallium with aluminium. However, the power output of these devices was low,
especially at the shorter wavelengths, and 0.63 micrometres appeared to
be a limit for practical lasers. This left shorter wavelengths out of reach,
and a gap at 0.9 to 1.1 micrometres in the infrared.

The lattice-matching constraint can be eased to 1 per cent for layers
no more than about 10 nanometres thick, which can accommodate much more
internal strain than thicker layers. This gives developers a much broader
choice of materials, and hence of wavelengths, as they in turn depend on
the material. For example, thin strained layers of gallium arsenide with
20 per cent of the gallium replaced by indium – a compound with atomic spacing
of 0.570 nanometres – can be grown on gallium arsenide, with 0.566-nanometre
spacing. Bellcore used this approach in its 140-element array, which emits
light at 940 to 983 nanometres, wavelengths not available from purely lattice-matched
lasers.

Developers have also turned to strained layers for visible semiconductor
lasers, which suffer from high threshold currents, limited output power
and short lifetime. At McDonnell Douglas Electronic Systems, based in Elmsford,
New York, lasers with a 7-nanometre strained quantum well of gallium-indium
phosphide produced a steady beam up to a record 475 milliwatts at 665 to
670 nanometres at room temperature. The current density needed to reach
laser threshold was below 375 amperes per square centimetre, the lowest
ever recorded at that wavelength. In October, Spectra Diode Laboratories,
a manufacturer based in San Jose, California, announced it had made a 1-watt
laser at 680 nanometres. Though the company claims a threshold current density
of 350 amperes per square centimetre for the device, the light generated
is at a slightly longer wavelength than that produced by the McDonnell Douglas
device.

At Bellcore, Chang-Hasnain’s group produced pulses of 650 milliwatts
at 634 nanometres from a laser with four strained quantum wells of indium-gallium
phosphide, 2.5 nanometres thick. In between were 4-nanometre barrier layers
that also contained aluminium. But the threshold current density was much
higher, 1700 amperes per square centimetre, and the laser could not produce
the high power in a steady beam. However, the shorter wavelength is better
for optical data storage and recording, and much brighter to the human eye,
making it more attractive for displays.

Strained layers were also used in the most dramatic semiconductor laser
breakthrough of 1991 – operation of the first green diode laser at room
temperature (91av, Technology, 17 August 1991). Several laboratories
had already produced shorter-wavelength blue light by passing the infrared
light from gallium arsenide lasers through materials that double its frequency.
However, that approach is inefficient, as it converts only a small part
of the infrared energy to visible light. A team at the 3M Corporate Research
Center, based at St Paul, Minnesota, produced light with a wavelength of
525 nanometres, the shortest ever from a diode laser, by making the semiconductor
from doped zinc selenide for the first time. The team’s quantum well laser
included strained layers to accommodate differences in atomic spacing within
the material. Developers hope the work will point the way to new semiconductor
lasers emitting visible light, and a new generation of laser applications.

Another innovation marks an even greater change in thinking about semiconductor
lasers than strained layers. For three decades, semiconductor lasers have
been designed to emit light from their edges. That is fine for individual
lasers, or for linear arrays of many laser elements, but something different
is needed if many lasers are to be integrated in a two-dimensional array
on a single chip.

The simplest approach is to redirect the laser light. In a standard
semiconductor laser, the reflective surfaces at the ends of the chip reflect
light back and forth horizontally along the junction plane so that the beam
emerges from the side of the wafer. The light can be redirected by etching
a mirror tilted at 45 degrees to the wafer surface. A more sophisticated
alternative is to etch tiny grooves in the bottom of the junction layer.
These grooves scatter or diffract some light at a steep angle, which can
then pass through the barrier layers. The latter design can be modified
to steer the combined beam from many laser elements over a small angle without
moving parts. This is done by shifting the phases of the separate currents
supplying each of the elements.

Both approaches have been demonstrated, and both have their limitations.
Because they emit light from only a small fraction of the surface, the number
of laser elements that can be squeezed onto a given area of a wafer is limited.
Tilted mirrors are difficult to etch, while the scattering grooves occupy
much more space on the wafer than the laser elements.

A more direct approach is to place the reflective surfaces above and
below the junction layer, so that the laser beam is generated perpendicular
to the junction layer and wafer, instead of along it (see Figure 3). The
beam from such a vertical-cavity laser does not have to be redirected to
emerge from the wafer surface. The emitting areas can be circular, and so
generate a symmetrical beam diverging by only 3 to 6 degrees, instead of
the 10 to 30 degrees of edge-emitting lasers, which require corrective optics.

Vertical-cavity lasers can also be packed more closely on the wafer
than other surface emitters. Typically, each laser element is 10 to 20 micrometres
across, and about 300 micrometres from its neighbours. The laser elements
can be made smaller and pressed much closer if the material between them
is etched away. Two years ago at Bell Laboratories, Jack Jewell etched lasers
as small as 1.5 micrometres across, spaced only a few micrometres apart,
thus packing more than a million lasers onto an area of only 0.5 square
centimetres.

There is a trade-off for this high packing density. The power a laser
can generate depends partly on the volume of material from which it extracts
light energy. If the laser mirrors are placed in the junction plane, as
in an edge-emitter, they can extract light from an area of the junction
a few micrometres wide and a few hundred micrometres long. Mirrors above
and below the junction plane extract light from a much smaller area, typically
a few micrometres to 20 micrometres in diameter. In practice, this means
that the mirrors on vertical-cavity lasers must reflect light much more
efficiently than those on edge emitters.

Powers available from vertical-cavity lasers smaller than about 20 micrometres
will remain low – the highest so far is about 3 milliwatts per laser element.
Larry Coldren of the University of California at Santa Barbara says that
10 milliwatts may be possible eventually, which should be sufficient for
the major expected uses of vertical-cavity lasers, in arrays for high-speed
communications or computing. However, Coldren warns of problems, including
removing waste heat from such a small volume, making current flow easily
and uniformly through the structure, and confining light and electrons for
efficient laser operation.

The vertical-cavity design, such as the 140-element array, also allows
the laser elements on one chip to generate different wavelengths. While
semiconductor composition determines the range of possible wavelengths,
the exact wavelength is set by the distance between the laser mirrors. The
round-trip distance must be an exact multiple of the wavelength, so changing
the spacing changes the wavelength or the multiple. If the mirrors are hundreds
of wavelengths apart – as they are in edge-emitting lasers – the multiple
changes. However, if the mirrors are only a few wavelengths apart, the wavelength
changes predictably (the longer the spacing, the longer the wavelength).
To produce its multi-wavelength array, Bellcore varied the thickness of
one layer diagonally across the wafer, so that it was thickest at one corner
and thinnest at the opposite corner. It was this smooth variation in mirror
spacing that provided the 0.3-nanometre shift in wavelength between adjacent
laser elements.

Advanced semiconductor laser technology will not instantly consign other
lasers to the scrap heap, any more than integrated circuits wiped out valves
and discrete transistors in a day. Semiconductor lasers are far from matching
the highest powers or shortest wavelengths available from other lasers.
What they will do is help to create new applications for lasers, in areas
such as optical computing and communications. They are likely to bring surprises,
and perhaps even new uses for older laser technology, just as integrated
circuits created new roles for electronics.