Antibunching and excitation lifetimes in CdSe nanocrystals
by user
Comments
Transcript
Antibunching and excitation lifetimes in CdSe nanocrystals
Antibunching and excitation lifetimes in CdSe nanocrystals Margaret Cosgriff March 19, 2009 2 1 Abstract Using a beam-splitter and two photon detectors, we measured the cross-correlation function of a series of single nanocrystals, and found that the nanocrystals exhibit antibunching (zero correlation at zero time delay) as expected, indicating that the nanocrystals were indeed emitting only one photon at a time. Modeling the nanocrystals as a three-level energy system (ground, excited, and "dark"), we measured the characteristic time delay between photons emitted by a nanocrystal, which corresponds to the inverse of the transition rate from the excited state and the ground state. We performed this measurement using 532nm and 405nm wavelength light to excite the nanocrystals. Nanocrystals excited by 532nm wavelength light had a mean time constant of 22.7 ± 5.3ns, and those excited by 405nm wavelength light had a mean time constant of 22.4 ± 6.3ns. 3 4 2 Introduction Colloidally synthesized nanocrystals can be manufactured to fluoresce at many different frequencies within a wide range, making them very useful in a variety of experiments. As semiconductor crystals, they have applications in the manufacture of electronics.15 ,16 N anocrystals are also useful as a replacement for organic dyes in studies of cell biology: they are non-toxic, last longer than organic dyes, and, unlike many organic dyes, nanocrystals that fluoresce at different wavelengths can be excited by a single wavelength. This property allows researchers to observe multiple structures simultaneously, by tagging them with different colors of nanocrystals and exciting them with a single laserY Nanocrystals are also good candidates to be single-photon emitters, and could therefore be used in quantum computing, as well as in the classic single-photon interference physics experiment, and other fundamental tests of quantum mechanics. However, nanocrystals also "blink" (stop emitting for varying periods of time), complicating such applications. is overlap Candut:tian bmd Fermi level metal s ernie onduet or Bandgap insulator Figure 1: Electron band structure of metals, semiconductors, and insulators.13 The optical properties of a bulk semiconductor depend on its electron band struc- 5 ture. An incoming photon can excite an electron from the valence band of the semiconductor into the conduction band, leaving a "hole" in the valence band. This electron is generally excited well into the conduction band and "relaxes," losing energy nonradiatively while staying in the conduction band. The hole and the excited electron can move freely within their respective bands, until they eventually recombine, emitting a photon. The characteristic time it takes the electron-hole pair to recombine determines the semiconductors fluorescence rate. The energy of the emitted photon is very close to the energy bandgap between the conduction band and the valence band, and is always lower than the energy of the exciting photon due to this relaxation process, a fact that is crucial to the design of fluorescence microscopy experiments. v=:::o v=oo V=oo V=oo v=o o L x Figure 2: Diagram of a one-dimensional infinite potential well. 8 These energy bands, however, only exist in bulk semiconductors. In nanocrystals, which are semiconductor clusters a few nanometers across, the energy bands are split into discrete energy levels. Nanocrystals behave essentially as quantum wells, as the surface of the crystal acts as a (near-) infinite potential barrier. Because the size of a potential well determines the allowed energy levels, the diameter of a nanocrystal controls the wavelength of light at which it fluoresces. And, because the diame- ter of nanocrystals can be controlled during their creation, the relationship between nanocrystal size and fluorescence frequency means that nanocrystals can be manufac- 6 tured to fluoresce at many different frequencies. Unfortunately, nanocrystals, also called quantum dots, generally do not fluoresce continuously. Whereas in an ideal fluorescent material, a steady illumination will produce a steady fluorescence at a rate dependent on illumination intensity, absorption cross-section of the material, and exciton (electron-hole) decay rate, nanocrystals exhibit digital "blinking" behavior, switching between an on-state where the dot emits light, and an off-state where it does not. IS Excitation still occurs in this off-state, but the resulting additional electron-hole pair recombines nonradiatively.6 The statistics of this blinking behavior has recently become the subject of significant study, which has shown that the distributions of on- and off- times follows a power law, in contrast to the generally exponential distribution of these times in dye molecules and other blinking systems. 14 ,IS Several models exist that attempt to explain this power law time distribution. The models all start with a relatively simple three-level system, in which electrons transition directly between a ground and an excited state, as in normal fluorescence, and also from the excited state to a dark state, and then back to the ground state, which does not result in photon emission. The dot stops fluorescing when either an electron or a hole escapes the dot or is trapped in a surface state, leaving the dot with an overall charge. In a charged dot, excitons can recombine via a nonradiative Auger process, which occurs on a much faster timescale than radiative processes, making the dot much less likely to fluoresce. 6 The models then differ on the source of the power law statistics. Traditional rate-limited behavior is characterized by exponential distributions of lifetimes. These rates are related to transitions between quantum levels by Fermi's Golden Rule, and even complex distributions of rates result in exponential distributions. Power law behavior, in contrast, is characteristic of "random walk" processes. 7 Models of nanocrystal blinking try to address the question of what is occurring physically that can be described as a random walk. One model holds that an excited nanocrystal ejects an electron that performs a physical random walk before returning to the nanocrystal. While the nanocrystal is charged, it is dark, and the coulomb force keeps the ejected electron from escaping to infinity. Another model has an electron tunneling to one of many surface trap states, which are distributed such that the offtime distributions follow a power law. A third model has an excited electron making the transition to and from a trap state of varying energy only when the two states are in a very narrow resonance range, with the energy levels themselves performing a "random walk." 10,18 Another feature of quantum dots is that they exhibit anti bunching, which we observe through a measurement of the correlation function g(2) (T). Our measure- ments use a Hanbury-Brown/Twiss (HBT) configuration, where light is emitted from a source is split and directed to two detectors, and those two detectors (in the case of antibunching) never register photons at the same time. Antibunching in quantum dots is strong evidence that they are single photon emitters. If the two detectors never, or almost never, detect photons from a quantum dot at the same time, it is highly unlikely that the dot ever emits two photons at once. Beyond the qualitative observation of the resulting dip at time T = 0, measurement of the g(2) (T) function can also be used to determine the distribution of excited state lifetimes among nanocrystals, and the wavelength dependence of this distribution. Antibunching behavior has been experimentally confirmed in CdSe, ZnS, and InAr nanocrystals. 2 ,15 Most such experiments make use of pulsed excitation, but antibunching has also been observed from quantum dots under continuous-wave excitation. 12 I studied the antibunching behavior of CdSe nanocrystals at two laser wavelengths, 8 one at 532nm and the other at 405 nm. I measured the second order correlation function g(2) (T) from the fluorescence of these nanocrystals to derive the transition rates between the I -exciton excited state and the ground state of these quantum dots. 9 10 3 Theory A common test that a fluorescent emitter is a single dot is to check for anti bunching. A standard method for measuring anti bunching, and also state lifetimes, is to measure the second order correlation function (2)(T) g = g(2) (T), given by the equation J I(t)I(t + T)dt J 12(t)dt (1) We did not measure this function directly, because given the time-scale of fluorescence ('"'-'IOns), we would need to be able to measure I(t) with sub-nanosecond precision, which was not possible with our equipment, or without increasing fluorescence rates so high that we would have damaged our samples. Instead, we used an HBT setup (explained in Section 4) to split the fluorescence from the sample into two channels, and measured the time delay between a detection in one channel and the next detection in the other channel. This type of measurement is a very good approximation of the cross-correlation function assuming the count rate is low over the time interval being observed (which in this experiment it was: typical count rates were '"'-'50kHz, and the full time interval over which we observed g(2) (T) was 500ns). Under that assumption it is very unlikely that two photons would be detected in the second channel during that interval; to have multiple photon counts contribute to the correlation function, the count rates would have to be at least 2MHz.9 We model a nanocrystal as a three level system, with transition rates from the ground to excited state (k 12 ), from the excited state to the ground state (k2d, from the excited state to a dark state (k 23 ), and from the dark state to the ground state (k3d. The behavior of the dark state that generates power law blinking does imply 11 Channell .Een ~ OJ E H Figure 3: Model graph of intensity vs. Channel 2 time for our two detectors. Assuming a low count rate (so that each "count" only represents one photon), g(2)(7) is just the distribution of ~t's. Our experimental setup, which only records the ~t's between a detection in the "first" channel and the next detection in the "second," would miss ~t3 in this example, but since the count rate is assumed to be low, ~t3 and similar intervals are unlikely to be within the range of ~t's (±250ns) that we are considering. that the form of k3I is more complicated than the constant value we assume it to be here, but the time-scale on which blinking occurs is too slow to affect the correlation function on the time scale of our experiment. The correlation data can therefore be fit to a function derived from rate equations for the purpose of extracting lifetime information during an on-time cycle. When the nanocrystal transitions from the excited state to the ground state, it emits a photon; all other transitions do not produce light. N I , N 2 , and N3 are the probabilities of the the nanocrystal being in the ground state, excited state, and .. . dark state, respectively, and N I , N 2 , and N3 are the corresponding time derivatives. 12 Excited state Dark state k12 Ground state Figure 4: Diagram of a three level system, the energy model we used to describe a nanocrystal. The k ij parameters are transition rates between the energy levels i and J. Assuming constant transition rates, this implies the following relations: (2) (3) (4) (5) We expect solutions of the form 13 (6) Plugging in, we find an equation for A: Of the four transition rates, is k12' k21 is the largest , on the order of 10 7 8 - 1. Next largest which is proportional to the intensity of the exciting light and the absorption scattering cross-section, on the order of 10 58-1, with order of 10° to 103 . 6 This relationship, k21 » k12 » k23 and k 23 , k31 ' k31 much smaller, on the allows several simplifying approximations to be made, giving the following expression for the three possible values of A. Thus the equations for N1 (t) , N2 (t) , and N3 (t) (after some algebra) are (8) (9) (10) 14 where The intensity of the light emitted by a nanocrystal is proportional to k 12 N 2 (t) , the number of electrons in the excited state times the rate at which they transition to the ground state. Since A3t « 1 for the range of time we are considering, this rate is (11) The cross-correlation function g(2) ( T) should be proportional to this expression. 15 16 4 Experimental Methods In a traditional optical microscope the sample material is illuminated by a lamp. The light reflected or scattered from the sample is magnified by an objective lens, generating an image that is observed by a camera or human eye through an eyepiece. Although it allows for direct visual examination of the sample, the presence of diffuse, broad-spectrum lamplight makes measurements of light emitted by a fluorescent sample almost impossible. A solution to this problem is to excite the fluorescent sample with a light from a narrow band of wavelengths, often using filtered light from a mercury lamp. This method is called fluorescence microscopy. The wavelength of light emitted by a fluorescent material is always longer than the exciting wavelength, a difference known as the Stokes shift. Not all materials fluoresce; some only scatter incident light. In order to examine these materials using fluorescence microscopy, the material can be stained with a fluorescent dye, or labeled with a semiconductor nanocrystal. Since nanocrystals generally do not react with proteins, they are therefore useful in studying biological systems. The Stokes shift allows the signal from the fluorescing sample to be separated from the exciting wavelength with the use of a dichroic mirror or filter that transmits one wavelength but reflects or absorbs the other. In epifluorescence microscopy, the signal from the sample passes through the same optic that focused the exciting light onto the sample, reducing the number of optical elements required, and ensuring that the detection system is focused on the sample region that is being illuminated. The image resolution can be improved still further if a laser instead of a lamp is used to excite the sample. Light from a lamp is incoherent, and can at best be 17 Radius: 2' A HIA 9A 1.5 2.5 2 3 Photon Energy (eV) Figure 5: "Absorption (thick lines) and fluorescence spectra (thin lines) for CdSe nanocrystals at T=77K." 11 Note that the absorption peak is always at a higher energy than the fluorescence peak. focused down to a lOOj..lm diameter spot on the sample plane. In order to focus on a smaller area, a coherent light source, such as a laser, must be used to illuminate the sample. Laser light, which is also wavelength-specific, can in fact be focused down to the diffraction limit, a spot on the order of lOO-300nm for the 532nm wavelength light used in this experiment. This thousand-fold improvement in spatial resolution allows for the creation of scanning microscopes, in which either the exciting light or the stage on which the sample is mounted move, so that different parts of the sample 18 barrier filter dichroic beam spUtter exciUtUon fUter exciting light objective emitted fluorescence light Figure 6: Fluorescence mIcroscopy diagram. The high-energy (in this case blue) exciting light reflects off the dichroic mirror, is focused by the objective, and absorbed by the specimen. The specimen then emits lower energy (in this case green) light, which passes through the dichroic mirror. 7 are excited, and fluoresce, at different times. Both the intensity of the signal and the relative position of the exciting light and the sample are known at all times during the scan, creating a map of the sample. The dichroic mirror effectively removes most of the exciting light noise from the signal, but it cannot stop light from other parts of the sample from interfering with 19 the measurement. To reduce this problem, confocal microscopes have the light from the sample pass through a pinhole. Light from points to the side or even (for threedimensional samples) above and below the desired point on the sample will not focus at the pinhole, and will therefore be largely blocked. This component helps increase the signal-to-noise ratio of the sample data at the cost of decreased overall signal. Taken together, the laser light source for tight focusing and the confocal geometry for blocking background noise allow the performance of single-photon-counting experiments, which are crucial for detecting antibunching. I I Figure 7: Confocal microscopy.19 In the upper image, light from outside the focal plane does not focus onto the pinhole, and is blocked. In the lower image, the pinhole is wider, allowing more light through, including light from outside the focal plane. In our experiment, we used a 532nm Nd:YAG laser and a 405nm diode laser to illuminate our sample. The laser light reflects off two mirrors, which were adjusted to send the beam through two irises, defining the light path. We placed several easily removable optical density filters in the path of the beam to adjust the power reaching 20 NdYAG Laser ~ HeNe Laser ~ Diode Laser 405 n m Piezo Sample Nano Stag e ='""'-7"----"~_T_"""'""'= Microscope Objective i! Flip Mirrors To PC Optical Density Filters Beam Expander +-. Dichroic Mirror t ~L. Beam Splitter Adjustable Aperture I'- - - - - -' Filter Lens Figure 8: Optical layout for the microscope when it is in confocal mode. the sample. We used a power of 5-10mW to illuminate the sample in wide-field imaging mode (described below), and O.5j.JW while in confocal mode, when the beam is much more focused at the sample plane. We estimate that the light intensity at the sample is the same for both modes, approximately 200W /cm 2 . The beam then passes through an expanding lens and is raised to the microscope entrance height by a periscope. The beam is expanded by a lens before it enters the microscope objective so that it can be more tightly focused onto the sample by the objective lens, a process known as "filling the back aperture." The microscope we used is an Olympus microscope with a 60x 1.42NA oil immersion objective. It has a removable lens where the laser enters the microscope (not shown) that allowed us to switch easily between wide-field mode 21 (with the lens) and confocal mode (without). After this lens is a dichroic mirror, which reflects almost all (>99%) of the laser light, and similarly passes almost all of the emitted red-orange light from the sample. In confocal mode, the light beam, still expanded, reflects off the dichroic mirror and enters the microscope objective, focusing to the diffraction limit of 1.22)..1 /d ~ 153nm after passing through index-matching oil to increase the light collecting parameter (the numeric aperture). In wide-field mode the light beam is no longer expanded when it enters the objective, having passed through the removable lens, and so illuminates a much wider area on the sample. The sample stage has two positioning drives: one coarse stage with centimeter travel, and a piezoelectric driven 3-axis stage with f"V 10nm positioning accuracy. Ap- proximately 30% of the fluorescence from the sample is collected by the objective, as in epifluorescence microscopy, and passes through the dichroic mirror and another lens that focuses the signal light onto the plane of the aperture, 100f-lm diameter pinhole. At this point approximately 20% of the photons emitted by the sample remain in the output signal. After the aperture, another lens re-collimates the output signal. The signal is then split by a 50-50 beamsplitter, and reaches the two avalanche photodiodes (APDs). Avalanche photodiodes use a semiconductor to detect photons. A photon entering the APD is absorbed by the semiconductor, which creates an electron-hole pair. By itself, the charge of a single such electron is not easily measured, but the APD applies a voltage to the semiconductor that separates the charges. The voltage accelerates the electron to the point that it excites other electrons in the semiconductor to the conduction band, a phenomenon known as the avalanche effect. This process creates enough charge to be measured, and the APD sends out a voltage pulse, indicating it has detected a photon. APDs are very sensitive; exposure to the ambient light in a 22 room lit by ordinary lamps will effectively destroy the detectors. To limit this risk, and the likelihood of detecting light not from the sample,we kept both detectors in an opaque box to further reduce the risk of false detections and damage to the devices (only light exiting the 100p,m pinhole can reach the detectors). The signal from the two APDs went to a time-to-amplitude converter (TAC) to measure the cross-correlation function of the sample fluorescence. The cross- correlation function g(2) ( T) can be approximated, for sparse data, by the time intervals between consecutive pulses from the two APDs. The signal from one detector goes to the TACs "start" channel, while the other, after passing through a length of delaying wire, goes to the "stop" channel. The TAC, as its name suggests, converts the time between a pulse in the start channel and a pulse in the stop channel, if less than a certain maximum time , to an output voltage pulse, with an adjustable conversion factor which we set to 10V per 500ns. The signal from the TAC goes to a multichannel analyzer, which uses a program called Gamma to plot the distribution of incoming voltages. This distribution, after accounting for the conversion factor and the delay in the stop signal, approximates the autocorrelation function of the fluorescence emitted by the sample. Procedure Our samples consisted of CdSe nanocrystals on a quartz slide. The nanocrystals come in a toluene solution, which we diluted in toluene three times: the first two times at a ratio of 1:100, and the third time at a ratio of 1:10. We flame-cleaned a quartz slide to remove surface contaminants and water, and spin-coated it with 8-12p,L of the diluted nanocrystal solution. We then mounted the slide on a custom-made slide holder, and placed the slide in a sealed container to dry it , by pumping the air from the 23 container and waiting for several minutes. The nanocrystals are protected by a shell of carbon chain molecules, but this protective coating dissolves in a dilute solution, so it is important to minimize the time between diluting the nanocrystal solution and drying the sample. This also means that samples degrade over time, and are generally useless after a day or two. We had previously used glass slides instead of quartz, which we cleaned by bathing them in a mixture of 70% sulfuric acid and 30% hydrogen peroxide, which we dubbed a "piranha clean." This method was effective, but it was also time-consuming, so we experimented with flame cleaning, which seemed to be just as effective. Unfortunately, when we tried to flame-clean glass slides, they tended to warp and break, so we had to switch to quartz slides. We also experimented with using a coating of polymethylmethacrylate (PMMA) to protect the dots from bleaching (signal degradation over time due to laser damage to the surface of the dot), but our sample of PMMA itself fluoresced, and may have scattered signal from the dots, creating a signal-to-noise issue that we could not resolve. Because of these problems, we stopped using PMMA. After the sample dried, we refreshed the index-matching oil on the objective, cleaning off any remaining oil and adding a new drop. We then mounted the slide holder on the microscope stage, and carefully raised the objective until the bead of oil touched the bottom of the slide. Once the oil and the slide were in contact, we put the microscope in wide-field mode, and, using the microscope lamp, tried to focus on a bit of dust on the upper surface of the slide, as an approximation of the vertical position of the nanocrystals. After finding and focusing on a piece of dust, we turned off the lamp and the room lights, turned on (or unblocked) the Nd: YAG laser, checking to make sure that about 5mW of power was getting through, and switched the output light from the eyepiece to a CCD camera mounted above the eyepiece arm of the 24 mIcroscope. Since the nanocrystal solution we used was so dilute, it was very unlikely that a dot would be in our confocal scanning range, but in wide-field mode we could use the camera to center a dot at a position we had previously determined to be in our confocal scanning region. The camera focal plane was very close to the focal plane in confocal mode which allowed us to position the dots in camera mode. Without the wide-field mode and the CCD camera, it would be very difficult to align a single quantum dot with the confocal detection system. Once we had found and focused on a group of dots in camera mode, we blocked the laser light and added more optical density filters, so that the power entering the microscope was approximately .5/-lW. With an approximately diffraction-limited spot, this corresponded to approximately 200W/cm 2 . We then removed the wide-field lens, and turned on the power to the APDs. Using a LabView program, we scanned the largest area possible (25/-lm by 25/-lm) , looking for a region of high emission, which would indicate the presence of a dot. If the area was uniformly dark, we moved the microscope stage and repeated the scan, until we found a bright spot. We then did a narrow scan of the bright spot, adjusting the the vertical position of the piezostage to refine the focus further. If the bright spot appeared to be a dot (the emission intensity switched back and forth between a high "on" state and a low "off" state, with no intermediate level of intensity), we then turned on the TAC and dwelled on the center of the bright spot, gathering data. 25 26 5 Results We collected data from 13 emitters (either single dots or clusters of two or three) illuminated by the 532nm laser, and 17 emitters illuminated by the 405nm laser. The intensity vs time data was collected as a series of timestamps, which we converted to intensity by re-binning: summing the number of counts in each 10ms interval. 350 300 '2 :E 250 '" S 0 ,...., 200 .... <l.l ,3< '" ~ 150 ::l 0 u 100 50 0 0 50 100 150 200 250 300 Time (sec.) Figure 9: Intensity vs time graph of a single dot (8_06_08_16). This dot exhibits blinking, and the overall downward trend in the 'on' intensity indicates that the dot was also bleaching. The correlation data is a list of bin counts from the multi-channel analyzer (MCA), with each bin corresponding to a specific TAC voltage. The TAC settings and MCA resolution (1024 channels over a range of 0-10V) allow conversion of the signal to time correlation data. The raw data, therefore has a time-bin resolution of ~g~:c~ = 0.488ns/ch. With typical lifetimes of 20-30ns (and visible curvature out to ±100ns), a 27 lower resolution ('"'-'5ns/ch) would improve the signal-to-noise ratio while still allowing a valid fit. We therefore also re-binned the correlation data, condensing 10 channels into one point. We plotted the intensity vs time and correlation vs time delay data for each emitter using Igor, a data analysis software package, and fit the correlation data to a curve of the form C(T) = A(l- e-blt-tol) + c, (12) 80 60 ~ p '-' u 40 20 0 -200 -100 0 100 200 't (ns) Figure 10: Cross-correlation function of a single dot (8_06_08_16) illuminated by 532nm wavelength light, with a fitted curve. A = 61.5 ± 5.8, b = 0.0393 ± .0061 , c = The parameters for the curve are 14.4 ± 6.0, and to = -.1 ± 1.7. where b is the inverse of the time constant. The constant c can correspond to the background count, but a high enough value (approximately half of the maximum function value) indicates the presence of two dots, rather than one. "A" corresponds 28 to the intensity of the emitted light, and to IS an offset from zero, accounting for possible delays in the electronics. The correlation plot in Fig. 10 shows clearly that the sample was a single quantum 0, an offset of 14.4 ± 6 counts is far less than half of the 61.5 count emitter: At T = amplitude. A Poissonian source would generate a constant correlation amplitude given by rlr 2i:ltT where rl and r2 are the detector count rates, i:lt is the bin size (in seconds) and T is the total data collection time. 3 For the data in Fig. 9, rl = 7433 Hz, r2 = 4266 Hz , i:lt = 4.89 x 1O- 9 s, and T = 300 s, resulting in a Poissonian count of 46.5. The fact that the correlation is super-Poissonian for times greater than 100 ns is a consequence of long-time photon bunching. 1 Ten of the emitters illuminated by the 532nm laser, and 14 from the 405nm proved to be single dots with sufficient fluorescence output to produce a high enough signal-to-noise ratio. Of the remainder, one data set proved to be the product of two dots, and the rest were from dots too weak to produce a useful cross-correlation function. 29 200 150 ~ p u 100 '-' 50 0 -200 -100 100 0 200 't(ns) Figure 11: Cross-corrleation function of what is probably two dots (8_06_08_8), illuminated by 532nm wavelength light. The parameters for the curve are A 112.0 ± 5.8, b = 0.0615 ± 0.0095, c = 55.2 ± 11.5, and to = -3.6 ± 1.1. Note both the higher "A" value, indicating a higher overall intensity, and the 'c' value (corresponding to background emission), which is comparable to the 'A' value of the previous graph. That the intensity at zero time-delay of this emitter is of comparable magnitude to the maximum intensity of the emitter in figure 8 suggests that this emitter consists of two dots. 30 Time constants (ns) The correlation data in all of the single 532 nm 405 nm dots measurements rises from the anti- 14.9±4.4 23.7±2.8 26.8±4.3 32.1±7.2 26.2±2.8 14.6±2.1 20.8±2.6 16.6±5.5 12.8±4.9 34.8±9.7 23.4±5.5 29.2±5.1 22.8±5.2 19.4±2.6 24.4±3.6 22.4±4.0 29.7±3.5 22.7±3.1 24.4±3.9 14.9±4.4 15.4±4.8 22.8±10.4 21.2±4.5 bunching minimum at T = 0 to the super- Poissonian level with a time constant "\2 ;:::::; k21 . Thus, from a data curve fitting pro- cedure in Igor , we can extract the single dot fluorescence rate k 21 . For each excitation wavelength, the result of the data analysis is accumulated in the histogram plots of Fig 12. These two distributions are centered around the same average value (22.4ns vs 22.7ns) but the 405nm data seem to be more broadly distributed. However, given the small number of measurements, it is difficult to draw a meaningful conclusion from the data about a re- 24.0±2.9 Fluorescence lationship between excitation wavelength lifetime data from single and the time constant of a fluorescing dot NCs or the shape of its distribution. Table 1: 31 4- (a) 405 nm - --.. ~ 3II 62 ...... -r- = § 1- r- - U o 10 15 20 25 30 35 Time Constant (ns) 4- (b) 532 nm - --.. 8 3II 62 ...... = § 1- U r- r- r- o f- I 10 15 20 25 30 35 Time Constant (ns) Figure 12: Histogram of time constants of (a) dots illuminated by 405nm light (mean time constant 22.4 ± 6.3ns), and (b) dots illuminated by 532nm light (mean time constant 22.7 ± 5.3ns). 32 6 Discussion As a means of identifying single emitters, the fluorescence, two detector correlation (RBT) experiment is very useful. While it is possible to make a qualitative observation of the fluorescence intensity variability, the anti-bunching signal at T = 0 is a definitive, quantitative measure. The vast majority of fluorescence emitters that we measured using the ND600/quartz substrate were single NC's with only one measurement being from a two- N C source. The simple three-state model is valid for very short time scales, on the order of a few hundred nanoseconds after emission. At longer times, however, a spectral diffusion trap-state model is more useful to describe the distribution of on- and off-times of a fluorescing nanocrystal. The lifetime measurements did not show a quantifiable difference with excitation energy (see figure 12). The excitation with 532nm wavelength light is at an energy (2.33eV) near the absorption edge of the dots we were using, and excites the dots to the lS e delocalized energy level, while 405nm wavelength light is well above the absorption edge and excites the dot to the 1P3 level.lO The dots excited by the 405nm laser should exhibit a broader distribution of time constants due to the larger effects of spectral diffusion at higher energy levels. l l The distribution of on-times has been found to follow a power law for times up to 3s, followed by an exponential cutoff when the nanocrystal is excited with 532nm and 488nm light, but the on time distribution drops off earlier when the nanocrystal is excited with 405nm light (see figures 14 and 15). This narrower distribution of on-times implies a wider distribution of high or excited energy states in nanocrystals excited with 405nm light, which would allow the nanocrystals to more rapidly excite and de-excite when they are "on," leading to the broader lifetime distribution our data tentatively displays. 33 Our data suggests a slightly larger standard deviation for the 405nm time constants than the 532nm time constants (6.3ns vs 5.3ns), but it would require data from upwards of 100 more dots to draw a meaningful conclusion from that difference. At both excitation energies there was a significant amount of variation from dot to dot in the measured time constant. Some of the time constant variation between dots may be due to variation in the size of the dots we used, or in the surface environment of the slides, which could create large variations in the local electric fields experienced by the dots. The time constant of a dot is also the time signature of the width of its emission line. 4 Variations in dot size and slide surface environment could therefore cause inhomogeneous broadening of this line. In considering future directions for this research, the variation of fluorescence decay rate with time could serve as an independent measure of spectral diffusion, complimenting spectral measurements. 5 Currently, we have to use an entire data set to fit a g(2) (T) function because only approximately 10% of the photons emitted by the dots reach the detectors, and the TAC only registered a count if two of these photons arrived within 500ns of each other. Although this sparseness in our data makes the data we recorded a good approximation of the cross-correlation function, it also makes it impossible, with our current setup, to observe possible time variation in the time constants of individual dots. A pulsed laser excitation source for this purpose has been proposed for funding. 34 Figure 13: "Experimental fluorescence spectra for CdSe-nanocrystals[oo.] for various excitation energies at 77 K" 11. Note the broadening of the fluorescence peak as the excitation energy increases. Spectral diffusion could thus affect the width as well as the frequency of the nanocrystals' emission spectrum peaks. 35 1 ~---r----~--~----~----r---~ NC600 0 ~ ~ (a) >. ~ -1 c Q) "0 >- -2 .1: :g -3 .J:l ... 488 nm, 175 W/cm 2 e0. .4 - c o mon = 1.42, 'Ton = 9 s • 532 nm, 300 Wlcm 2 - - - mon = 1.44, 'Ton = 9 s 0> -5 .2 ·6 ~_--L-_--'--<:--_-'--_-:-'::_ _- ' - - _ " " ' " -1.0 0.0 2.0 log(on t im~ (s)) Figure 14: Log-log graph of on time probability density vs. on time for NC600 dots excited by 488nm and 532nm wavelength light 5. ,--. ,--. ";' 'I) - ~ 0 :>. 'I) c: oD "0 :>. -1 -2 .0 -3 .0 0 I... -4 ro CL c: 0 ~ -5 0\ 0 -6 -1.0 0.0 1.0 2.0 Figure 15: Log-log graph of on time probability density vs. on time for NC600 dots excited by 405nm and 532nm wavelength light 5 . Note the earlier drop off for 405nm wavelength exciting light. 36 7 Conclusion We used a two-detector RBT setup to examine the fluorescence of CdSe nanocrystals. Much of the optical arrangement we used had been assembled previously for another nanocrystal experiment, but we assembled the post-microscope setup, aligning the confocal pinhole, beamsplitter, and APDs, and building an opaque box to protect the APDs from ambient light. We coated flame-cleaned slides with a diluted solution of nanocrystals, and dried them, before mounting them on the microscope. We excited the nanocrystals with 532nm and 405nm light, and measured the cross-correlation function g(2) (T) of individual emitters using two avalanche photodiodes and a time- to-amplitude converter, using Igor to transform the raw TAC data to a useable form, and to fit the predicted form of g(2) (T) to our data. We confirmed that CdSe nanocrystals exhibit antibunching, and are therefore single photon emitters. Of the 25 emitters we observed that produced a strong enough signal to meaningfully analyze, one proved to be a cluster of two nanocrystals, whose g(2) ( T) function dropped to half-maximum at T = 0, and the rest had cross-correlation functions that clearly exhibited antibunching, dropping to a small fraction of the maximum g(2) (T) value at zero time delay. The excitation lifetimes we extracted from our cross-correlation data did not seem to vary between excitation wavelengths. Nanocrystals excited by 532nm light had a mean time constant of 22.4ns, while those excited by 405nm light had a mean time constant of 22.7ns. There was a suggestive difference in the distribution of these time constants (standard deviation of 5.3ns for the 532nm light, vs 6.3 for 405nm light), but this would have to be confirmed in a much larger study. 37 38 8 Acknowledgements I would like to thank Professor Catherine Crouch for her advice and assistance, and the use of her lab, Robert Mohr for his help and moral support, Orion Sauter for his work creating the programs that recorded and processed our data, and of course my advisor Professor Carl Grossman for his constant help and support throughout this endeavor. 39 40 9 References 1. Bernard J, Fleury L, Talon H, Orrit M: Photon bunching in the fluorescence from single molecules: A probe for intersystem crossing. Journal of Chemical Physics 1993; 98: 850-859. 2. Brockmann X, Giacobino E, Dahan M, Hermier JP: Highly efficient triggered emission of single photons by colloidal CdSe / ZnS nanocrystals. Applied Physics Letters 2004; 85: 712-714. 3. Brouri R, Beveratos A, Poizat J-P, Grangier P: Photon antibunching in the fluorescence of individual color centers in diamond. Optics Letters 2000; 25: 12941296. 4. Corney A: Atomic and Laser Spectroscopy. Oxford University Press: 1977; 231-233. 5. Crouch CH, Mohr R, Emmons T: Excitation energy dependence of fluorescence intermittency in CdSe/ZnS core-shell nanocrystals. Journal of Physical Chemistry, submitted. 6. Efros AL , Rosen M: Random Telegraph Signal in the Photoluminescence Intensity of a Single Quantum Dot. Physical Review Letters 1997; 78: 1110-1113. 7. Epi-fluorescence with the Zeiss Universal. University of Victoria 2005. http://web.uvic .ca / ail/techniques / epi-fluorescence. html. 8. Esham B: Infinite Potential Well. http://en.wikipedia.org/wiki/Particle in a box. 9. Fleury L, Segura JM, Zumofen G, Hecht B, Wild UP: Nonclassical Photon Statistics in Single-Molecule Fluorescence at Room Temperature. Physical Review Letters 2000; 84: 1148-1151. 41 10. Frantsuzov PA, Marcus RA: Explanation of quantum dot blinking without the long-lived trap hypothesis. Physical Review B 2005; 72: 155321-1- 155321-10. 11. Hoheisel W, Colvin VL, Johnson CS, Alivisatos AP: Threshold for quasicontinuum absorption and reduced luminescence efficiency in CdSe nanocrystals. J. Chern. Phys. 1994; 101: 8455-8460. 12. Kiraz A, Falth S, Becher C, Gayral B, Schoenfeld WV, Petroff PM, Zhang L, Hu E, Imamoglu: Photon correlation spectroscopy of a single quantum dot. Physical Review B 2002; 65: 161303-1- 161303-4. 13. Kuiper P: Isolator-Metal. http://en.wikipedia.org/wiki/File:Isolator-metal.svg. 14. Pelton M, Smith G, Scherer NF, Marcus RA: Evidence for a diffusion-controlled mechanism for fluorescence blinking of colloidal quantum dots. PNAS 2007; 104:1424914254. 15. Santori C, Fattal D, Vuckovic J, Solomon GS, Waks E, Yamamoto Y: Submicrosecond correlations in photoluminescence from InAs quantum dots. Physical Review B 2004; 69: 205324-1- 205324-8. 16. Service RF: Small Clusters Hit the Big Time. Science 1996; 271:920-922. 17. Service RF: Semiconductor Beacons Light Up Cell Structures. Science 1998; 281: 1930-1931. 18. Stefani FD, Hoogenboom JP, Barkai E: Beyond quantum jumps: Blinking nanoscale emitters. Physics Today 2009; 62: 34-39. 19. Use of Confocal Microscopy. Trinity College Dublin 2006. http://www.tcd.ie/Bio chemistry / confocal / why confocal. ph p. 42