重新评估硅纳米晶体中的发光寿命分布
摘要
纳米晶体集合中的发光动力学因各种过程而变得复杂,包括不均匀加宽样品中辐射和非辐射率的大小依赖性和粒子间相互作用。这会导致非指数衰减,对于硅纳米晶体 (SiNC) 的特定情况,已使用 Kohlrausch 或“拉伸指数”(SE) 函数广泛建模。我们首先推导出 exp[− (t /τ) β ]。然后,我们比较了通过假设发光衰减或种群衰减遵循此函数计算的分布和平均时间,并表明 β 的结果显着不同 远低于 1。然后我们将这两种类型的 SE 函数以及其他模型应用于来自具有不同平均尺寸的两个热生长的 SiNC 样品的发光衰减数据。平均寿命在很大程度上取决于实验设置和选择的拟合模型,这些模型似乎都不能充分描述集合衰减动力学。然后将频率分辨光谱 (FRS) 技术应用于 SiNC,以直接提取寿命分布。速率分布的半宽度为 ~ 0.5 十进制,主要类似于有点高频偏斜的对数正态函数。 TRS和FRS方法的结合似乎最适合揭示具有宽发射光谱的NC材料的发光动力学。
介绍
胶体纳米粒子可用于多种应用,包括催化、医疗和光电应用 [1,2,3,4]。半导体纳米粒子对光发射、光伏和光催化应用特别感兴趣 [5, 6]。由于硅的可调发射特性 [7] 以及硅的丰度和生物相容性 [8],硅纳米晶体 (SiNCs) 是当前关注的焦点。为了开发基于纳米粒子的技术,需要深入了解相关的光电特性,而时间分辨光谱通常是实现此目的的宝贵工具。
SiNC 的发光寿命通常用拉伸指数 (SE) 函数建模,其基本形式为 exp[ - (λt ) β ],其中色散参数β 取 0 到 1 之间的值,λ 是一个速率参数,t 是时候了。该函数通常被描述为“比指数慢”,并且暗示着衰减率的不对称分布向着更长的寿命方向发展。一旦 β 和 λ 通过拟合发光衰减曲线找到参数,可以近似重建相应的衰减率分布[9]。
在过去的 20 年里,硅和其他半导体量子点中 SE 发光衰减的起源引起了激烈的争论,而且争论最近还在继续 [10]。已经对衰减动力学中 SE 的出现提出了各种解释,包括载流子隧穿和捕获在紧密间隔的纳米晶体 [11]、不均匀加宽的尺寸分布 [12]、尺寸相关的电子-声子耦合 [10] ,以及非辐射复合的势垒高度分布 [13],后者类似于先前对多孔硅的建议 [14]。显然,要更广泛地理解SiNCs以及半导体纳米晶体中的发光机制,需要了解速率分布。
在以前关于 SiNC 的许多文献中,拉伸指数衰减是先验假设的,通常没有分析其他可能的分布。 SE 倾向于在视觉上很好地拟合(即,最佳拟合线似乎与“肉眼”数据很好地匹配)。此外,在绝大多数以前的工作中,例如 [15],对于是否实际模拟了种群衰减或发光衰减缺乏明确性。这些通过导数相关,应该使用正确的表达式来理解样本中的衰减时间尺度 [16]。此外,由于宽的集合发射光谱,探测器的响应度函数会对 SiNC 中测量的发光衰减曲线产生显着影响。尽管如此,响应度很少被考虑在内,因此很难比较不同调查的结果。最后,以前没有研究尝试在硅纳米晶体的分析中使用频率分辨光谱 (FRS)。原则上,FRS允许在不先验假设模型的情况下提取寿命分布。
本文的目的是建立一种方法来测量、模拟和解释硅纳米晶体的发光动力学。希望这有助于更好地理解文献中经常相互矛盾的结果的巨大多样性,导致不同测量之间更好的一致性或至少更高的一致性,并更好地理解发光机制。
基础理论
我们比较了三种模型:广泛用于 Si 纳米晶体的拉伸指数、最近首次应用于 SiNC 的对数正态衰减分布 [17] 和双分子衰减。对于任何模型,发射概率密度函数,用强度函数g的积分表示 (t ),在时间 t′ 与 t′ 处剩余的激发分数有关 根据[16]。
$$ {\int}_0^tg\left({t}^{\hbox{'}}\right) dt=1-\frac{c_t}{c_0}, $$ (1)其中 c t 和 c 0 是 t 时刻被激发的 NC 的数量 和最初。概率密度函数描述了时间 0 和 t 之间发射的光子的分数 相对于发射的光子总数。如果种群衰减遵循一阶速率方程(即“单分子”重组),我们有 dc t /dt = − λc t , 其中 λ =1/τ 0 , 导致通常的 c t /c 0 =exp[− λt ] 和 g (t ) =λ⋅ exp[− λt ] 取等式两边的时间导数后。 1. 导数是必要的,因为在窗口dt'中测量的发光强度 与该区间内激发分数的变化成正比。
如果我们同时考虑辐射率和非辐射率,那么我们替换总衰减率 λ 与 λ R + λ NR 所以 g (t ) =(λ R + λ NR )exp[− (λ R + λ NR t ] =λ R exp[− (λ 里 + λ NR )t ] + λ NR exp[− (λ R + λ NR )t ] 其中只有第一项是可测量的,产生时间分辨光谱 (TRS) 的测量强度
$$ g(t)={\lambda}_R\exp \left[-\left({\lambda}_R+{\lambda}_{NR}\right)t\right]。 $$ (2)用于拟合数据的衰减函数,I t =A· exp(− λt ) + dc,使用额外的任意前置因子 A 进行缩放 ,这取决于检测效率和激发的纳米粒子数量,并将导致适当的规模。通常在衰减函数中加入一个直流偏移作为另一个拟合参数。
在拉伸指数衰减的情况下,激发发射体的比例根据
$$ \frac{c_t}{c_0}=\exp \left[-{\left({\lambda}_{SE}t\right)}^{\beta}\right]。 $$ (3)其中 λ 东南 是拉伸指数衰减率(等于 1/τ 东南 ). 将其插入等式。 1 和前面一样取两边的导数得到一个发射概率函数为
$$ g(t)={\beta \lambda}_{SE}^{\beta}{t}^{\beta -1}\exp \left[-{\left({\lambda}_{SE} t\right)}^{\beta}\right]。 $$ (4)一种估计频率分布的方法H (λ ) 导致方程。 3 使用拉普拉斯逆变换 [9] 显示,产生的分布随着 β 的减小而变宽 并且偏向高频。
不幸的是,在方程式中。 4、不可能将前因分为辐射和非辐射部分。这意味着方程。 4 仅针对 λ 正确归一化 NR = 0 [16],从 PL 衰减曲线获得的寿命分布只能通过这种方式来理解。此外,前因子中有一个时间依赖项;因此,与发光衰减相比,种群衰减具有不同的时间依赖性 [16, 18]。为了获得τ的值 东南 和 β 对于可以从中提取适当平均寿命的种群衰减,必须使用等式。 4 模拟观察到的衰变,我们替换 g (t ) 由测量的衰减函数 I t :
$$ {I}_t=A{\beta \lambda}_{SE}^{\beta {t}^{\beta -1}\exp \left[-{\left({\lambda}_{SE }t\right)}^{\beta}\right]+\mathrm{dc}。 $$ (5)在方程式中。 5、一个缩放参数(也可以吸收β 和 λ 前置因子中的项)和直流偏移作为拟合参数插入。平均寿命为
$$ \left\langle {\tau}_{SE}\right\rangle =\frac{\tau_{SE}}{\beta}\Gamma \left[\frac{1}{\beta}\right], $$ (6)其中 Γ 表示 Gamma 函数,平均衰减时间为
$$ \left\langle t\right\rangle ={\tau}_{SE}\frac{\Gamma \left(2/\beta \right)}{\Gamma \left(1/\beta \right)} . $$ (7)在以前的许多工作中,通常使用“标准”拉伸指数 exp[− (λ 东南 t ) β ] 模拟发光衰减而不是种群衰减。因此,我们有一个归一化的强度函数由
给出 $$ g(t)=\frac{\lambda_{SE}\beta }{\Gamma \left(1/\beta \right)}\exp \left[-{\left({\lambda}_{SE} t\right)}^{\beta}\right]。 $$ (8)等式 8 被归一化,使得 t 之间的积分 =0且∞等于1。对应的拟合模型很简单
$$ {I}_t=A\exp \left[-{\left({\lambda}_{SE}t\right)}^{\beta}\right]+\mathrm{dc}。 $$ (9)等式 9 被广泛应用并且通常非常适合 SiNC 发光数据,尽管事实上(如等式 4)等式。对于 100% 的绝对量子效率 (AQY),8 是严格归一化的。一个经常被忽视的一点是人们无法提取τ 东南 (=1/λ 东南 ) 和 β 从由方程建模的发光衰减。 9 并使用它们来计算方程的平均时间。 6 和 7。本质上,方程。图 4 和图 8 是不同的强度衰减模型,预计会有不同的种群衰减函数、平均时间和衰减率分布。
为了找到会导致由方程给出的强度函数的种群衰减。 9,我们应用我们从方程中得到的相同过程。 4 到等式。 5,但反过来,就是:
$$ \frac{c_t}{c_0}=1-\frac{\lambda_{SE}\beta }{\Gamma \left(1/\beta \right)}{\int}_0^t\exp \left[ -{\left({\lambda}_{SE}t\right)}^{\beta}\right]\cdot \mathrm{dt}。 $$ (10)经过几个步骤,方程的解决方案。 10 是
$$ \frac{c_t}{c_0}=\frac{1}{\Gamma \left(1/\beta \right)}\Gamma \left[1/\beta, {\left({\lambda}_{ SE}t\right)}^{\beta}\right]。 $$ (11)方程 11 是从方程给出的强度衰减得到的种群衰减。 8. 以通常的方式求平均寿命导致
$$ \left\langle {\tau}_{SE}\right\rangle ={\tau}_{SE}\frac{\Gamma \left(2/\beta \right)}{\Gamma \left(1 /\beta \right)} $$ (12)和平均衰减时间
$$ \left\langle t\right\rangle ={\tau}_{SE}\frac{\Gamma \left(3/B\right)}{2\Gamma \left(2/\beta \right)} . $$ (13)最后,频率分布是 (1 /λ )·H (λ ),其中和以前一样,H (λ ) 是参考文献中计算的分布。 [9] 对于方程给出的种群衰减。 3. 这些结果总结在表1中。
图>两个 SE 公式之间的差异是显着的(图 1)。在文献中,人们经常发现强度衰减由A·建模 exp[− (t /τ 东南 ) β ] + dc(即方程 9),然后使用方程计算平均时间。 6 和 7。这在数学上似乎是不正确的,因为方程。图 6 和图 7 源自等式给出的强度衰减。 4,不等式。 8. 例如取τ 东南 =100 μs 和 β =0.7,如图 1 所示,对于由 exp[− (t /τ 东南 ) β ],我们发现平均时间常数为 199 μs(方程 12),而使用方程为 127 μs。 6. 平均衰减时间也有类似的差异(方程 7 和 13)。此外,还有一种称为 Higashi-Kastner 方法的方法,用于估计特征寿命 [19],该方法已应用于 SiNC,作为应用 SE 衰减模型的替代方法 [20, 21]。在该模型中,特征延迟时间 t d , 简单地将衰减数据的峰值绘制为 I t ·t 对比 t .这被建议等价于 (1 /β ) 1 /β ·τ 东南 从方程获得。 9 [20]。
<图片>结果与讨论
基本特征
由于与 SiNC 相关的低对比度和来自无定形碳载体的重叠斑驳对比度,无法应用使用明场图像的基于计算机的粒子计数算法,并且必须使用像素计数软件“通过肉眼”估计直径(样品明场 TEM 图像如图 2a、d 所示,手动粒子计数结果与对数正态分布(图 2c、f)拟合,以获得 2.9 nm 的线性平均直径(平均值和标准偏差自然对数μ =1.057 和 σ =0.1555) 和 5.4 nm (μ =1.663 和 σ =0.1917),分别为 1100 和 1200 °C 的退火温度。这些样品此后将被称为“小”和“大”SiNC。通过所选 NC 的高分辨率成像进一步检查尺寸(图 2b、e),其中晶格条纹可用作识别 NC 并估计其直径的另一种方式。傅里叶变换红外 (FTIR) 光谱和 XPS 数据表明制备的 SiNCs 成功地被十二烯官能化;然而,小SiNCs比大SiNCs氧化得更多,因此显示出较小的功能化程度(附加文件1:图S1和S2)。
<图片>结论
The most common models used for SiNC luminescence decay were described theoretically. The population decay corresponding to the “simple” stretched exponential luminescence decay, exp[− (t /τ ) β ], was derived and expressions for the characteristic mean times were found. This model was compared against the alternative model in which the population decays according to the simple SE. Two dodecene-functionalized SiNCs samples were then prepared from thermal nucleation and growth, followed by etching and alkane surface functionalization. These samples consisted of particles with mean diameters of 2.9 and 5.4 nm, respectively. The basic PL spectrum and TRS was measured using standard methods. The TRS data were fit with several distributions in order to establish whether any of them can be considered “true” and to find which one yields the best fit. While the simple SE luminescence decay fits the TRS data reasonably well, the distribution of residuals shows that it is not strictly accurate. None of the fitting models fully captures the shape of the measured decay rate distribution; they also show large deviations in the peak position and the shape of the distribution, as well as disagreement in the average time constants. Furthermore, the ensemble mean time constants were dependent on the responsivity curve of the detection system. This leads to serious questions about how to interpret the PL decay from ensembles of thermally-grown SiNCs.
Quadrature frequency-resolved spectroscopy was then employed with the intent to find the lifetime distribution directly for SiNC ensembles formed by thermal annealing of a base oxide. The spectrum was found to be not much wider than the intrinsic QFRS response function, requiring a deconvolution in order to extract the SiNC rate distribution. This yielded a distribution whose shape was nearly symmetrical (on a semilog scale) for the small NC sample and about half a decade wide, whereas it was slightly more skewed for the large NCs. We find that FRS techniques are suited to the study of SiNC luminescence dynamics and, after deconvolving the system response from the data, FRS yields the decay rate distribution directly. The most significant problem is the required deconvolution, but the Richardson-Lucy method was found to produce fairly robust results. While the detector response function can in principle be corrected from the FRS data, there is no simple means to do this for wide-PL-band TRS data. Still, as long as the data compared are from the same detector then the results should at least be internally meaningful. Hopefully in the future, these issues will be more fully considered when analyzing inhomogeneously broadened NC luminescence lifetimes, rather than defaulting to the simple stretched exponential model (Eq. 9) to describe and characterize the dynamical processes at work in the PL spectrum.
方法
The SiNCs were synthesized according to a recently-proposed method [21]. Briefly, 4 g of hydrogen silsesquioxane (HSQ) was annealed at 1100 or 1200 °C for 1 h in a flowing 5% H2 + 95% Ar atmosphere, resulting in composites of SiNCs embedded in a silica matrix. These composites were mechanically ground into a fine powder using an agate mortar. The powder was shaken for about 8 h with glass beads using a wrist action shaker. The powders were suspended in 95% ethanol and interfaced to a vacuum filtration system equipped with a filter. To liberate the H-SiNCs, the silica matrix was removed via HF etching. An approximately 200 mg aliquot of the composite was transferred to a Teflon beaker to which 2 mL of ethanol, 2 mL of water, and 2 mL of 49% HF aqueous solution were added in order to dissolve the silica matrix. After stirring the suspension for 40 min, the liberated H-SiNCs were extracted as a cloudy yellow suspension using toluene and isolated by centrifugation at 3000 rpm for 5 min. The resulting hydrogen-terminated SiNCs were suspended in 10 mL dry toluene, and then transferred to an oven-dried Schlenk flask equipped with a magnetic stir bar. Subsequently, 1 mL of 1-dodecene (ca. 4.6 mmol), as well as 20 mg of AIBN were added. The suspension was subjected to three freeze-pump-thaw cycles using an Ar charged Schlenk line. After warming the suspension to room temperature, it was stirred for 24 h at 70 °C, and 10 mL of methanol and 20 mL of ethanol were subsequently added to the transparent reaction mixture. The resulting cloudy suspension was transferred to a 50 mL PTFE vial and the SiNCs were isolated by centrifugation at 12,000 rpm for 20 min. The SiNCs were re-dispersed in 10 mL toluene and isolated by addition of 30 mL ethanol antisolvent followed by another centrifugation. The latter procedure was carried out one more time. Finally, the dodecyl-SiNCs were re-dispersed in 5 mL dry toluene and stored in a screw capped vial (concentration ~ 0.5 mg/mL) for optical studies.
TEM samples were prepared by depositing the freestanding nanoparticles directly onto an ultrathin (ca. 3 nm) carbon-coated copper TEM grid. The NCs were imaged by bright-field TEM using a JEOL JEM-2010 and HRTEM was done on a JEOL JEM-ARM200CF. Fourier transform infrared spectroscopy (FTIR) was performed in a Nicolet 8700 from Thermo Scientific. X-ray photo-electron spectroscopy was measured in a SPECS system equipped with a Phoibos 150 2D CCD hemispherical analyzer and a Focus 500 monochromator. The detector angle was set perpendicular to the surface and the X-ray source was the Mg Kα line.
Luminescence spectra were excited with a 352-nm Ar + ion laser, which was pulsed (50% duty cycle, 50–250 Hz) using an Isomet IMDD-T110 L-1.5 acousto-optic modulator (AOM) with a fall time of ~ 50 ns. The used setup is schematically depicted in Fig. 7. The laser beam passes the acousto-optic modulator and one of the diffracted beams is selected by an iris. A beamsplitter reflects the main part of the pulsed laser beam into the sample cuvette and the incident power on the sample was ~ 8 mW spread over an area of ~ 4 mm 2 . The luminescence was collected with an optical fiber (numerical aperture 0.22), sent through a 450-nm longpass filter and is guided to the appropriate detector. The PL spectrum was measured by an Ocean Optics miniature spectrometer whose response function was corrected using a calibrated radiation source (the HL-3 + -CAL from Ocean Optics). The quantum efficiency was measured using an integrating sphere with 405-nm excitation, using a solution diluted to have an absorbance of ~ 0.15 at that wavelength.
Diagram of the experimental setup. M mirror, AOM acousto-optic modulator, BM beamsplitter, PD photodiode, MC monochromator, S spectrometer, PMT photomultiplier tube, APD avalanche photodiode, LIA lock-in amplifier
图>The luminescence dynamics were measured with two different detectors. The first detector was the Thorlabs 120A2 avalanche photodiode (50 MHz roll-off), which was interfaced to a Moku:Lab (200 MHz) in digital oscilloscope mode. The second detector was a Hamamatsu h7422-50 photomultiplier tube interfaced to a Becker-Hickl PMS400 multiscalar. The error in the luminescence decay times was obtained by repeating the measurements three times, yielding a standard error in the mean lifetime calculated using the stretched exponential fit (Eq. 4) of 1 μs. All fits to the decay data were done in Origin using the least linear squares with the Levenberg-Marquardt algorithm, and were repeated in Matlab using the same method. For wavelength-dependent decay measurements, the luminescence was sent through an Acton MS2500i monochromator prior to detection, with the half width of the detected radiation set to ~ 3 nm.
For QFRS measurements, the AOM was set to produce a sinusoidal oscillation. A part of the incident beam was deflected into a Thorlabs PDA10A photodiode (200 MHz) in order to generate the reference signal. The SiNC PL response was simultaneously collected and sent to the APD. The reference signal was obtained using the beamsplitter, and along with the corresponding PL signal, was analyzed using the Moku:Lab in the lock-in amplifier mode to measure the in-phase and quadrature components of the signal.
Finally, we also searched for a short-lifetime component in the luminescence, as has sometimes been reported previously and attributed to oxidation [22]. This system used a 405-nm picosecond diode laser (Alphalas GmbH) to excite the NCs, and a Becker-Hickl HPM-100-50 PMT interfaced to an SPC-130 pulse counter system. This setup has a response time of ~ 100 ps. No evidence of a nanosecond decay was observed in these SiNCs.
缩写
- APD:
-
Avalanche photodiode
- AQY:
-
Absolute quantum yield
- FRS:
-
Frequency-resolved spectroscopy
- LN:
-
Lognormal
- NC:
-
纳米晶
- PL:
-
光致发光
- PMT:
-
Photomultiplier tube
- QFRS:
-
Quadrature frequency-resolved spectroscopy
- SE:
-
Stretched exponential
- SiNCs:
-
Silicon nanocrystals
- TRS:
-
Time-resolved spectroscopy
纳米材料