Integrated tunable quantum-dot laser for optical coherence
tomography in the 1.7µm wavelength region
Citation for published version (APA):
Tilma, B. W. (2011). Integrated tunable quantum-dot laser for optical coherence tomography in the 1.7µm wavelength region. Technische Universiteit Eindhoven. https://doi.org/10.6100/IR712655
DOI:
10.6100/IR712655
Document status and date: Published: 01/01/2011 Document Version:
Publisher’s PDF, also known as Version of Record (includes final page, issue and volume numbers) Please check the document version of this publication:
• A submitted manuscript is the version of the article upon submission and before peer-review. There can be important differences between the submitted version and the official published version of record. People interested in the research are advised to contact the author for the final version of the publication, or visit the DOI to the publisher's website.
• The final author version and the galley proof are versions of the publication after peer review.
• The final published version features the final layout of the paper including the volume, issue and page numbers.
Link to publication
General rights
Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights. • Users may download and print one copy of any publication from the public portal for the purpose of private study or research. • You may not further distribute the material or use it for any profit-making activity or commercial gain
• You may freely distribute the URL identifying the publication in the public portal.
If the publication is distributed under the terms of Article 25fa of the Dutch Copyright Act, indicated by the “Taverne” license above, please follow below link for the End User Agreement:
www.tue.nl/taverne
Take down policy
If you believe that this document breaches copyright please contact us at:
openaccess@tue.nl
providing details and we will investigate your claim.
Integrated tunable quantum-dot laser
for optical coherence tomography
in the 1.7 µm wavelength region
PROEFSCHRIFT
ter verkrijging van de graad van doctor aan de
Technische Universiteit Eindhoven, op gezag van de
rector magnificus, prof.dr.ir. C.J. van Duijn, voor een
commissie aangewezen door het College voor
Promoties in het openbaar te verdedigen
op dinsdag 28 juni 2011 om 16.00 uur
door
Bonifatius Wilhelmus Tilma
geboren te Amsterdam
Dit proefschrift is goedgekeurd door de promotor: prof.dr.ir. M.K. Smit
Copromotor: dr. E.A.J.M. Bente
The research presented in this thesis was supported by the IOP Photonic Devices program managed by Agentschap NL (Ministerie van Economische Zaken, Landbouw en Innovatie), Technologiestichting STW and the NRC photonics grant.
Copyright © 2011 Bonifatius Wilhelmus Tilma Printed in The Netherlands
A catalogue record is available from the Eindhoven University of Technology Library Tilma, Bonifatius Wilhelmus
Integrated tunable quantum-dot laser for optical coherence tomography in the 1.7µm wavelength region / by Bonifatius Wilhelmus Tilma. –Eindhoven : Technische Universiteit Eindhoven, 2011.
Proefschrift. – ISBN: 978-90-386-2499-0 NUR 959
Trefw.: halfgeleiderlasers / geïntegreerde optica / 3-5 verbindingen / afstembare laser / quantum-dots
Subject headings: semiconductor lasers / integrated optics / III-V semiconductors / tunable laser / quantum-dots
Jen dy’t fan wysheid hâldt makket syn heit bliid.
Spr. 29:3a
5
Contents
Contents ... 5
1
Introduction ... 9
1.1 Optical coherence tomography ... 10
1.1.1 Requirements on tunable laser for FD-OCT... 12
1.2 Approach to the laser design ... 14
1.2.1 Laser simulation ... 15 1.3 Thesis Outline ... 17
2
Integration technology ... 19
2.1 Introduction ... 19 2.2 Fabrication Technology ... 21 2.2.1 Active-passive layerstack ... 212.2.2 Active-passive processing scheme ... 22
2.3 Specific processing used in this thesis ... 26
2.3.1 Waveguide structure at 1700nm ... 26
2.3.2 Top buffer layer ... 28
2.3.3 Doping passive waveguides ... 28
2.3.4 InP etch stop layer in the film layer ... 29
2.3.5 Planarization and etch back Polyimide ... 29
2.3.6 Metallization ... 30
2.4 Chip mounting for measurement ... 30
3
Measurement and analysis of the gain in quantum-dot amplifiers 33
3.1 Introduction ... 333.2 Gain measurement method, device design and fabrication ... 35
3.3 Gain measurement results ... 37
3.4 Quantum-dot amplifier gain model ... 40
3.5 Simulations ... 46
3.6 Comparison with measurements... 49
6
4
Tunable wavelength filters in the 1.6 to 1.8 µm wavelength region 51
4.1 Introduction ... 51
4.2 Tuning principle ... 54
4.3 Design ... 56
4.4 Layerstack and waveguide design ... 57
4.4.1 High resolution filter ... 58
4.4.2 Low resolution filter ... 59
4.4.3 Mask layout ... 62
4.5 Fabrication ... 63
4.6 Measurements ... 64
4.6.1 Static filter characteristics ... 65
4.6.2 Calibration method of the phase modulators ... 66
4.6.3 Calibration HR-filter ... 68
4.6.4 Calibration LR-filter ... 70
4.7 Tuning results ... 71
4.7.1 HR-filter ... 72
4.7.2 LR-filter ... 73
4.7.3 Filter Tuning speed ... 74
4.8 Conclusion ... 75
5
Tunable quantum-dot ring laser simulations ... 77
5.1 Introduction ... 77
5.2 Laser model ... 78
5.2.1 QD amplifier rate equation model ... 78
5.2.2 Segmented ring laser model ... 80
5.3 Simulations ... 82
5.3.1 Simulation 1: complete output spectrum ... 83
5.3.2 Simulation 2: 1nm around laser peak (start-up) ... 85
5.3.3 Simulation 3: 1nm around laser peak (0.2nm tuning)... 87
7
6
Integrated tunable quantum-dot laser in the 1.7 µm wavelength
region ... 91
6.1 Introduction ... 91
6.2 Design and fabrication ... 93
6.3 Laser characterization ... 95
6.3.1 Laser tuning – influence of the gain spectrum ... 97
6.3.2 Laser tuning – influence of the cavity mode structure ... 100
6.3.3 Laser tuning speed ... 102
6.4 Laser coherence length and effective linewidth ... 103
6.5 Conclusion ... 106 6.5.1 Tuning bandwidth ... 107 6.5.2 Laser linewidth ... 108 6.5.3 Scan rate ... 108 6.5.4 Output power ... 109 6.5.5 Overall conclusion ... 109
References ... 111
List of abbreviations ... 121
Summary ... 123
List of publications ... 125
International Journals ... 125 International Conferences ... 125 Local Conferences ... 127Acknowledgement ... 129
Curriculum vitae ... 133
9
1 Introduction
The control of mankind on light sources has had large influences on human development. Two significant technology breakthroughs in history are the invention of the light bulb in the 19th century and the experimental proof of the laser in 1960 by Theodore
Maiman [49]. The control of emission characteristics of light sources has its impact on many developed technologies such as the use of lasers in telecommunications, surgery and military applications as well as other more end-user technologies such as the barcode scanner, the optical disc reader and the laser printer.
Well-controlled light sources have also been introduced in a variety of measurement techniques owing to the often extremely precise measurement results that can be achieved [71]. Examples of measurement techniques in which light sources are already used range from geographical measurement techniques such as airborne laser relief scanning [27], structure health monitoring for example with fiber Bragg sensing [38], and different medical measurement techniques such as optical coherence tomography (OCT) in ophthalmology [23] spectral imaging in dermatology and hematology [86] and Sidestream Dark Field Imaging in cardiovascular applications [30]. The required optical properties and the related choice of light sources are different for each of the measurement techniques.
One of these measurements techniques, Optical Coherence Tomography (OCT), is an imaging technique where cross-section images can be obtained at micrometer scale resolution. The cross section images can be reconstructed from the phase and intensity information in backscattered light from within a sample. This information can be retrieved in an interferometer setup where the light scattered by different materials in the sample interferes with light from a reference path. One of the major advantages of OCT is that it is particularly useful for imaging of biological tissues with a resolution of several micrometers. Images can be obtained in vivo without physical contact. In ophthalmology this is a very useful property. It allows for obtaining detailed images of the retina, the light-sensitive inner backside of the eye, and the anterior segment (front side of the eye). Since useful diagnostics can be made from such OCT images, the technique is used often in ophthalmology [72]. To image the retinal structure the wavelength ranges around 800nm and 1050nm light are often used. Typically 1300nm is used to image the anterior segment. The imaging depth into the tissue that can be achieved using OCT is generally dependent on the scattering coefficient. Absorption typically tends to be less important than scattering in biological samples. The scattering coefficient decreases in most cases with increasing wavelength. This increase in imaging depth with increasing wavelength is however interrupted due to the strong water absorption in the two water absorption bands between
Chapter 1
10
1400nm and 1500nm and between 1900nm and 2200nm. The spectral range between 1600nm and 1800nm is however a promising wavelength range for OCT which can be used for deeper imaging in human tissue. The relation between wavelength and imaging depth and the necessary spectral width needed for a 10µm resolution is given in Fig. 1-1 [22]. A quantitative comparison of the OCT imaging depth between 1300nm and 1600nm was investigated and presented in [40].
Fig. 1-1 The calculated imaging depth (solid curve) and bandwidth requirement (dashed curve) on the OCT light source to maintain a resolution of 10µm as a function of wavelength. The image is calculated for typical parameters of an OCT system, the scattering coefficient and the water absorption spectrum. The imaging depth is defined as the depth at which the contributions of the single and multiple scattering to the OCT signal are equal. (Figure from [22])
In this introduction chapter, an introduction to the different OCT measurement techniques is first presented. Given a particular choice of OCT type, the requirements on the light source are determined from the requirements on the image and the imaging setup. In section 1.2 we discuss the general approach of the realization of a tunable laser for a frequency domain OCT setup. The tunable laser is realized using tunable intra-cavity filters. The requirements on the filters are determined in this section from a simple time domain laser model. Finally an outline of the thesis is presented.
1.1 Optical coherence tomography
OCT is an interferometric three dimensional imaging technique often used in ophthalmology. The measurement technique relies on the analysis of back-scattered light from a tissue interfered with light from a reference arm. The basic scheme of an OCT measurement setup is given in Fig. 1-2. The choice of light source, light detector and the
Introduction
11
interference setup is dependent on the type of OCT measurement. The main OCT types are time domain OCT and frequency domain OCT. Both types will be briefly discussed.
Reference mirror Light source
Detector
Tissue
Fig. 1-2 Basic measurement scheme of an OCT setup. The choice of light source, detector and the details on the interference path depends on the type of OCT.
Time domain OCT
In a time domain (TD) OCT measurement, also known as low coherence interferometry, a broadband light source is used to illuminate the tissue. This broadband light source has a short coherence length that is typically in the order of a few to ten microns. The light reflected from the tissue will only interfere constructively with the light coming from the reference arm if the difference in path length is within the coherence length of the broadband light source. An increase in signal intensity is measured at the detector in that case. By changing the path length in the reference arm the depth in the tissue from which the reflection intensity is detected can be selected. The depth image can be measured directly by measuring the interfered light as a function of arm length difference. A major drawback of this OCT technique is the physical arm length change necessary in the setup. This change in path length can for example be introduced with a moving mirror. In most cases the problem with such a mechanical moving mirror is that the scan speed seriously limits the speed of the overall measurement.
Frequency domain OCT
In frequency domain (FD) OCT the depth image is reconstructed from measuring a spectrally resolved interferometer signal. This can either be executed with a wavelength scanning light source and recording the intensity of the individual reflected wavelength components in time, which is the solution proposed in this thesis, or by illuminating the tissue with a broadband light source and measuring the intensity of the individual wavelength components with a spectrometer. In the latter case, the spectrometer can either be a monochromator with a single photo-detector or a dispersive element, typically a grating, which distributes the different wavelengths on to a detector array. Solutions based
Chapter 1
12
on detector array are much preferred since it is a more efficient detection setup by two to three orders of magnitude. In either case, the depth image can be reconstructed by Fourier transforming the (spectral) wavelength components to the time or space domain.
The first advantage of the measurement technique with the tunable laser is that no light is lost in a spectrometer system so less power can be used or the signal intensity is higher. The second advantage is that the signal level on the detector is much higher since all power is at one wavelength. The signal from the detector is integrated only over the time that the laser is at one particular wavelength. This means that the contribution from noise in the detector is lowered significantly. More information about the different OCT techniques can be found in [22].
The preferred option for FD-OCT is therefore to use a narrow linewidth laser source for the light source and use a single photo-detector. An additional advantage of using a tunable laser is that the measurement can benefit from the often relatively narrow linewidth and therefore long coherence length of these lasers, the importance of which will be discussed in the next section. The design and construction of a suitable tunable laser is however far from trivial. Currently setups using bulk optics are used in most cases to realise the tunable laser source.
1.1.1 Requirements on tunable laser for FD-OCT
Within this thesis the realization of a tunable laser is presented for a frequency domain OCT system. The requirements on such a laser are imposed by the requirements on the imaging performance of the OCT system. The following requirements on the imaging system are given: 2.5mm imaging depth, 10µm imaging depth resolution and an approximately 1 second measurement time for a 200 by 200 pixels image. Each of these requirements imposes its requirements on the tunable laser system:
2.5mm imaging depth: The imaging depth that can be achieved in the OCT
measurement is dependent on the scattering coefficient and the absorption coefficient in the tissue given a specific detection limit for the light detector. As presented above, the wavelength range between 1600nm and 1800nm is promising for imaging approximately 2.5mm deep into typical biological tissue. For the laser this means a tuning bandwidth somewhere in this 1600nm to 1800nm range. For imaging at 2.5mm depth a certain coherence length of the laser is necessary to be able to resolve the image deeper into the tissue. For an image depth of 2.5mm the coherence length Lcoh
must be at least 5mm times the refractive index of the material (n≈1.3 for typical human tissues, which have high water content [31]). This implies a coherence length in vacuum of approximately 6.5-7mm. The coherence length of a laser is directly dependent on the linewidth of the laser. For a tunable laser an effective linewidth can be defined which is the full-with-half-maximum (FWHM) of the laser peak in the output spectrum. The necessary effective linewidth of the laser can be calculated with:
Introduction
13
Lcoh=λ2/2πδλ where δλ is the effective linewidth. The necessary linewidth to obtain
6.5mm coherence length in vacuum is δλ<0.07nm.
10µm image depth resolution: The resolution of the reconstructed image in depth is
inversely proportional to the tuning bandwidth. The image resolution can be calculated (in vacuum) to be: ΔL=0.44λ02/Δλspan [22]. Here ΔL is the image depth resolution (in
vacuum), λ0 the central wavelength of the tuning band and Δλspan the total tuning
bandwidth. Depending on the refractive index of the tissue the image resolution can be calculated accordingly. For an image resolution of 10µm at a wavelength around 1700nm this means in typical tissues a necessary bandwidth of approximately 100nm, as can be seen in Fig. 1-1.
Reflected signal level for reconstruction: The optical signal reflected from the tissue
should be above a certain level to discriminate the signal from the noise floor of the detector. This reflected power is directly dependent on the output power of the light source. It is experimentally estimated at the Academic Medical Center (AMC) of the University of Amsterdam that the output power of the light source should be at least in the 1-10mW range to be able to reconstruct an image from a typical biological tissue. Variations in the output power of the light source as a function of wavelength can be compensated for in the analysis as long as the variation is reproducible.
Image scan rate: The measurement time necessary for a single depth profile (this is
called an A-scan) is given by the time for a single wavelength scan. The time needed for a single scan is dependent on the scan speed and the scanning bandwidth. In case of a 100nm bandwidth, 1000 data points have to be measured to get to 10µm resolution in 2.5mm depth (see image dept and image dept resolution above). Therefore the sampling has to be each 0.1nm. In ophthalmology 3D images need to be recorded. For patient comfort and the minimization of movement artifacts [39] these need to be recorded as fast as possible. For example, with a scan rate of at least 20kHz up to 50kHz an image of approximately 200 by 200 pixels can be built up in the time span of 2 seconds down to 0.8 second. This means a sampling rate of at least 20MHz (50ns samples) for the measurement of 1000 data points in a single scan of 50µs, i.e. a scan speed of at least 2nm/µs. The exact necessary scan rate obviously depends on the preferred number of pixels (this is called the B-scans) in the image. For real-time imaging the calculation time also plays a role but this is not taken into account here. To summarize the requirements on a suitable tunable laser for OCT imaging with a 2.5mm image depth should satisfy the following:
a) Tunable in the 1600nm to 1800nm wavelength region. b) Effective linewidth less than 0.07nm
c) Tunable over at least 100nm d) Output power at least 1-10mW
Chapter 1
14
1.2 Approach to the laser design
A possible approach for the realization of a tunable laser which fulfils the requirements stated above is the integration of the laser on a single semiconductor chip. This is possible due to the recent development of new quantum-dot (QD) gain material in the 1700nm region with a wide gain spectrum [54]. This gain material can be implemented in the active-passive indium-phosphide (InP) integration technology used at the Inter-University Research School on Communication Technologies Basic Research and Applications (COBRA) [7][60], together with electro-optically tunable wavelength selective devices [47]. In this thesis a tunable laser is presented based on this active passive integration technology in combination with the specially designed QD. A schematic picture of the proposed laser design is given in Fig. 1-3. The design is based on a ring laser built up from a QD amplifier to generate and amplify the light in the cavity and an intra-cavity tunable wavelength filter to select the wavelength in the cavity. Part of the light is coupled out of the ring to the output.
QD-amplifier Tunable λ-filter
output output
Fig. 1-3 Schematic picture of the proposed monolithically integrated tunable laser. The ring laser basically consists of a quantum-dot amplifier to generate and amplify the light in the cavity and an intra-cavity tunable wavelength filter to select the wavelength in the cavity. Part of the light is coupled out of the ring to the output.
A ring laser topology is chosen above a linear laser topology. A linear laser has the disadvantage that the light that passes through a tunable filter at the passband wavelength, is reflected on a cleaved facet and returns through the same tunable filter again before it returns to the amplifier again. This light can however be attenuated with more than 25dB through losses in the filters and in reflection from the facet. Using amplifiers after the filter and before the cleaved facet will decrease this problem. In a ring laser however the light passes alternately through the amplifier and the filter. In principle the round trip path length in the ring laser can be twice as short as that in the linear laser which reduces the round trip time and therefore the switching speed of the laser because of the shorter feedback time. Other minor advantages of a ring laser are the arbitrary choice of output coupling and the design is independent of the location of the cleaved facets of the chip. A possible
Introduction
15
disadvantage of a ring laser is the initial bidirectional operation which reduces the output power from one direction. To overcome this disadvantage the ring can be forced to operate unidirectional, for example by means of a feedback loop [81] that reflects a small part of the output light back into the ring cavity [61]. This feedback loop will further be discussed in the chapters 5 and 6.
In the ring laser spontaneous emitted light is first generated in the QD amplifier over a large (100-200nm) wavelength region (around 1700nm). This light travels through the ring in both directions and is filtered by the tunable wavelength filter. After one roundtrip the light which is attenuated by the waveguide losses and the filter is amplified again in the QD amplifier. Due to the passband filtering in the tunable filter the light at the passband wavelength will be stronger than the other spectral components. This effect will be amplified each roundtrip. Due to the mode competition above laser threshold the laser peak will be even narrowed further than the passband characteristics of the filter. Changing the central wavelength of the passband of the tunable filter results in a suppression of the existing laser peak and the appearance of another laser peak at the new passband wavelength. The switching speed between two passband wavelengths is dependent on the suppression of the initial laser peak and on the amount of small signal gain necessary to amplify the new laser wavelength.
1.2.1 Laser simulation
In order to obtain a basic understanding of the operation of the laser and, in particular, to determine initial requirements for the filter characteristics, a simple time-dependent multimode laser model was developed. This model is based on simple rate equations for the average carrier concentration in the amplifier and the cavity average photon densities of five laser modes. The longitudinal mode spacing was assumed to be 0.05nm, based on expectations of a 16mm long laser cavity. Modes adjacent to the target wavelength should be sufficiently suppressed within the required 50ns after switching of the tunable filter. This results in a 50ns switching and a linewidth of less than 0.1nm as required for the OCT application discussed above. In this model the parameters for the amplifier structure were that of a bulk amplifier that could supply sufficient gain. This was necessary since at that time no data were available on the quantum-dot amplifiers. The losses for the different modes were calculated separately, based on typical values for arrayed waveguide grating (AWG) type of filters which can be used for the intra-cavity filtering. The continuous wave (CW) and dynamical behavior of the laser can then be simulated for various filter properties. From the CW state simulations one can find a set of minimum required loss values for the unwanted laser modes. When the passband of the filter is scanned the situation becomes more involved. As the filter is scanned the laser must change from one mode to the next. To study the speed at which the laser mode intensities can follow the filter tuning, one also has to consider the relative losses of the modes, the cavity length, the laser pump level, overall cavity losses and the spontaneous emission intensity. The
Chapter 1
16
multimode dynamics in the laser and the semiconductor optical amplifier (SOA) and the limits on the switching speed are further investigated in chapter 5. Two results from the initial simplified simulations covering five consecutive laser modes (wl1 to wl5) are given in Fig. 1-4 to illustrate the main issues on the wavelength scanning. On the left hand side the evaluation of the output power is shown when the laser starts up and reaches the CW state. All modes are building up, but the mode with the smallest loss in the filter (wl3) will win the competition and suppress the other modes. On the right hand side the evaluation of the output power is depicted when the laser is already on and the filter is switched from wl3 to wl2. In this case wl3 will decrease and wl2 will increase. In this simulation the tuning of the filter is simulated by changing the wavelength dependent losses in the filter comparable to the change in losses when a pass-band of a filter is spectrally shifted. In these figures we immediately see that the laser can switch from one wavelength to another within 50ns. Here we choose to have a filter with a loss difference of 0.06dB in the electric field (0.12dB in power) at 0.05nm from the center wavelength. For a typical parabolic filter shape of an AWG this means a filter with a 0.5nm FWHM (Power).
Fig. 1-4 Simulated output power levels for the different laser modes. (a) The laser starts up with the filter at its center wavelength at wl3. Note that the modes at wl1 and wl5 overlap as well as the modes at wl2 and wl4. (b)The evolution of the modes when the filter switches from wl3 to wl2 at t=0. The legend gives the relative loss values for the different modes. The vertical line marks the 50ns switching time.
The minimum requirement on the tunable filter is that it needs to have a filter shape with a FWHM less than 0.5nm, assuming a parabolic filter shape, and a free spectral range of more than 200nm to suppress all other wavelengths in the neighboring 100nm.
Introduction
17
1.3 Thesis Outline
The further research on the monolithically integrated tunable laser is described in this thesis and is organized as follows:
Chapter 2 is concerned with the active-passive fabrication technology used to realize the monolithic integrated devices presented in this thesis. The use of standard active-passive integration technology of COBRA in the 1700nm wavelength region is discussed, as well as its implications for the passive waveguide performance requirements. The chapter further contains information on the special chip mounting and wire bonding necessary to connect the chip to the electronics and perform measurements.
Chapter 3 is concerned with a study into the performance of the QD amplifiers. A series of QD amplifiers have been fabricated from which the small signal gain was measured. This chapter contains the description of the measurement method and the results obtained for the small signal gain as a function of current density. The results of the fitting of a QD rate equation model to the measured gain spectra are then presented. The change in gain spectrum with increasing injection current density is explained and discussed.
Chapter 4 contains a description of the design, fabrication and characterization of the tunable filters necessary for the tunable laser. A combination of two filters fulfils the required filter characteristics as stated above. These filters can be tuned in the 1600nm to 1800nm region with a series of electro-optic phase modulators. The characterization of the filters is described. First the calibration of the filters and the integrated phase modulators is discussed, followed by the results obtained for the characteristics and tuning of the filters.
Chapter 5 contains a description of a model of the complete tunable laser as designed. Simulations are performed to explore the laser capabilities such as the laser linewidth and the switching speed. The model is based on the QD rate equation model discussed in chapter 3. This model is now embedded in a segmented ring laser model.
Chapter 6 consists of a presentation of the full design and characterization of the monolithically integrated tunable laser. This chapter is builds upon the results presented in chapter 3 and 4. The full tuning capabilities of the laser are presented, as well as a detailed conclusion on the performance and the improvements which can be made. The chapter ends with a demonstration of the laser in a free space Michelson interferometer setup, the first step towards an OCT measurement.
Chapter 1
18
The research presented in this thesis was supported by the IOP Photonic Devices program managed by Agentschap NL (Ministerie van Economische Zaken, Landbouw en Innovatie), Technologiestichting STW and the NRC photonics grant.
19
2 Integration technology
Abstract – In this chapter the active-passive integration technology used to fabricate
the different devices presented in this thesis is discussed. This integration technology based on a generic integrated technology used at COBRA is slightly adapted for the work presented in this thesis. The special design and fabrication considerations made are discussed. The chapter ends with a description of the mounting and bonding of the chip necessary to perform measurements on the chip.
2.1 Introduction
The integration of optical devices on semiconductor material has some major advantages above the use of bulk optics, especially if multiple components can be integrated together on a single chip. The integration can often benefit from advantages such as: compactness, less power consumption, faster, robust, potentially cheaper, and in some cases compatible with electronic integration technologies.
There are however many different integration approaches, using different material systems and even more different fabrication technologies. Some of the main material systems are Silicon (Si) [37], Gallium- Arsenide (GaAs) and Indium- Phosphide (InP) [78]. Each of these material systems has its own capabilities.
For a light source in the 1600nm to 1800nm wavelength region InP would be the first choice. InP has already been used the last couple of decades for the integration of light sources in the 1550nm telecom wavelength region. An InGaAsP layer can be grown lattice matched on InP resulting in a direct bandgap material with an emission wavelength in the 1550nm wavelength region. This emission wavelength in lattice matched “bulk” InGaAsP is however limited to approximately 1670nm [26]. Recent breakthroughs in new types of gain materials, the so called Quantum Well (QW) gain material and the Quantum-dot (QD) gain material [54] make it possible to extend the wavelength region in InP/InGaAsP towards the 1600nm to 1800nm wavelength region. For quantum wells this can be done by the growth of thin layers of non lattice matched (strained) material in the InGaAsP film layer. A proper choice of material and growth conditions allows for fabrication of InAs quantum wells that can emit light with wavelengths up to 2300nm [58]. Quantum-dots on the other hand can also be grown in the InGaAsP film layer. These quantum-dots are grown by metal-organic vapour-phase epitaxy (MOVPE) using self-organized strain induced island formation. The size of these dots, and so also the corresponding emission
Chapter 2
20
wavelength, can be tuned potentially up to emit light at 2000nm wavelength due to the recent development of a method to control the sizes of the quantum-dots through the insertion of ultrathin GaAs interlayers beneath the quantum-dots [54]. These quantum-dots have already been used to fabricate lasers in the 1500nm to 1600nm wavelength region [2].
Both quantum wells and quantum-dots have their own advantages and disadvantages. The advantage of QWs is that a higher gain per unit length can be obtained per well than can be obtained with QDs per dot layer. Also the growth process is relatively straightforward. The gain bandwidth of QWs is however limited to approximately 20 to 40nm which is ideal for narrowband lasers but not for widely tunable lasers of over 100nm bandwidth. Theoretically it may be possible to stack different QWs with different emission wavelengths to broaden the bandwidth, however this has not been explored for such long wavelength quantum wells to our knowledge. QDs on the other hand have a wide gain spectrum due to the inhomogeneous broadening. In the gain material there is a variety of dots with different dot sizes and since each dot size has its characteristic gain wavelength, the QD amplifier can amplify light over a broad spectrum up to 200nm (Chapter 3) [69]. This makes the QDs very suitable for widely tunable lasers. The drawback of QDs is however the low gain per unit length per layer with respect to QWs.
The integration of light sources and other components in the InP/InGaAsP material system has already been explored extensively for the use in telecommunication systems. Large scale integrated circuits have already been successfully brought to the market [51]. The recent activities on the development of a generic integration technology platform, both in Europe and the United states, are expected to boost the integration technology of InP/InGaAsP. The standardization of a limited number of integrated component building blocks and the availability of multi-project wafer runs should also open up the market for smaller companies. The prospects are that the generic integration technology platform will reduce the development costs by at least one order of magnitude.
The tunable laser presented in this thesis uses this standard integration technology developed at COBRA. It relies on the experience COBRA has with the so called “active-passive” integration technology used for photonic integrated circuits in the 1550nm telecom wavelength region. With the active-passive integration technology is meant the integration of active components together with passive components on a single chip. In this technology one is able to combine active components that generate, amplify or absorb light, with passive components which are transparent for light at 1550nm and are used to manipulate the light. For the use of this integration technology for components operating in the 1600nm to 1800nm wavelength region two major aspects have to be explored: the integration of long wavelength QDs in the active regions and the applicability of standard passive components in the 1600nm to 1800nm wavelength region. The QDs themselves have already successfully been integrated in the COBRA active-passive integration technology [75] and their applicability will be discussed in this chapter.
This chapter continues in section 2.2 with a description of the fabrication technology starting with the fabrication of the active-passive wafer followed by a step-by-step
Integration technology
21
description of the active-passive processing scheme used. Section 2.3 describes the adaptations made to the standard integration technology necessary to fabricate the tunable laser in the 1600nm to 1800nm wavelength region. The chapter ends with a description of the mounting of the chip before measurement.
2.2 Fabrication Technology
The fabrication of a semiconductor integrated optical device can basically divided into two main parts: the fabrication of the semiconductor wafer with a specific layerstack and the fabrication of the optical devices on the wafer. In this section the active-passive layerstack is discussed first, as well as the fabrication of the active-passive wafer. Then the processing scheme that was used to fabricate the tunable laser is explained in detail.
2.2.1 Active-passive layerstack
The complete monolithically integrated tunable laser consists of active layerstack components (semiconductor optical amplifiers (SOAs)) as well as passive layerstack components (waveguides, arrayed waveguide gratings (AWGs), multi-mode interferometers (MMIs) and phase modulators (PHMs)). To be able to integrate these active and passive components on a single chip, a single wafer with active areas and passive areas is required. These active-passive wafers where fabricated at COBRA, in collaboration with the MiPlaza division of Philips Research, using the butt-joint integration approach [11]. (see Table 2-1) In this integration approach, based on MOVPE, the active layerstack is grown first on an n-doped InP substrate, ending with the 20nm not-intentionally-doped (n.i.d.) InP layer on top of the InGaAsP film layer (Q1.25) (performed at COBRA). Within the Q1.25 waveguiding film layer five InAs QD layers are stacked with an ultrathin GaAs interlayer underneath each QD layer to control the size of the dots [2]. The QD are optimized to emit light in the 1600 to 1800nm wavelength region, as presented in [28][54]. The QD layers are 40nm separated from each other with InGaAsP (Q1.25). Furthermore, the film layer also contains a 10nm n.i.d. InP etch stop layer 20nm underneath the QD layers to be used in the etch-back process. The active areas are then defined in a SiNx layer
by means of a photolithography process and the layerstack is etched back down to the InP etch-stop layer 20nm underneath the QD layers (performed at COBRA). In the first re-growth step (performed at Philips) the passive InGaAsP (Q1.25) film layer is grown ending with a 20nm n.i.d. InP layer. After the removal of the SiNx mask a common
cladding is grown in the second re-growth (performed at Philips). The final layer structure is shown in Table 2-1.
Chapter 2
22
Passive Layerstack Active Layerstack
Layer Material Doping
[cm-3] Thickness [nm] ColorcodeColorcode Thickness [nm] Doping [cm-3] Material Layer
E p i 3
E3-6 p-InGaAs 1.51019 280 280 1.51019 p-InGaAs E3-6
E p i 3 E3-5 p-Q1.4 7.21018 10 10 7.21018 p-Q1.4 E3-5 E3-4 p-Q1.2 4.71018 10 10 4.71018 p-Q1.2 E3-4
E3-3 p-InP 11018 1000 1000 11018 p-InP E3-3
E3-2 p-InP 51017 300 300 51017 p-InP E3-2
E3-1 i-InP n.i.d. 180 180 n.i.d. i-InP E3-1
E p i 2
E2-2 i-InP n.i.d. 20 20 n.i.d. i-InP E1-8
E p i 1 E2-1 i-Q1.25 n.i.d. 330 100 n.i.d. i-Q1.25 E1-7
5x40 n.i.d. QD/i-Q1.25 E1-6 20 n.i.d. i-Q1.25 E1-5 10 n.i.d. i-InP E1-4 E
p i 1
E1-3 i-Q1.25 n.i.d. 170 170 n.i.d. i-Q1.25 E1-3 E1-2 i-InP n.i.d. 70 70 n.i.d. i-InP E1-2 E1-1 n-InP 51017 430 430 51017 n-InP E1-1
E1-0 n-InP 1-41018 ≈350µm ≈350µm 1-41018 n-InP E1-0
Table 2-1 Active-passive layerstack. The thick lines show the different growth steps: First the active layerstack is grown on the substrate E1-layers. Then the passive areas are selectively etched back till the InP stop layer (E1-4) in the film layer. In the second growth step the passive waveguide layers are grown (E2-layers). In the third growth step the common cladding layers and contact layers are grown (E3-layers).
2.2.2 Active-passive processing scheme
The active-passive integration technology used at COBRA limits the number of different components/waveguide structures that can be used on a single chip. These components are passive waveguides, PHMs and SOAs. The passive waveguides and the PHMs can be fabricated as shallowly etched or deeply etched, whereas the SOA can only be shallowly etched. A shallowly etched component has a low contrast ridge waveguide where the area around the ridge is etched away down to 100nm into the film layer. In a deeply etched waveguide this area around the waveguide is etched away all the way through the film layer. Thus a high contrast ridge waveguide is formed that is suitable for small bend radii. For the tunable laser it was decided to use only the shallowly etched waveguides, PHMs and SOAs to minimize the complexity of the processing. In the
Integration technology
23
following step-by step description of the processing scheme diagrams are added in Fig. 2-1 to illustrate the process. In these pictures the different components are presented from left to right; SOA, PHM, passive waveguide with electrical lead crossing it, bond pad and an electrical isolation waveguide at the far right-hand side. Note that the dimensions of the picture are not realistically scaled.
a) Optical waveguide lithography - A 50nm thick SiNx layer is deposited on the wafer
using a PECVD process. The optical waveguide pattern is defined in a 750nm thick layer of positive photoresist (HPR504) using optical contact lithography. This pattern is transferred to the SiNx layer using a CHF3 reactive-ion etching (RIE) dry etch
process. After removal of the photoresist the optical waveguide pattern remains in the 50nm SiNx which is used as a hard-mask in the subsequent InP RIE etching steps.
b) First InP RIE etch step - In the first InP RIE dry etch step the height difference
between the top of the isolation sections in the waveguides and the etch depth of the shallow waveguides in the film layer is realisedusing an optimized CH4/H2 RIE
process alternated by an O2-descum process. In this etch step also the estimated lag
effect has to be taken into account. This is an effect which results in small etch depth differences between areas directly beside a ridge waveguide and areas further from the waveguide where the etch depth is measured. The aiming etch depth is 380nm (200nm top cladding, 100nm film layer and 80nm lag effect).
c) Isolation section lithography - The SiNx mask is removed from the areas where
isolation sections have to be made in the optical waveguides. First the isolation sections are defined in a 3µm thick layer of positive photoresist (AZ4533) using optical lithography (dark field mask). Afterwards the SiNx is etched away wet
chemically using a buffered hydrofluoric acid (BHF) solution.
d) Second InP RIE etch step - In the second InP RIE etch step the height difference
between the top of the passive waveguides and the top of the isolation sections in the waveguides is made using the same etch process as the first InP RIE etch step. Here it is important to realize that at the end of the thirds RIE etch the top of the passive waveguides must be underneath the highly doped InGaAs contact layer to minimize the waveguide losses. This means in this second RIE etch step the etch depth should be a little less than the difference between the top of the passive waveguide and the isolation section. In this etch step the aim is to etch 1200nm. The top of the passive waveguides is then 100nm underneath the contact layer.
e) Contact layer lithography - The SiNx mask is removed from the passive waveguides
which do not need the InGaAs contact layer for metal contacting. It is removed by defining the contact areas in a 3µm thick layer of positive photoresist (AZ4533) using optical lithography and etching away the SiNx wet chemically from the passive
Chapter 2
24
f) Third InP RIE etch step - In the third and last InP RIE etch step the height difference
between the top of the contact layer and the top of the passive waveguides is made using the same etch process as the first and second InP RIE etch steps. More important is the final etch depth of the shallow waveguides in the film layer which has to be 180nm (including lag effect) which means it is 100nm near the ridge. The expected etch step is 400nm however the etch depth of the shallow waveguides determines the final etch step. After removing the remaining SiNx mask wet chemically using a BHF
solution the waveguide topology is finished.
g) Planarization - The wafer is planarized by spinning six layers of polyimide (PI2723)
which are cured after each layer at a temperature up to 325C. The polyimide is first etched back using a CHF3/O2 RIE dry etch process till approximately 100nm above the
top of the contact layer. The areas where contact openings have to be made in the polyimide are defined in a 2.5µm thick layer of negative photoresist (MaN-440) using optical lithography. Finally the polyimide is selectively etched further using the CHF3/O2 RIE dry etch process till approximately 100nm underneath the top of the
contact layer.
h) Metallization - The metallization pattern is defined in a 2.5µm thick layer of negative
photoresist (MaN-440) with a lift-off profile using optical lithography. To clean the surface of the contact layer to get a low contact resistance, the top 100nm of the contact layer is etched away wet chemically using H2SO4:H2O2:H2O in proportions
1:1:100. Then the Ti/Pt/Au 60/75/300 nm metal contacts are deposited by means of e-beam evaporation and lift-off in acetone. The backside of the wafer is metalized with evaporated Ti/Pt/Au 60/75/200 nm to create one common n-contact. No waver thinning process was employed before this backside metallization. The wafer is annealed at 325C for 30s. More than one annealing cycle at 325C for 30s did not improve the contact resistance, only increased the risk of defects in the p-metal.
i) Plating - To assure a uniform current injection in the long SOAs, the thickness of the
p-side metal contacts on the SOAs is increased using an electroplating process. For this process first a 100nm Au seed layer is sputtered on the wafer. Then the plating pattern is defined in a 3µm thick layer of positive photoresist (AZ4533) using optical lithography (dark field plating mask). A 1.7µm thick layer of gold is deposited on the SOA contacts using an electroplating process. After the photoresist has been removed the metallization pattern is again defined in a new 3µm thick layer of positive photoresist (AZ4533) using optical lithography (same mask as for the p-side metallization). The seed layer in between the p-metal contacts is now removed by etching the gold layer in a potassium cyanide (KCN) solution. The reason to selectively etch back the seed layer is to keep the seed layer on the unplated PHMs and their leads to the bonding pads to improve the reliability of the electrical connection when crossing a height difference in the polyimide.
Integration technology
25
j) Cleaving - Finally the approximately 350-360µm thick chip was cleaved out of the
wafer and mounted on a copper chuck using an electrical and thermal conductive epoxy resin. Further details of this mounting are given in the section on chip mounting at the end of this chapter.
SOA PHM WG PAD Iso
a
SOA PHM WG PAD Iso
b
SOA PHM WG PAD Iso
c
SOA PHM WG PAD Iso
d
SOA PHM WG PAD Iso
e
SOA PHM WG PAD Iso
f
SOA PHM WG PAD Iso
g
SOA PHM WG PAD Iso
h
SOA PHM WG PAD Iso
i
SOA PHM WG PAD Iso
j
Fig. 2-1 Different processing steps. The different components are presented from left to right; SOA, PHM, passive waveguide with electrical lead crossing it, bond pad and an electrical isolation waveguide at the far right-hand side. The used color codes are as presented in Table 2-1.
(a) Optical waveguide lithography (b) First InP RIE etch step (c) Isolation section lithography (d) Second InP RIE etch step (e) Contact layer lithography (f) Third InP RIE etch step (g) Planarization (h) Metallization (i) Plating (j) Cleaving. Note that the dimensions of the picture are not realistically scaled.
Chapter 2
26
2.3 Specific processing used in this thesis
The active-passive layerstack and the processing scheme described above are used for the realization of the tunable laser. They are based on the experience of previous chip fabrication at COBRA. In this section the differences compare to the standard integration technology will be discussed.
2.3.1 Waveguide structure at 1700nm
The standard integration technology described above is optimized for the use in the 1550nm wavelength region. The layerstack consists of a 500nm waveguide core layer to confine the optical mode in between the InP top and bottom cladding. In the standard integration technology the waveguides are fabricated to be 2µm wide and the area around the waveguides is etched 100nm into the film layer to confine the light in the lateral direction. Using these waveguides in the 1600nm to 1800nm wavelength region has some influence on the mode size and the waveguide losses. In this subsection we describe what we can expect and what influence it has on the waveguide losses.
The waveguide wide of 2.0µm at 1550nm is chosen just underneath the 3rd order
lateral mode cut-off width. The existence of the antisymmetric 2nd order lateral mode in the
waveguide is not considered problematic. It has considerably higher losses than the lowest order mode due the higher intensity of the fields at the edges of the waveguide. And the antisymmetric field distribution couples less easily to the fundamental mode than the 3rd order mode. For the use of these waveguides in the 1700nm region the width is
increased to 2.2µm. The width is chosen as large as possible without enabling higher order modes to minimize the overlap of the fundamental mode with the rough side walls.
The intrinsic material losses changes proportionally to the change in refractive index of the material. For the InGaAsP (Q1.25) film layer this change in refractive index is from 3.364 at 1550nm to 3.354 at 1700nm (-0.8%) [24]. This change is due the increasing energy distance from the absorbing band edge in the material and thus the optical absorption of the material is decreasing. But this is almost negligible when the loss at 1700nm is compared to that at 1550nm.
The other difference is the change in mode size. The change from 1550nm to 1700nm resulted in an increase in mode size due to the longer wavelength. This results in a larger overlap of the optical mode with the doped cladding layers. Especially the p-doped top cladding has a high loss as can be seen in Table 2-2 [14][77]. To estimate waveguide losses due to doping in the cladding layers we calculated the confinement in each of the layers. These confinement factors are given in Table 2-3 for a 2.0µm waveguide at 1550nm and 1700nm and for a 2.2µm waveguide at 1700nm. The total waveguide losses can be calculated with:
Integration technology 27 i i i n p , (2-1)
Here αp,n is the total waveguide losses, Γi the confinement in the ith layer and αi the
waveguide losses in layer i due to doping. The calculated waveguide losses due to doping in the different layers are also given in Table 2-3 for the different configurations.
Doping type Doping level
[cm-3] [dB/cm] Loss αi 1550nm Loss αi [dB/cm] 1700nm p 1e18 99.0 128.2 p 5e17 49.5 64.1 p 1e16 (n.i.d.) 1.0 1.3 n 1e16 (n.i.d.) 0.1 0.1 n 5e17 5.8 7.0 n 2.5e18 42.5 51.1
Table 2-2 Material losses for different p and n doping levels calculated [14] for 1550nm wavelength and 1700nm wavelength. For not-intentionally-doped (n.i.d.) layers a 1e16 cm-3 doping level is
estimated. Layer Doping [cm-3] Thickness [nm] Confinement Γi 2.0µm 1550nm Loss Γiαi [cm-1] 2.0µm 1550nm Confinement Γi 2.0um 1700nm Loss Γiαi [cm-1] 2.0µm 1700nm Confinement Γi 2.2um 1700nm Loss Γiαi [cm-1] 2.2µm 1700nm p-InP 1e18 (p) 1000 0.0036 0.0821 0.0060 0.1760 0.0064 0.1876 p-InP 5e17 (p) 300 0.0276 0.3151 0.0347 0.5120 0.0362 0.5346 i-InP 1e16 (p) 200 0.0863 0.0197 0.0908 0.0268 0.0936 0.0276 i-InGaAsP 1e16 (n) 500 0.7318 0.0110 0.6912 0.0124 0.6917 0.0125 i-InP 1e16 (n) 70 0.0573 0.0009 0.0599 0.0011 0.0587 0.0011 n-InP 5e17 (n) 430 0.0859 0.1042 0.1042 0.1668 0.1010 0.1617 n-substrate 2.5e18 (n) - 0.0072 0.0703 0.0129 0.1523 0.0121 0.1428 Total losses [dB/cm] 2.7 4.6 4.6
Table 2-3 Calculated confinement and loss components in the different layers for a 2.0µm waveguide at 1550nm and 1700nm and for a 2.2µm waveguide at 1700nm
Furthermore the increase in mode size also results in an increase in overlap of the optical mode with the etched side walls. This causes increasing waveguide losses as well, due to light scattering on sidewall roughness. The estimated increase in losses due to sidewall roughness has not been calculated.
If we assume that the waveguide losses are mainly due to doping and scale comparably with the calculated losses due to doping, an increase of approximately 75% in the waveguide losses can be expected in a 2.2µm waveguide at 1700nm in comparison to the 2.0µm waveguide at 1550nm.
Chapter 2
28
2.3.2 Top buffer layer
In the active-passive layerstack growth process the first epitaxial growth ends with an InP layer (E1-8 in Table 2-1) on top of the InGaAsP film layer (E1-7 in Table 2-1). In active-passive layerstacks that were used previously in our research institute, this first epitaxial growth ended with a 200nm p-doped (31017 cm-3) InP buffer layer above the
InGaAsP film layer [4][20][34]. The p-doping in this buffer layer was included to improve the current flow to the active regions. After the etch back and the selective area re-growth of the passive film layer, a 200nm n.i.d. InP buffer layer was grown on top of the passive film layer. This regrown layer is n.i.d. to minimize the passive waveguide losses.
In the active-passive wafer fabricated for the tunable laser we chose to end the first growth with a 20nm n.i.d. InP layer above the film layer (layer E1-8 in Table 2-1). This layer is grown this thin to minimize the height of the selectively re-grown passive film layer. This in turn reduces the height variations that appear at the active-passive butt-joint. In the second re-growth (layers E2 in Table 2-1) this 20nm InP is completed with an extra 180nm n.i.d. InP layer (layer E3-1 in Table 2-1) to end up with a 200nm n.i.d. InP buffer layer above the film layer for both the active and the passive areas. This n.i.d. buffer layer is also chosen to minimize the waveguide losses in the waveguides. The fact that the buffer layer above the amplifiers is now also n.i.d. can result in slightly higher voltage over the amplifiers during operation when compared to previously used wafers with a lightly p-doped cladding.
2.3.3 Doping passive waveguides
For the passive layerstack one has to make a tradeoff between minimum waveguide losses and optimal phase shift efficiency in the phase shifters. An n-doping in the film layer results in a higher phase shift efficiency [10][82][83] due to the increasing carrier effects as will be discussed in chapter 4. However, this n-doping in the film layer also results in a higher optical loss due to the increasing free carrier absorption as can be seen above. The calculated film layer loss increases from 0.08dB/cm for a 11016cm-3 doping level to
0.59dB/cm for a 61016cm-3 doping level (1700nm)[14]. Measured phase shift efficiencies
for doping concentrations of 61016cm-3 vary between 0.2rad/Vmm [35] and 0.4rad/Vmm
[82]. The decrease in phase shift efficiency expected when reducing the doping in the film layer from 61016cm-3 to 11016cm-3 is however very small [83].
Because of the long path lenghts of passive waveguides in the tunable laser circuit we chose to have a 11016cm-3 n-doping in the film layer to reduce waveguide losses. The
expected small decrease in phase shift efficiency can be compensated with slightly longer phase modulators (chapter 4).
Integration technology
29
2.3.4 InP etch stop layer in the film layer
During the active-passive wafer fabrication the film layer is selectively etched-back after the first growth in the areas where passive waveguides have to be realized (section 2.2.1). To have a uniform and precisely defined etch back over the complete etch-back area, a 10nm InP etch-stop layer is included in the film layer, 20nm underneath the QD layers where the etch-back has to stop. The selective wet-chemical etch-back will stop on this InP etch-stop layer. The InP etch stop layer is removed in the passive area during the bromine-methanol cleaning step prior to the regrowth. The influence of this layer on the optical field is assumed to be negligible. The electrical influence is not known exactly; however, most probably the carriers will tunnel through this 10nm InP layer.
2.3.5 Planarization and etch back Polyimide
After the etching of the waveguide structures, the wafer needs to be planarized before metal evaporation. For the planarization a method is used that consists of spinning six layers of polyimide alternated with a curing at 325C and a surface treatment as described above in section 2.2.2(g). The advantage of applying six layers over the method that is often used with one or two layers of polyimide is the uniformity of polyimide. After six layers the height variation is reduced to approximately 150nm whereas these variations after two layers can be more than 500nm. These height variations need to be less than 200nm to be able to open all contacts approximately at the same time when etching back the polyimide.
Another important advantage of these six layers is the surface roughness of the polyimide after etching back. The roughness of the surface is proportional to the etch-back time. In case of six layers of polyimide this etch time is approximately 35minutes. With two layers of polyimide applied 3minutes can be enough which leaves a much smoother surface. To be able to make a good adhesion between the metallization and the polyimide the surface of the polyimide should be rough. Previous chips fabricated with only one or two layers of polyimide or BCB did not have this strong adhesion between the metal and the polyimide which resulted in release of contact pads from the polyimide during the bonding process or during measurements with probe needles. Various chips with these six layers of polyimide did not show this release of contact pads.
The tunable laser was fabricated using the six layers of polyimide. For the separate QD amplifiers fabricated on a different wafer for the QD gain measurements described in chapter 3 only two layers of polyimide where used. The six polyimide layers where not necessary in this case. This was due to the absence of separate probe pads and the uniform distribution of the amplifiers on the wafer resulting in smaller height differences.
Chapter 2
30
2.3.6 Metallization
For the metallization a distinction has been made between the metal contacts on the amplifiers and the metal contacts on the PHM and the associated bonding pads. The metal contacts of the amplifiers need to be plated to minimize the electrical losses in the contacts [34]. However, the PHM and their associated bonding pads do not need this plating due to the low current through the PHM in reverse bias. Furthermore, bonding is preferably done on non-plated pads due to the fact that plated gold contains more residuals from the plating bath. These residuals make the adhesion of a bonding on plated gold more difficult. For this reason, only the amplifiers where fabricated with a plating on top of the evaporated Ti/Pt/Au.
During the fabrication of the tunable laser the gold seed layer has not been completely removed after plating but has been etched away selectively with the p-contact metallization mask. This introduces an extra lithography step (not an extra mask) but has the advantage that the seed layer remains on the PHM contacts and their contact pads. This prevents that part of the evaporated gold is etched away when etching away the seed layer. Furthermore it improves the electrical leads from the PHM to the contact pads especially where a step has to be made between different heights in the polyimide. The seed layer improves this due to the tilted evaporation of the seed layer whereas the evaporated Ti/Pt/Au layers are evaporated right from the top.
2.4 Chip mounting for measurement
The tunable laser described in this thesis has to be controlled with 36 voltage sources connected to the PHMs as will be described later in chapter 6. The most common connection method used in a research environment for such number of connections is by means of a multi-probe connection. The multi-probe is typically placed directly on the gold connection pads on the fragile chip. To avoid the need for direct probing, which easily damages the fragile chip, a wire bonding connection method was used for the measurements described in this thesis. The optical chip is glued on a copper mount together with a printed circuit board (PCB). The PHMs on the chip are connected individually via the bond pads on the chip to bond pads on the PCB by means of wedge-bonding. On the PCB high bandwidth (1GHz) multi-pin connectors, each connecting eight voltage signals, are positioned to connect the laser to the electronics [70]. A picture of the measurement setup is given in Fig. 2-2.
The PCB is designed to transport the voltage signals from the connectors to the bond pads on the PCB. The electrical leads from connectors to bond pads on the PCB have been shielded to avoid electrical crosstalk at 100MHz (maximum) operation. An 8-layer PCB is designed to transport all electrical signals to the small (10mm wide) connection area close to the optical chip. Two layers contain the signal lines from the connectors to the individual
Integration technology
31
bond pads. The individual ground signals from each of the signal lines are carried by two other layers in the PCB. These ground signal lines all come together in a common ground bond pad in the bottom layer of the PCB directly underneath the bond pads. This common ground pad is glued to the copper mount and thus connected to the p-side of the optical chip. The four other layers on the PCB are used as vertical shielding layers in between the signal and ground layers to avoid electrical crosstalk. The vertical crosstalk between signal and ground leads is minimized by alternating these signal and ground leads with shield leads connected to the common shielding. The PCB frequency response capabilities have not been tested separately.
Fig. 2-2 Image of measurement setup. The setup contains the following components: 1) Tunable laser chip (within dotted lines). 2) Printed circuit board. 3) Bond pads on PCB to connect the contacts of the PHMs on the laser chip to the PCB. 4) Multi-pin connectors to connect the control electronics of the PHMs to the PCB. 5) Probe needles to inject current in the three QD amplifiers. 6) Fiber stages to collect the light from the laser chip with lensed fibers.
The copper mount is fixed on a measurement setup. Within this measurement setup cooled temperature stabilized water can be guided just underneath the copper mount to remove heat generated in the chip. The optical chip can be approached with lensed fibers from both sides of the chip to collect the light from the cleaved facets.
The described connection method of the optical chip to the electronics was used successfully for the tunable laser over more than 4 months. This demonstrates that this connection method is very stable and robust for the measurement of optical chips with a large number of electrical connections.
1 2 3 4 4 5 5 5 6 6
33
3 Measurement and analysis of the gain in
quantum-dot amplifiers
Abstract - Small signal modal gain measurements have been performed on
two-section ridge waveguide InAs/InP (100) quantum-dot amplifiers that we have fabricated with a peak gain wavelength around 1700nm. The amplifier structure is suitable for monolithic active-passive integration and the wavelength region and wide gain bandwidth are of interest for integrated devices in biophotonic applications. A 65nm blue shift of the peak wavelength in the gain spectrum has been observed with an increase in injection current density from 1000A/cm2 to 3000A/cm2. The quantum-dot amplifier gain
spectra have been analyzed using a quantum-dot rate-equation model that considers only the carrier dynamics. The comparison between measured and simulated spectra shows that two effects in the quantum-dot material introduce this large blue shift in the gain spectrum. The first effect is the carrier concentration dependent state filling with carriers of the bound excited and ground states in the dots. The second effect is the decrease in carrier escape time from the dots to the wetting layer with decreasing dot size.
3.1 Introduction
The behavior of semiconductor quantum-dots (QDs) has been studied extensively over the last fifteen years. The use of QDs in monolithic optical devices, such as semiconductor optical amplifiers (SOAs) and semiconductor lasers, can have a number of benefits over the use of bulk or quantum well gain material. These advantages are due to the three dimensional carrier confinement in QDs. Some of the benefits that have been experimentally demonstrated are a comparatively low lasing threshold current density [46] and ultra fast gain recovery in optical amplifiers [12][13]. Another major advantage of QDs over bulk or quantum well material is the wavelength tunability. The wavelength of a generated photon depends first of all on the discrete energy levels in the quantum-dot and can be changed only slightly due to the temperature dependent homogeneous broadening [62]. The discrete energy levels in the QD are however directly dependent on the size (and shape) of the dots [36] which means that the distribution of the size of the dots in an amplifier determines to a large extent the gain spectrum. In this chapter we present results on InAs QD on an InP (100) substrate. The average QD size in this system can be controlled in the growth process which makes it possible to tune the average wavelength
Chapter 3
34
[2][28][54]. Most results that have been published up to now on this material are in the 1550nm telecom region where wide gain spectra (e.g. 80nm wide) have been observed. However, the average dot size can also be tuned to provide gain material which has a centre emission wavelength in the 1700nm region or even longer wavelength regions.
Such wavelengths in combination with a wide gain bandwidth are desirable for other applications such as tunable lasers for gas detection [79] or optical coherence tomography [23]. This extension of the wavelength range of an amplifier structure that can be used in active-passive optical integration schemes [7][75] opens up the possibility to develop integrated light sources for these applications.
In order to be able to use this gain material optimally in devices for applications in these longer wavelength ranges, we have measured the small signal gain spectra of InAs/InP(100) ridge waveguide amplifiers as a function of injection current and modeled this QD material to understand its behavior. This can be utilized for instance for increasing the gain bandwidth of the optical amplifier by using multiple amplifier sections which are operated at different injection current densities [80].
In this chapter we start with presenting the measurement method, the InAs/InP QD devices used and the characterization results in the 1600nm to 1800nm region. Then the QD amplifier model that was used to explain the dependency of the observed gain spectra on injection current is presented. The modeling of QD materials is more complicated than the modeling of bulk or quantum well gain material due to the inhomogeneous character of the QD gain material. Extensive QD models have been presented especially for InxGa1-xAs/GaAs QD materials [25][63] but also for InAs/InP QD materials [29]. Here we
have simplified a commonly used QD amplifier model to calculate the small signal gain spectrum of the amplifiers and to fit a number of the model parameters to the experimental data. Three important parameters are the electron-hole transition energies of the wetting layer (WL), the excited state (ES) and the ground state (GS). From InAs QD on GaAs substrate these transition energies can be determined with photoluminescence (PL) measurements or comparable measurement techniques. In these PL measurement results two individual peaks can be observed, one for the GS and one for the ES. PL measurements from InAs QD on InP substrate show however only one wide individual peak and no feature is observed that can be attributed to the wetting layer. The peaks from the GS and ES do overlap, which makes it impossible to extract the transition energies of the dots and the WL with this PL measurement technique [1][18]. Using the model we have been able to determine the transition energies of the GS and ES and an effective transition energy for the wetting layer.
The fitted results are discussed in the last section together with the explanation that was found for the large shift in peak wavelength in the gain spectrum with the change in injection current density.
