Henini M. Handbook of Self Assembled Semiconductor Nanostructures for Novel devices in Photonics and Electronics

Подождите немного. Документ загружается.

Quantum Dot Superluminescent Diodes 569

ultra-narrow (  1 Å) emission lines corresponding to the recombination of excitons, biex-

citons and other multi-excitonic states. The lines from a large ensemble of QDs (typically

 10

–10

QDs are embedded in the active region of a waveguide) then merge in a broad

(30–100 nm) Gaussian-shaped emission line. At room temperature and/or high excitation

levels typical of device operation, a signiﬁ cant homogeneous broadening (10–20 nm) of

the emission line of each QD also contributes to the total spectral width. While inhomo-

geneous broadening has always been perceived as the major obstacle to achieving high

performance in lasers, it represents a clear advantage for SLED applications. As discussed

thoroughly in section 19.2, the growth conditions can be tuned to increase the QD size

dispersion, and thus the gain spectral width. Additionally, several QD layers can be stacked

with different centre wavelengths ( “ chirped multilayers ” ), resulting in an even wider com-

bined gain spectrum.

2. Limited density of states: The number of electronic states in a QD ensemble is directly

related to their areal density (there are two spin-degenerate states on the lowest ( “ ground ” )

state for each QD). The areal density depends on the growth conditions and is typically lim-

ited to few 10

dots/cm

, particularly due to a limitation on the average strain in the sys-

tem where In-rich QDs are needed to shift the wavelength towards the 1300 nm region. By

stacking multiple layers, this can be increased to the few 10

dots/cm

range. This density

of states (DOS) is roughly a factor of 10 lower than the typical DOS in the ﬁ rst subband

of a single QW in a comparable energy interval. This in turn results in lower maximum

modal gain corresponding to full inversion of the states (in the few 10 s cm

 1

range for

QDs, as compared to 100–200 cm

 1

for a single QW). This lower gain is a major limitation

for lasers, producing gain saturation and consequently poor frequency response. On the

contrary, the relatively low density of states can be an advantage for SLEDs. In fact, to

achieve population inversion over a large energy range at reasonable current levels, a

low density of states per unit energy is preferable. The ground state (GS) of a single InAs/

GaAs QD layer emitting around 1300 nm is typically ﬁ lled with  100–200 A/cm

, and

a gain spectrum  100 nm wide, combining ground and excited states (ES, see right part

of Fig. 19.3 ), is obtained with less than 1 kA/cm

. In QWs, as mentioned above, ﬁ lling all

states in this energy range would require 10 kA/cm

or more. Ultimately, the wide emis-

sion spectrum of QD SLEDs is the consequence of the relatively small density of states per

unit energy. Achieving high-output powers obviously requires a certain level of chip gain

( gL  6–10), which is nowadays achievable by stacking 10–20 layers of QDs.

19.2 QD growth for SLEDs

Large spectral width is one of the most important features of SLEDs [2, 3] , as the correspondingly

short coherence length can signiﬁ cantly improve the spatial resolution in coherence-based sys-

tems. The large spectral width can be realized by using the techniques such as multiple-quantum

well (QW) engineering and QW intermixing [4–7]. However, increasing the SLED spectral width

over 100 nm in the 1300 nm wavelength region is still a challenge. As mentioned in section 19.1,

and originally proposed in [8] , the naturally occurring size dispersion in self-assembled growth

of quantum dots (QDs) can be beneﬁ cial for SLEDs requiring larger spectral width. Generally,

inhomogeneous size distribution of QDs in the active region is disadvantageous for laser applica-

tions. However, for the wide spectra device applications such as SLEDs, it becomes an effective

and intrinsic merit due to the broadened gain spectrum. Size dispersion in the tens of per cent

range results in a spectral width over 100 nm from the QD active region as predicted theoretically.

Therefore, considerable efforts have been invested in developing QD SLEDs [9–22] . Initial reports

employed the uniform stacked In(Ga)As/GaAs QDs which are routinely used in laser devices,

showing a narrow spectral width and/or a short emission wavelength [10–13, 15] . To increase

the spectral width, deliberate increase of the dot size dispersion is a straightforward method

[8, 17, 21] . In addition, employing multiple QD layers with different amounts of InAs deposition

in the QDs is another option [9, 17] . Large photoluminescence (PL) spectral width over 100 nm

can be obtained by using both methods [9, 21] . However, for SLED applications, a large PL

CH019-I046325.indd 569CH019-I046325.indd 569 6/27/2008 5:21:29 PM6/27/2008 5:21:29 PM

570 Handbook of Self Assembled Semiconductor Nanostructures for Novel Devices in Photonics and Electronics

linewidth is not sufﬁ cient, high radiative efﬁ ciency and uniform optical gain must be obtained

across the entire spectral range. For example, varying the InAs coverage also affects the QD den-

sity, producing a non-uniform gain spectrum. As an alternative, stacking InAs-chirped multiple

QD (CMQD) structures with constant InAs coverage and varying In composition in the InGaAs

capping layer was proposed by us [4, 17] and also used by other groups [18–20] . In fact, when

the QDs are covered/surrounded by an InGaAs layer, a red shift of the emission wavelength is

observed, depending on the composition and thickness of the InGaAs layer [23, 24] . Because the

emission wavelength of each QD layer in the chirped structure can be individually controlled by a

change in the InGaAs matrix, the composite emission of the chirped structure will yield a broad

gain spectrum. In such CMQD structures, the QD areal density is not signiﬁ cantly affected by the

InGaAs layer, so that the available gain and the wavelength are effectively decoupled and a ﬂ at

gain spectrum can be obtained. The combined spectral broadening depends on the spectral width

of the individual QD layer, the spectral separation among each QD layer and the number of the

stacked layers. Using the CMQD structure, a composite PL spectral width of 78 nm was obtained

from the ground state (GS) of uniform QDs. Incorporating such a CMQD structure into a SLED

device structure and combining the emission from GS and excited state (ES) of QDs, SLEDs with

spectral width of 121 nm, covering from 1165 to 1286 nm, were demonstrated [14] . In such

structures a dip may appear in the spectrum between GS and ES, due to insufﬁ cient broadening

of the GS and ES lines. The dip can be effectively eliminated by optimizing the chirped stacking

structure and QD growth conditions. In particular, replacing uniform QDs by QDs with increased

size dispersion in the CMQD structure, smooth and broad output spectra up to 115 nm without

dip were obtained from SLEDs [17] . In the following paragraphs, the growth optimization of the

CMQD structure and single-layer InAs QDs with increased size dispersion are described.

19.2.1 Chirped multiple QD structure

High uniform InAs QDs used for laser devices have high peak gain but provide relative small gain

spectral width which limits their applicability in the SLEDs. The CMQD structure was proposed to

take advantage of the optimized gain characteristics of these QDs while increasing the total gain

linewidth in a controllable way [14] . In this part, we describe the growth of the CMQD structure.

The samples are grown on (100) GaAs substrates by using solid-source molecular beam epitaxy

(MBE). InAs QDs form by continuous deposition of InAs material (nominal thickness 3 monolay-

ers) at the growth temperature of 530°C and growth rate of 0.163 μ m/h, which are then cov-

ered by a few nm of thin InGaAs capping layer. In the CMQDs structure, the InAs/InGaAs QD

layers are separated by 40 nm GaAs barriers to avoid as much as possible strain-driven growth

coupling between different dot layers. Typical PL test structures are ﬁ nally completed capping

the dots with a 100 nm thick GaAs barrier, and are then characterized at room temperature (RT)

exciting the dots with a 632.8 nm He–Ne laser and detecting the photoluminescence with an

uncooled InGaAs detector.

Varying indium content and thickness of the InGaAs capping allows a ﬁ ne tuning of the PL

characteristics of each QD layer, which is very useful for the optimization of CMQDs. The depend-

ences of the RTPL peak wavelength of the samples as a function of the In composition and the

thickness of the InGaAs layer are shown in Fig. 19.4 . Insets show the normalized PL spectra

of the samples with the different In composition or the thickness of the InGaAs layer. Between

0 and 5 nm, the PL peak wavelength is proportional to the thickness of the InGaAs layer as

shown in Fig. 19.4a . The PL peak wavelength increases from 1140 to 1300 nm upon increas-

ing the InGaAs layer thickness from 0 to 6 nm. Similar behaviours are also observed when the

In composition of the InGaAs layer is changed, which is shown in Fig. 19.4b . The signiﬁ cant PL

peak wavelength variation mainly appears in the range of In composition between 0 and 20%.

A further increase of In composition does not contribute to the further PL peak wavelength shift.

These ﬁ ndings are in good agreement with previous reports [23, 24] . The red shift of the QD

emission with increasing In composition and thickness of the InGaAs layer can be attributed to

a reduction in the strain in the QDs [24] , to the suppressed In segregation from the QDs into the

cap layer, and to the activated spinoidal decomposition of the cap layer [25] .

CH019-I046325.indd 570CH019-I046325.indd 570 6/27/2008 5:21:29 PM6/27/2008 5:21:29 PM

Quantum Dot Superluminescent Diodes 571

By appropriately choosing In composition or thickness of the InGaAs capping layers, the emis-

sion of CMQDs should cover a wide spectral range. This idea can be easily validated by imple-

menting several QD layers with different capping properties in the same structure. Figure 19.5

(continuous line) shows the normalized RTPL spectra of a CMQD structure containing ﬁ ve QD

layers. The In composition in the InGaAs layer is varied from 0.09 to 0.15 in steps of 0.015,

which results in a wide PL spectral width of 78 nm. The PL emission is dominated by GS transi-

tion at around 1270 nm, while the high energy tail merging with the main peak comes from ES

transition. The achieved spectral width is much larger than the typical spectral width of 44 nm

from the single-layer QD samples (dashed–dotted line), which veriﬁ es the validity of using the

CMQD structure for spectral broadening.

PL wavelength (m)

1.26

1.20

1.32

1.14

0.1 0.2

In composition in InGaAs layer

0.30.0

0.4

900 1000 1100 1200 1300 1400

a: 0.00

b: 0.05

c: 0.09

d: 0.16

Wavelength (nm)

bc d

GaAs

x 

(b)

1.26

PL wavelength (m)

1.20

1.32

1.14

Thickness of In

0.15

0.85

As (nm)

(a)

abc d

InGaAs

d(nm)

a: 0.0

b: 1.5

c: 3.0

d: 5.0

1100 1200 13001000 1400

Wavelength (nm)

810

Figure 19.4 (a) RTPL peak wavelength as a function of the thickness of the In

0.15

0.85

As layer. (b) RTPL peak

wavelength as a function of the In composition of the 5 nm thick InGaAs layer. Insets show the normalized PL spectra

of the samples with different In composition or thickness of the InGaAs layer.

1.0

0.5

GaAs barriers

1.5

0.0

1200

InGaAs

1250

1300

5 chirped layers

Single layer

QDs

1150

1350

Wavelength (nm)

Normalized intensity (a.u.)

Figure 19.5 Normalized RTPL spectra of a single layer of InAs QDs and a chirped multiple InAs QDs sample.

Inset : Schematic band diagram of the chirped QD structure.

CH019-I046325.indd 571CH019-I046325.indd 571 6/27/2008 5:21:29 PM6/27/2008 5:21:29 PM

572 Handbook of Self Assembled Semiconductor Nanostructures for Novel Devices in Photonics and Electronics

19.2.2 Increasing the inhomogeneous spectral width of a single layer of QDs

Although using the CMQD structure is effective to increase the spectral width, the number of

QD layers needed for a given total linewidth without a spectral dip, and the required accuracy

in the In content, depends on the linewidth of each layer. Increasing the inhomogeneous linew-

idth of each layer helps to obtain a broad and uniform spectrum. In this paragraph, we show the

effect of growth conditions on the linewidth, and ﬁ nd optimized conditions for large linewidth

and high radiative efﬁ ciency.

Growth temperature and amount of InAs deposited to form the dots have a strong impact

on the QD size dispersion over the ensemble and on the material quality. Figure 19.6 shows the

RTPL spectra of 2.2 ML InAs QDs deposited at various growth temperatures from 485 to 530°C

and at a growth rate of 0.1 ML/s (monolayer per second). QDs are capped with a 5 nm thick

InGaAs layer and covered by a 100 nm thick barrier. The spectral width of the QDs increases and

the PL efﬁ ciency deteriorates when lowering the growth temperature. This can be attributed to a

reduced diffusion length of adatoms at lower growth temperatures [26] . Nucleation centres on

the epitaxial surface increase, giving rise to a high density of small dots. The size distribution of

these dots is inhomogeneous which results in the broad emission spectrum. There is an optimized

growth temperature around 510°C which gives a broad spectrum emitting near 1300 nm. Its PL

integrated intensity is comparable to that of the samples grown at higher temperature.

533°C

520°C

510°C

500°C

485°C

InAs  2.2 ML; Gr  0.1 ML/s

1.0

Intensity (a.u.)

1.1 1.2 1.3 1.4 1.5

Wavelen

th (m)

Figure 19.6 RTPL spectra of the 2.2 ML InAs QDs deposited at various growth temperatures for a growth rate of

0.1 ML/s.

Figure 19.7 shows the RTPL spectra of the QDs with different InAs thicknesses deposited at

growth temperatures of 485 and 510°C. With increasing InAs thickness, the PL peaks of the

samples grown at both temperatures shift systematically to the long wavelength side due to an

increased dot size. For the QDs deposited at 485°C, the spectra are broad. However, the QDs show

very low PL efﬁ ciency and the PL emission peak is blue shifted as compared to the QDs grown

at 510°C, which prevents accessing the 1300 nm range. Although the spectral widths are large,

the QDs grown at 485°C may not be suitable for device applications. For the QDs deposited at

510°C, the spectral width of the QDs becomes small with increasing InAs thickness due to the

improved dot size uniformity [27] . A broad spectral width up to 96 nm is only obtained when

the InAs thickness is  2.2 ML due to the dot size dispersion. There might exist several groups of

QDs with different sizes, each of them emitting at different wavelengths, thereby giving rise to a

broad spectrum. The PL integrated intensities of these QDs are comparable to that of the 2.4 ML

thick InAs QD layer, which has the strongest PL intensity. In fact, 2.4 ML thick InAs QDs have

been successfully used in low threshold current density lasers, implying high radiative recombi-

nation efﬁ ciency. It should be noticed that the PL intensity of QDs with thicknesses larger than

2.4 ML degrades, probably due to the crystal defects such as dislocations which may have been

generated in larger islands, so that they grow faster or even at the expense of the coherent ones

CH019-I046325.indd 572CH019-I046325.indd 572 6/27/2008 5:21:30 PM6/27/2008 5:21:30 PM

Quantum Dot Superluminescent Diodes 573

[28, 29] . In contrast, there is no dislocation in QDs with InAs thicknesses less than 2.4 ML,

implying efﬁ cient radiative recombination.

By incorporating such dispersed InAs QDs into a CMQD structure, even larger spectral width

and higher gain can be expected. Two approaches were developed to stack QD layers hav-

ing chirped emission wavelength, by varying (i) the In composition in the InGaAs layer and

(ii) the InAs thickness. The RTPL spectra from the two types of structure are depicted in Fig.

19.8 . A spectral width of 91 nm (solid line) was achieved in a seven-layer stack with identical

InAs QDs (2.2 ML thick) capped by three different In compositions of an InGaAs layer (12%,

15% and 17%; structure A). The value is larger than the value of 78 nm achieved by stacking

2.4

10x

2.8 ML

2.6 ML

2.4 ML

2.2 ML

2.0 ML

510°C

485°C

1.6

Intensity (a.u.)

0.8

3.2

0.0

1.0 1.21.1 1.3 1.40.9 1.5

Wavelength (m)

Figure 19.7 RTPL spectra of the QDs with different InAs thicknesses deposited at growth temperatures of 485°C

and 510°C.

1.5

1.0

2.0

0.5

1.04 1.12 1.20 1.28 1.36

Wavelength (m)

PL intensity (a.u.)

Figure 19.8 RTPL spectra from the two types of chirped QD structures. Structure A: seven-layer stack with

identical thick InAs QDs (2.2 ML) capped by InGaAs layers with three different In compositions (12%, 15%

and 17%); structure B: seven-layer stack with two different InAs thicknesses (2.1 and 2.2 ML) capped by InGaAs

layers with the same In compositions.

CH019-I046325.indd 573CH019-I046325.indd 573 6/27/2008 5:21:30 PM6/27/2008 5:21:30 PM

574 Handbook of Self Assembled Semiconductor Nanostructures for Novel Devices in Photonics and Electronics

chirped high uniform InAs QD layers. The largest spectral width of 124 nm was achieved in a

seven-layer stack with two different InAs thicknesses (2.1 and 2.2 ML) capped by InGaAs layers

with the same In composition (dash–dot line; structure B). These spectral widths correspond to

the combined GS  ES emission. For the sample with a different InAs thickness, the deconvolved

GS spectral width is still larger than 100 nm. The optimized CMQDs with 124 nm PL linewidth

(structure B), once incorporated in an SLED device structure, result in a smooth spectral width

(GS  ES) of 115 nm [17] as will be detailed in section 19.4.2.

19.2.3 Conclusions and perspectives

In this section we investigated the growth of size inhomogeneous InAs QDs and CMQD struc-

tures, in order to broaden the gain spectrum of the QD active region in the 1200–1300 nm

wavelength range. By intentionally increasing the size dispersion of InAs QDs, a broad PL spec-

tral width is achieved from a single layer of QDs. No evidence of degradation of optical proper-

ties was observed. The results conﬁ rm that the CMQD structure is effective for devices which

need a broad-gain spectrum such as SLEDs, semiconductor optical ampliﬁ ers and tunable lasers.

Although this discussion has been conducted on the InAs/GaAs material system for applications

in the 1300 nm region, the concepts and approaches can be easily extended to the other material

systems. In particular, it becomes possible to realize QD or quantum dash SLEDs covering various

wavelength ranges based on the material systems such as InAs QDs/quantum dashes on InP sub-

strate (1550 nm telecom wavelength) [30, 31] , InP/InGaP QDs on GaAs substrate (700–800 nm

red emission) [32, 33] , and InGaN QDs on GaN substrate (400–500 nm blue emission) [34, 35] .

19.3 Wide-spectrum InAs/GaAs QD SLEDs

19.3.1 Gain and length requirements

The typical structure of QD-SLEDs emitting in the 1.2–1.3 μ m wavelength region is made of a

certain number of QD layers separated by 35–40 nm GaAs spacers to avoid strain and coupling

effects, and embedded in the GaAs intrinsic region of a p-i-n junction. The p - and n -regions are

1  x

As layers, which provide both optical and electrical conﬁ nement. Starting from these

structures, devices are realized by dry-etching tilted or bent waveguides (see inset of Fig. 19.11 )

which are then passivated by BCB planarization or by steam oxidation of the exposed AlGaAs

cladding depending on the Al content in the alloy. Even though high Al content results in a high

optical conﬁ nement, a low Al concentration can be desirable for the applications, to minimize

the reliability issues related to the aluminium oxidation and to improve the far-ﬁ eld aspect ratio

of the emission. Ridges are a few μ m in lateral size to ensure a single lateral mode operation and

therefore an efﬁ cient coupling into tapered single-mode ﬁ bres (the typical coupling efﬁ ciency

is 40–50% when the SLED operates in the ampliﬁ cation regime), which is highly desirable for

applications. The last fabrication steps are p -contact (Ti/Pt/Au) and n -contact (Ni/Ge/Au/Ni/Au)

deposition, on top and on the substrate side, respectively, and ﬁ nally cleaving chips of the proper

length.

The most important parameters related to the superluminescent diode output power are the

chip length L and net gain g , which relates to modal gain g

mod

and internal loss α

through the

relation g  g

mod

 α

. As mentioned in section 19.1.3, in a simpliﬁ ed picture where the output

facet reﬂ ectivities are assumed be negligible and the gain and spontaneous emission rate are

assumed to be independent from positions inside the waveguide, an exponential increase in out-

put power is expected when increasing g or L :

out



⎡

⎣

⎤

⎦

(19.4)

In Fig. 19.9 a the output power measured over a series of QD SLEDs with similar geometrical

characteristics (tilted waveguides with ridge widths of 3–4 μ m, 4 mm long) are reported in the

CH019-I046325.indd 574CH019-I046325.indd 574 6/27/2008 5:21:31 PM6/27/2008 5:21:31 PM

Quantum Dot Superluminescent Diodes 575

regime of emission where GS  ES versus the measured maximum net gain g

max

. In spite of

the many differences between the samples (number of QD layers, cladding compositions, band-

widths, etc.) the data points show a clear exponential increase with increasing gain, in agree-

ment with Eq. 19.4, and provide a path to the achievement of high optical powers. Outputs of

10 mW or higher can be achieved only when g

max

is of the order of 18–20 cm

 1

(i.e. chip gain

max

L  8). A ﬁ t is reported in the plot, which was made using Eq. 19.4, where β P

/g was

assumed constant and used as a ﬁ tting parameter and L was ﬁ xed to 4 mm. This corresponds to

a spontaneously emitted output power per unit length coupled into the guided mode at a gain of

20 cm

 1

of 25 μ W mm

 1

The effect of the device length on the L – I characteristics of tilted waveguide superlumines-

cent diodes can be observed in Fig. 19.9b . The Figure shows the L – I characteristics of three

devices fabricated from the same wafer and cleaved to 1 mm, 3 mm and 4.5 mm length. These

devices contained ﬁ ve identical QD layers with a relatively narrow gain spectrum (20–40 nm

around 1.3 μ m, depending on the injection and ampliﬁ cation regime) embedded into a GaAs/

0.8

0.2

As waveguide. Measurements were performed in pulsed regime, at room temperature.

The output power of the 1 mm long devices saturates around 300 μ W and does not show any evi-

dent exponential characteristic in the L – I curve, which would be typical of the superluminescent

emission. The 3 mm long device in contrast reaches output powers in the mW range. For this

device two different regimes of ampliﬁ cation can be observed, one at low and one at higher injec-

tion, which are related to recombination from different bound states in the QDs. The double expo-

nential increase is an intrinsic phenomenon in QD superluminescent diodes where the relatively

low density of states of the lower energy levels leads to state ﬁ lling and subsequent population

of the higher energy levels. This phenomenon, which can be exploited in quantum dots for the

achievement of very large bandwidths combining the emission from different states, will be

the subject of a deeper analysis in the next paragraphs. The L – I characteristics of the 4.5 mm

long SLED reported in Fig. 19.9 b show even higher ampliﬁ cation with output powers exceeding

10 mW mainly composed of ground state emission.

The two parameters having an important effect on the maximum modal gain

max

mod

are the

number of QDs contributing to the emission and the gain spectral bandwidth. The number of

dots introduced in the device active region, which are related to the DOS of the system, is directly

proportional to the modal gain. Figure 19.10 a shows

mod

max

measured over three samples contain-

ing ﬁ ve, ten and 18 QD layers, at the peak wavelength, all of them having similar PL linewidths

of about 80 nm. These devices show a typical

mod

max

of 1–2 cm

 1

per dot layer at the peak posi-

tion, which becomes slightly higher for samples with a narrower inhomogeneous broadening.

The line in the plot is a guide to the eye and has been intentionally considered sub-linear. This

is because the three waveguiding structures are not identical and the width t

opt

of the optical

0.1

4.5 mm

3mm

1mm

0.01

0.3 0.6 0.90 1.2

Current (A)

Output power (mW)

(b)

100

0.1

Output power (mW)

(a)

12 14 16 18 20622

Net gain (cm

1

)

Figure 19.9 (a) Output power measured over a series of 4 mm long, tilted QD-SLEDs in the regime of GS  E S

emission, versus the corresponding maximum net gain. The data are ﬁ tted using Eq. 19.4. (b) L – I characteristics of

tilted-waveguide SLEDs for varying device length (fabricated from the same wafer).

CH019-I046325.indd 575CH019-I046325.indd 575 6/27/2008 5:21:31 PM6/27/2008 5:21:31 PM

576 Handbook of Self Assembled Semiconductor Nanostructures for Novel Devices in Photonics and Electronics

mode in the growth direction is expected to be larger for the ﬁ ve and ten QD layer samples, lead-

ing to a smaller conﬁ nement factor (

mod

max

is inversely proportional to t

opt

). t

opt

indeed depends on

the device geometry and in particular varies proportionally to the GaAs core region thickness,

and depends on the refractive index contrast between active region and claddings (higher Al con-

tent in the claddings means higher refractive index contrast and higher conﬁ nement and there-

fore smaller t

opt

). The ﬁ ve QD layer sample has 80% Al concentration against 35% for the other

two samples. Also, the GaAs core region is thicker in the 18 QD layer sample than in the others.

As the spacing between different dot layers must be kept large (35–40 nm) to reduce strain and

coupling effects, the higher number of QD layers requires the growth of a thicker active region,

thus resulting in a larger t

opt

Once the number of QD layers is ﬁ xed (each layer with approximately ﬁ xed QD density of

3.10

 2

), the material gain at the peak position depends only on the shape of the broadened

density of states, whose integrated value is proportional to the total number of dots. If we con-

sider Gaussian broadenings for the GS DOS, which is usually the case for single QD layers as con-

ﬁ rmed by the shape of the PL emission, the peak gain is reduced proportionally to the increase in

width of the broadening. The decrease in gain was observed measuring the peak gain in samples

with ten QD layers and a different PL linewidth.

mod

max

decreased from 20 to 15 cm

 1

for a low-

excitation PL linewidth increase from 50 to 80 nm.

Number of QD layers and inhomogeneous broadening are not the only two parameters affect-

ing the gain characteristics of QD-SLEDs. The thermal redistribution of carriers over the energy

levels of a semiconductor is one of the most important parameters limiting the gain perform-

ance of devices operating at room temperature. This is true for bulk materials, where electrons

and holes occupy the continuous energy bands with quasi-Fermi distributions, but also for QDs

where the presence of close excited state energy levels may reduce the occupation and therefore

the gain of the lower energy levels. In [36] , for example, is shown how the proximity of the WL

levels (especially in the valence band) of QDs emitting around 1000 nm generates a carrier redis-

tribution reducing the maximum modal gain. Furthermore, the presence of many closely spaced

energy levels for the holes in the valence band could produce a similar effect. A few years ago the

introduction of p -doping in the QD active region was proposed as one possible solution to over-

come this problem by forcing a certain number of extra holes to be in the dots [37] . Since this

moment, many groups all over the world have been studying the effect of p-doping on the tem-

perature, gain and modulation characteristics of quantum dot lasers. Here we report the effect

of p-doping on the gain characteristics of the QD material. Later on, the effect of p-doping on

the L – I and temperature characteristics of SLEDs will be discussed. Three samples with identi-

cal epitaxial structure, grown under the same conditions, but with different doping levels in the

active region were processed into SLEDs. Each of them contains ten InAs QD layers in a GaAs

active region embedded between 1.5 μ m thick Al

0.35

0.65

As cladding layers. In sample A the dot

Maximum net gain (cm

1

)

(a)

Modal gain (cm

1

)

(b)

1000 2000

3000

Current density (A/cm

)

10

20

30

40

10 15520

QD layers

Figure 19.10 (a) Maximum modal gain versus number of QD layers for samples showing a comparable PL

linewidth. (b) Modal gain of 2 mm long and 3 μ m wide tilted ridge-waveguide devices with identical epitaxial

structure but different doping levels in the active region.

CH019-I046325.indd 576CH019-I046325.indd 576 6/27/2008 5:21:32 PM6/27/2008 5:21:32 PM

Quantum Dot Superluminescent Diodes 577

layers are separated by undoped GaAs spacers, while in samples B and C carbon doping is intro-

duced inside a thin region within each GaAs spacer (10 nm thick regions, 10 nm above each QD

layer). The doping level was estimated to correspond to the introduction of about eight and 15

extra holes per QD, for samples B and C, respectively. The spectral characteristics showed very

similar behaviour for the three samples, except for small variations in the spectral centre wave-

length (1267, 1270 and 1300 nm for samples A, B and C, respectively), demonstrating that for

the considered doping levels the dispersion of the QD density of states is not substantially modi-

ﬁ ed. In Fig. 19.10 b the modal gain curves for TE polarization (electric ﬁ eld perpendicular to the

growth direction) at the GS peak position versus current density are reported for the three sam-

ples. Each line interpolates a series of 80 data points obtained through the analysis of the ampli-

ﬁ cation undertaken by a tunable laser injected in the tilted waveguides and tuned at the GS peak

wavelength. The measurements were veriﬁ ed to be reproducible over several nominally identical

devices. The plot shows an increasing maximum modal gain with increasing doping level. The

maximum value of g

mod

for samples A, B and C is 18, 22 and 25 cm

 1

, respectively. The increase

can be attributed to the modiﬁ ed carrier distribution in the valence band due to the introduction

of several acceptors per QD, which has the effect of pushing the quasi-Fermi level deeply inside

the band, and thus increasing the hole population contributing to the gain. We note that an

increased injection current is needed in p -doped devices to achieve the same gain level. P -doping

is expected to increase monomolecular (through dopant-related defects), radiative and Auger

(through increased hole population) recombination rates. A combination of these effects is likely

to produce the shift of the gain curves towards higher current. P -doping has an impact on the

optical losses as well, through increased free carrier absorption and photon scattering due to the

introduction in the active region of doping-related crystal defects. From the measured Fabry–

Perot fringe visibility and laser thresholds in untilted devices we estimate the optical losses as

 1.8 cm

 1

, 3.5 and 5 cm

 1

for samples A, B and C, respectively. The increase in modal gain

in p -doped structures B and C exceeds the increase in optical losses, resulting in a larger net gain

which, as we will see in the next section, results in higher output power for p -doped SLEDs.

As already discussed, the SLED output power does not depend simply on the material gain but

also on the distance travelled by photons inside the waveguide before output. In tilted devices this

can be assumed to be equal to the device length (single-pass ampliﬁ cation). In contrast, combin-

ing the low reﬂ ectivity of one facet with the high reﬂ ectivity of the opposite, in bent devices the

photon may cover a larger length than the cavity length before output. This leads to higher light

ampliﬁ cation and therefore higher output power, provided that the waveguide curvature does

not introduce signiﬁ cantly higher internal loss. With this purpose bent stripes were realized by

using an arc of circumference whose tangents at the two end points are 7 ° tilted and perpendicu-

lar to the opposite facets of a 2 mm long cavity, respectively. Longer cavities were also realized

by adding a straight piece of waveguide to the perpendicular facet side. This combines the low

reﬂ ectivity of the tilted facet with the high reﬂ ectivity of the perpendicular facet so that the light

travelling inside the waveguide may experience double-pass ampliﬁ cation. A comparison between

the L – I characteristics of a 2 mm long bent and 2 mm long tilted waveguide are reported in Fig.

19.11 . Both devices were fabricated with the same process and from the same wafer, thus dem-

onstrating that the bent structure is effective in increasing the light ampliﬁ cation if compared to

a single-pass guide. While the output power of the tilted waveguide is limited to few mW at high

injection, bent waveguides achieve 30 mW at the same injection.

In the following paragraphs, we will ﬁ rst describe the characteristics of SLEDs fabricated from

samples with a narrow gain, optimized for emission at 1.3 μ m, as they are the standard gain

material for high-performance QD lasers. This results in devices with narrow optical bandwidths

but at the same time provides useful information for the understanding of the device character-

istics. Then, in section 19.3.3 we will discuss the properties of SLEDs with chirped active regions

showing that through the optimization of growth conditions – as discussed in section 19.2 – the

device output may exhibit bandwidths larger than 100 nm. The large bandwidth is obtained in a

regime of two-state emission (GS  ES) that, for optimized chirping, results in a smooth spectral

superposition (ﬂ at top). The coherence properties and temperature characteristics of both types

of devices are addressed in sections 19.3.4 and 19.3.5.

CH019-I046325.indd 577CH019-I046325.indd 577 6/27/2008 5:21:32 PM6/27/2008 5:21:32 PM

578 Handbook of Self Assembled Semiconductor Nanostructures for Novel Devices in Photonics and Electronics

19.3.2 Narrow-gain devices

In this paragraph we report the spectral and L – I characteristics of SLEDs realized from samples

containing ﬁ ve and ten identical QD layers. More details about these devices can be found in [13,

15] . Tilted-ridge SLEDs fabricated from these samples were mounted p-side up on a temperature

controlled heat-sink, and tested under pulsed operation. In Fig. 19.12a the L – I characteristics of

a 3 mm long SLED are reported for the heat-sink temperature of 10°C. Two separated superlinear

behaviours can be identiﬁ ed in the curve: one at low injection (0–200 mA) and one at higher

injection (600–800 mA) which are related to the recombination from GS and ES energy levels of

the dot. The superlinear increase in output power together with the high power (of the order of

10 mW at high injection) are the signature of operation in the gain regime.

0 0.2 0.4 0.6 0.8 1 1.2

Current (A)

1200 mA

600 mA

300 mA

100 mA

1

2

3

4

5

1000 1100 1200 1300 1400

26 nm FWHM

22 nm FWHM

Output power (mW)

Wavelength (nm)

Intensity (a.u.)

(a)

0 200 400 600 800

0.1

100

Power (mW)

Current (mA)

(b)

Figure 19.12 (a) L – I characteristics of a 3 mm long tilted ridge-waveguide SLED containing ﬁ ve QD layers,

measured at 10°C in pulsed regime. The corresponding spectra are reported in the inset. (b) L – I characteristics of

4 mm long tilted ridge-waveguide SLEDs containing ten QD layers and different p -doping levels in the active region.

The circles identify the regime of emission where GS and ES contributions to the spectra are comparable.

0.1

100

200 300 400

500

600 700 800

Current (mA)

Output power (mW)

Bent

Tilted

Bent

Tilted

Figure 19.11 Bent versus tilted SLED structures (2 mm long chips): bent waveguides allow the achievement of

higher output powers at the same injection levels. A schematic drawing of the two structures is reported in the inset.

A better understanding of the L – I characteristics can be achieved observing the corresponding

spectra reported in the inset of Fig. 19.12a . At low injection the GS provides the main contribu-

tion to the emission with 26 nm full width at half maximum (FWHM) and up to 2 mW output

power. At higher currents this emission saturates because of the GS ﬁ lling and a second line due

to recombination from the ﬁ rst ES appears. This line is eventually dominant (exceeding 10 mW)

CH019-I046325.indd 578CH019-I046325.indd 578 6/27/2008 5:21:32 PM6/27/2008 5:21:32 PM