Motive – the Temperature Effect

In order to further realize the room temperature operation for advanced QD-based devices, an indispensible work of understanding the temperature effect on the QD systems turns out to be necessary. As non-zero thermal energy being naturally inventible, we must factor in its influence as considering both the device performance and reliability; it is especially more crucial for devices fabricated with quantum structures since the related energy of the systems is often in the order of eV, or even smaller to be meV; therefore, the investigation of temperature effect remains one of the most valuable topics in studying semiconductor QDs.

As a matter of fact, the role of temperature has been widely studied from different aspects. For instance, there are discussions which focus on the temperature effect directly on device operation [20, 21]. As for the researches on mechanisms, non-radiative recombination at high temperature being an unwanted path to deplete carriers is demonstrated to be significant in self-assembled QDs [22-24]; also, both steady-state and time-resolved studies show that carrier distribution will form local equilibrium among QDs as temperature rises [23, 25-27], and quantum tunneling is as well shown to be favored by thermal energy [28]. Moreover, the thermalized carriers can undergo intraband transition [29], and even populate optically inactive states which results an increase in carrier lifetimes of QDs [30]. The observation of these phenomena let us further picture the world of QDs and we accordingly carry out the study of this work.

Chapter 2

Spectroscopy

Before entering our main studies, we give essential introductions to the spectroscopy techniques and InAs QDs samples studied in this work. The spectroscopy covers steady-state and time-resolved determinations, where the two both give important information but from different aspects. On the other hand, the structure and morphology of the six studied samples are introduced so that one can have the central view on the InAs QDs in our study.

2.1 Carrier Dynamics in Semiconductor QDs

When it comes to direct-bandgap semiconductors, the luminescence spectroscopy remains one of the most important approaches to investigate the intrinsic behaviors of carriers in the system interested, and the behaviors are so worth studying since they are the key principles closely related to further device operation.

Fig. 2-1 schematically shows the three major procedures of carrier dynamics in QDs when electron-hole pairs (excitons) are generated by an external excitation:

capture, relaxation and recombination. After firstly captured by QDs, carriers may undergo a rather fast process to relax from higher quantum states to lower ones with the emission of phonons; also, the separated electrons and holes may have a chance to recombine, in this case, generating photons. Studies have found that the rate of relaxation and recombination process are usually quite different, typically being in the

-1 -1

Figure 2-1: Typical procedures for carrier transitions in semiconductor QDs.

2.2 Steady-State Photoluminescence (SSPL)

The steady-state luminescence spectrum is the fundamental spectroscopy for semiconductors which determines the energies of the emission photons generated from the recombination process. As shown in Fig. 2-2, photons with energy higher than the one of band gap (Eg) are input to pump the electrons from V.B. to C.B.; the holes and electrons relax to lower states and then recombine to generate PL. For the measurement of this kind, the detection focuses on different luminescence energies and is in steady-state condition. The excitation will generally be a photon source such as a laser because of the experimental convenience, giving the name to the analysis as photoluminescence; however, the excitation source can vary to be others like a bias voltage (electroluminescence) or a cathode source (cathodeluminescence).

Figure 2-2: Schematic plot of the principle of PL generation.

2.3 Steady-State Photoluminescence Excitation (SSPLE)

In addition to SSPL, the SSPLE is another useful steady-state spectroscopy to determine the energy levels in nano-structures. The major difference between the SSPL and SSPLE is the energy of the excitation source (laser). For SSPL, the photon energy of the laser is usually fixed while in SSPLE an energy-tunable photon source is used. It is well known that the input photons will only be absorbed by electrons in valence band to jump to conduction band when the energy of the input photons matches the energy difference between the two quantum states. Therefore, we can determine the energy levels by varying the energy of the input photon and monitor the PL intensity.

2.4Time-Resolved Photoluminescence (TRPL)

Unlike steady state measurement whose spectrum is resolved in frequency domain, time resolved photoluminescence deals with the spectroscopy in time domain, or the time dependence of photon generation. Imagine a specially-made ball whose color will change every millisecond. If one wishes to record all the colors that it reveals within a long time interval by a camera, the time interval between two shots of the camera must be less than one millisecond, or otherwise, some information will definitely be lost. Similarly, only by using a detection method with detection speed faster than it of the dynamic event itself can we ensure that the dynamic behaviors are completely recorded. In order to detect the event of carrier relaxation (picosecond order) and recombination (nanosecond order), the time resolution of the measurement techniques must be at least comparable (or faster) to them. Here we introduce the main principles of the three most commonly used methods for studying time-resolved spectroscopy in semiconductor nanostructures.

Pump-probe spectroscopy

The pump-probe spectroscopy is basically the application of optics. As shown in Fig. 2-3, a laser pulse is initially divided into two, the “pump” and “probe” beams, and a time delay Δt will be deliberately given between the two; the pump bean is targeted to excite the sample, while the other is used for probing the dynamic signal which will be collected by a detector. By adjusting Δt, one can get the time-dependent spectrum, which can be reflectivity, absorption or luminescence.

A commanding advantage of this approach will be the time resolution; ideally, its resolution is only determined by the pulse width of the laser [33], and by using the

laser with ultra-short pulse width, one can easily reach to an ultra-fast detection capability of femtosecond order or even higher.

Figure 2-3: Main principle of the pump-probe technique.

(Source: Z. Sun,” Time Resolved Pump-Probe Spectroscopy.”)

Streak Camera^[34]

A typical streak camera system is shown in Fig. 2-4. The signal generated by the sample will be focused to one photocathode to excite photoelectrons. The photoelectrons then are accelerated by an ultra high voltage to enter the sweep field, which provides a time-varying voltage. Finally the photoelectrons are reached on a screen for detection.

With this approach, one can directly get the data in a large detection range and it is very useful in some research aspects. Nonetheless, its time resolution can only be reached to the order of picosecond due to the limitation of the sweep rate of electronic devices.

Figure 2-4: Main principle of the steak camera technique.

(Source: “Guide to Streak Cameras,” Hamamatsu Photonics.)

Time Correlated Single Photon Counting ^[35]

The major principle of Time Correlated Single Photon Counting (TCSPC) is a fairly smart one, mainly making use of the concept of statistics. Referring to Fig. 2-5, the measurement is proceeded by averagely allowing one or no photon between two laser pulses. When a photon appears, the time interval (Δt) will be recorded to complete a detection event, and the procedure will carry on repeatedly. Although Δt may be different from every event, one can get a statistically reliable result after a large quantity of repeated detections, as shown in Fig. 2-6.

Generally, the time resolution of TCSPC method can be reached to the order of picosecond, usually limited by the speed of the detectors. A great advantage about TCSPC method is the capability of detecting signal with low intensity by choosing suitable detectors such as photomultipliers (PMTs) or avalanche photodiodes (APDs);

furthermore, the system is quite simple so that it is possible to additionally assemble it on the existing-experimental system by only adding several instruments.

Figure 2-5: Main principle of the TCSPC technique.

(Source: Reference. 35.)

Figure 2-6: Final statistical integration of signal in the TCSPC technique.

(Source: W. Becker, A. Bergmann,” Detectors for High-Speed Photon Counting.”)

2.5 Experimental Setup in This Work

Steady-state PL

The steady-state luminescence spectra are obtained by PL in this work, and the apparatus is schematically drawn in Fig. 2-7. The sample is placed in a low-temperature (low-T) chamber with a closed-cycled helium cryostat used for temperature control; to increase the signal-to-noise ratio, a chopper and lock-in amplifier are applied. A diode laser (PicoQuant LDH-P-780, peak wavelength at 780 nm) is used as an external excitation source, and after passing through few reflection mirrors, the laser hits the sample and makes it generate spontaneous emission to be focused back by another two lens into a monochromator (JobinYvon iHR-550). The luminescence is then resolved by a 1200 gr/mm grating and collected by an InGaAs solid-state detector whose response wavelength is from about 800 to 1500 nm (near infrared). Notice that the monochromator carries two exits and a guiding mirror is arranged to guide the light entering one of them.

Figure 2-7: Schematic plot of the steady-state PL and time-resolved PL system.

Time-resolved PL

The TRPL measurement is carried out by the TCSPC technique mentioned in the last chapter. The system is assembled additionally on the original set-up of the SSPL and the two actually share many common apparatus. The same diode laser is used for excitation except that the repetition rate is kept at 10 MHz in TRPL measurement while it is irrelevant in steady-state PL; moreover, both the laser beam and emitted luminescence undergo the same optical alignment as in the steady-state situation. As for the different aspects, first, the chopper and lock-in amplifier are not used and because of the dynamic measurement of TRPL, the detector must be able to “count”

the signal as well, so an InGaAs photomultiplier tube (PMT, Hamamatsu H10330-75) is applied here. After selecting the detected wavelength, the guiding mirror in the monochromator is set to guide the light to the exit facing the PMT. Another difference resides in the final signal processing done by many delicate electronic devices, introduced as follows.

Beside the general laser beam used for excitation, an additional pulse signal is sent as a synchronization (SYNC) trigger and directly enters a constant fractional discriminator (CFD) to be a reference signal; the CFD acts as a threshold level to filter out the noise and judge whether the signal is valid. On the other hand, the luminescence signal will be detected by the PMT and then magnified by a pre-amplifier before entering the CFD. In the end, both signals enter the TCSPC counting card (Time Harp 200) for final statistical integration, as shown in Fig. 2-7, too.

The framed part in Fig. 2-8 shows that the TCPSC counting card can be further

transforms the charge (voltage) signal to digital information of time to be inputted into the histogrammer. At this moment, one detection event is completed, and with numerous times of integration, one can get the time-dependent luminescence intensity like the one in Fig. 2-6.

Due to the system resolution is limited to be about 300 ps, it is only enough to detect the recombination dynamics but incapable for the relaxation. In other words, only the decay parts in the curves (Fig. 2-6) have the quantitative meaning whereas the rising will reveal merely the system resolution. For numerical fitting, we apply a bi-exponential decay function to evaluate the carrier lifetimes

) electronic devices inside the TCSPC card.)

(Source: W. Becker, A. Bergmann,” Detectors for High-Speed Photon Counting.”)

2.6 Fabrication of InAs QDs Samples

It cannot be overemphasized the essentiality of epitaxy in fabricating semiconductor nanostructures; if the crystal quality of the sample were already poor, the following study or further processed device would turn out to be in vain because of its natural imperfections. For instance, the interference of non-ideal effects due to poor crystal quality shall mask the original phenomena from our observation, not to mention its negative influence on device performance. Therefore, researchers have worked hard in these decades to improve the epitaxy techniques and have made a superior progress, including novel mechanical inventions such as molecular beam epitaxy (MBE) and metal organic chemical vapor deposition (MOCVD); nowadays, not only can we get samples with fine quality but many novel structures and materials can be put into new attempt.

The Stranski – Keastanov method [14, 36, 37]

To fabricate semiconductor QDs, the Stranski-Keastanov mode (S-K mode) epitaxy offers a reliable and convenient way. As shown in Fig. 2-9, a thin wetting layer is firstly deposited on a barrier layer (or simply a substrate); due to the lattice mismatch between the barrier and epitaxial material, strain quickly starts to pile up in the system; it is not until the critical moment that the structure can no longer sustain the accumulation of strain does the two-dimensional surface stop forming, changing into island or point-like shape to partially release energy of the system, and QDs are formed under this circumstance. A major key for this dimensional transition to occur is that the degree of lattice mismatch must be large enough for the two materials, for

The S-K mode approach has been widely applied in the growth of different nano-structures for both its uncomplicated procedures and capability for large number of output elements in one growth period.

Figure 2-9: Schematic plot of the Stranski-Keastanov mode formation.

Quantized Energies of QDs

By varying the epitaxial conditions, we can control the size of the formed QDs, which is closely related to their intrinsic energy levels. Let us first recall a simple case.

For a free particle inside a one-dimensional box of length L with infinite potential confinement, it is well-known that its ground state energy is inversely proportional to L², so is and the energy spacing between ground and first excited state. Qualitatively speaking, the energy and energy spacing will both be higher as long as the length of the box is shorter (Fig. 2-10(a)).

In fact, the case in QDs have a good analogy since it provides just a strong potential well for confining carriers, where the energy and energy spacing are higher (lower) for smaller (larger) QDs, as shown in Fig. 2-10 (b).However, except the size of QDs, there are still other factors which determine the energy levels such as the intermixing of matrix and QD materials; for just a gradual view and simplicity, we treat the energy level and energy spacing to be effectively higher (lower) as the size of

Figure 2-10: Comparison of the effective eigenenergies of (a) free particle in a box and (b) QDs. For the two different sizes of QDs, we show the energy EQD2g(e)> EQD2g(e), and also (EQD2e -EQD2g) > (EQD1e -EQD1g).

Sample structure

Six self-assembled InAs QDs samples grown by MBE on (100) GaAs substrate are studied in this work; their structures are basically the same shown in Fig. 2-11. A relatively thick GaAs buffer layer is firstly deposited on the substrate; the self-assembled QDs were embedded in GaAs matrix and sandwiched with Al(Ga)As confinement layers. Finally, uncapped QDs were grown on the surface with the same growth conditions of the embedded ones for morphology measurement. In addition, the QDs layer of the six samples were grown under different growth temperatures and

Figure 2-11: Structure of the InAs QDs studied in this work.

Morphology

The morphology measurement is done by using atomic force microscopy (AFM), one of the most powerful methods for obtaining surface information. The AFM of the six samples are shown in Fig. 2-12 (a) to 2-12 (f) with the scale of 1×1 μm². For simple comparison, information including growth conditions, average QD density (D) width, height, and the corresponding sample numbers are all listed in Table 2-1.

Figure 2-12: AFM images of samples. (a) A (LM4682), (b) B (LM4681), (c) C (LM3572), (d) D (LM3573), (e) E (LM3472) and (f) F (LM4596).

Sample # Tgrowth (^oC) MLs D (10¹⁰/cm²) Width (nm) Height (nm)

A (LM4682) 500 3.0 6.0 54.6 8.3

B (LM4681) 500 2.4 6.8 54.5 5.1

C (LM3572) 480 2.4 8.2 62.4 2.8

D (LM3573) 480 2.4 13 70.4 2.2

E (LM3472) 480 2.6 4.5 39 4.2

F (LM4596) 480 2.0 2.8 31.2 1.9

(a) (b)

(c) (d)

(e) (f)

Chapter 3

Experimental Observation

In this chapter, we begin to introduce the results of our experimental work: the steady-state and time-resolved spectra of InAs/GaAs QDs. For time-resolved spectra, although there are totally six samples studied, the discussion will be firstly focused on two of them (sample A and C) for clear comparison. We present the first observation of an anomalous spike in the temperature-dependent TRPL around 45 – 85 K, and the spike is shown to be considerably suppressed by high excitation power. Furthermore, several InAs QDs samples with different sizes are studied, and we find that the spike will only occur in the QDs with a particular range of luminance energies.

3.1 Basic Determinations

Steady-state PL spectra

Fig. 3-1 shows the normalized SSPL spectra of the six samples measured at T = 25 K under the excitation of a 780-nm laser, with their full-width-at-half-maximum (FWHM) and ground state emission peaks ranging from 1.08 to 1.30 eV, listed in Table 3-1; the non-delta like emission energy is caused by the size-nonuniformity of the ensemble QDs, but in average, we can tell that the size is the largest for Sample A and the smallest for Sample F. In addition, although their FWHM are different among the six samples, we will show that it is irrelevant to the observation.

The steady-state PL is also a necessary work for the determination of detection points for the following TRPL measurement; each TRPL spectrum is detected under a particular PL energy, and unless specially mentioned, the detection points which we

0.9 1.0 1.1 1.2 1.3 1.4 1.5 1.6

Table 3-1: List of peak energies and FWHM of the six samples studied in this work.

Steady-state PLE spectra (selective samples)

In addition to SSPL, the SSPLE measurements on four selective samples (sample A, C, D, and E) are done at T = 77 K (liquid nitrogen) using a Ti-sapphire laser shown in Fig. 3-2. The horizontal axis of the figure represents the energy difference between the energy of laser (E_exc) and a fixed detection PL (E_det); the detection PL points are chosen to be the ones corresponding to the (ground state) peak intensities of SSPL spectra at 77 K, and for a particular data point, it means the response of the state higher than ground state with spacing E – E . The result shows an analogy for all

respectively, which the other three samples also show the similar behaviors at higher

Figure 3-2: SSPLE spectra @ T = 77 K of the four selective samples.

Calculation of numbers of injected electron-hole pairs

The excitation power in the TRPL measurement actually correlates the numbers of injected electron-hole (e-h) pairs per QD per laser pulse, calculated by the laser pulse; P is the average input power; R is the repetition rate; Elaser is the photon energy of the laser; s means the area of the laser spot; r is the light reflection percentage at the sample surface; α is the absorption coefficient of GaAs at 780 nm.

Referring to Fig. 2-11 for the structure of the sample, only the two 150nm-GaAs layers adjacent to the embedded QDs will contribute the injected e-h pairs into QDs;

starting and ending points of these GaAs layers; notice that the Al_0.35Ga_0.65As layers do not respond to the excitation so their thicknesses are neglected. Moreover, by covering metal except several windows of different sizes (order of μm) left opened on the sample surface, the spot size of the laser can be estimated by adjusting the spot into different windows and checking the intensity of the PL signal. As a result, the spot size is estimated to be roughly 60 μm in diameter. For the other parameters, we list them in Table 3-2 with the calculated result under P = 25 μW for sample A (with

starting and ending points of these GaAs layers; notice that the Al_0.35Ga_0.65As layers do not respond to the excitation so their thicknesses are neglected. Moreover, by covering metal except several windows of different sizes (order of μm) left opened on the sample surface, the spot size of the laser can be estimated by adjusting the spot into different windows and checking the intensity of the PL signal. As a result, the spot size is estimated to be roughly 60 μm in diameter. For the other parameters, we list them in Table 3-2 with the calculated result under P = 25 μW for sample A (with