1. Introduction

Silicon-based light emitting materials have been a focus of research in the field of silicon photonics because of the potential impact of a silicon-based laser [1,2]. Due to its indirect bandgap, bulk silicon is a poor light emitter. Thus, the development of an electrically pumped, complimentary-metal-oxide-semiconductor (CMOS) compatible silicon-based laser or light emitting device (LED) presents a challenge. Nonetheless, research has demonstrated that quantum-confined silicon is capable of emitting light at room temperature [3–8]. Silicon nanostructures, such as quantum wells, wires, and dots (called nanocrystals), are all a means for attaining quantum confinement in silicon. Among the most promising of these materials are silicon nanocrystals (Si-nc) embedded in SiO₂ [5], because researchers have observed optical gain [9] and stimulated emission [10].

While many fabrication techniques exist for silicon nanocrystals, the most common approaches employ a silicon rich oxide (SRO) or Si/SiO₂ superlattice deposited by plasma enhanced chemical vapor deposition (PECVD) and annealed at a high temperature to precipitate silicon nanocrystals in the host material [4–6]. Several research groups investigated the electroluminescence (EL) and current injection properties in Si-nc-based devices [11–13]. These devices are typically biased with DC signals, although the insulating nature of the Si-nc material makes conduction difficult and can require high voltages to achieve light emission. Light emission under DC excitation is typically associated with an impact excitation process, which can cause damage to the insulating material resulting in devices that are inefficient, unstable, and ultimately unreliable. Pavesi, et al. demonstrated that bipolar injection at low voltages results in devices that are more efficient although still biased under DC conditions [14]. Walters et al. demonstrated that bipolar injection could be achieved using an alternating or pulsed bias voltage and termed the resulting emission “field-effect luminescence” [15,16]. This type of electrical pumping allows the devices to be driven at voltages much lower than necessary under DC pumping. This presents an opportunity to make silicon-based light emitting devices more efficient and reliable.

In this work, we present the results of experiments demonstrating bipolar injection into silicon nanocrystals within a Si/SiO₂ superlattice deposited by PECVD. We study the resulting EL as a function of frequency and time in order to understand the behavior of the devices. We analyze time-resolved EL measurements to help explain the frequency dependence exhibited by the LEDs, and finally, we compare the light output of a device under pulsed excitation to the light output under DC excitation to illustrate the benefit of a pulsed pumping scheme.

2. Experimental details

Metal-oxide-semiconductor (MOS) devices were fabricated on a p-type silicon substrate having a resistivity of ~0.01 Ωcm. Silicon nanocrystals embedded in SiO₂ served as the oxide material for the MOS devices. A 20-period superlattice consisting of amorphous Si (a-Si) and SiO₂ was deposited by PECVD with each period ~3-4 nm thick for a total thickness of ~60 nm. The material was then annealed at 1100°C for 1 hour in an N2 ambient to precipitate the nanocrystals. More details on the nanocrystal material can be found in another work [17]. A 100 nm indium tin oxide (ITO) layer was deposited via sputtering and circular contacts with 300, 400, and 500 µm radii were formed using a liftoff process.

Photoluminescence (PL) measurements were excited using a 532 nm laser. Spectral PL and EL measurements were collected from the top of the device using an iHR320 spectrometer and an electrically cooled silicon CCD from Horiba Jobin Yvon. Time resolved EL measurements were collected from the top of the devices using a Hamamatsu photomultiplier tube and analyzed using a Tektronix TDS6604 oscilloscope. All measurements were taken at room temperature.

When probing the devices, forward bias was achieved by applying a positive voltage to the substrate and reverse bias was achieved by applying a positive voltage to the ITO. An Agilent 33220A function generator and a Tegam 2340 high voltage amplifier were used to apply square wave biases ranging from 30 V to 80 V peak-to-peak.

For optical power measurements, a two-lens system with a numerical aperture of 0.5 was positioned above the devices to collect the emission. The emission was then detected with a model 818 SL silicon detector head connected to a Newport model 1830-C optical power detector.

3. Results and discussion

Previously, we demonstrated EL from a Si-nc material in which the application of a DC voltage excited the EL signal [17]. In this work, we demonstrate bipolar injection into the same Si-nc material using a pulsed pumping scheme. By applying square wave voltages at varying frequencies, we are able to sequentially inject opposite carriers into the silicon nanocrystal material, which subsequently recombine radiatively. Figure 1 shows the measured PL and EL from the prepared devices. The EL was excited using a square wave with a root mean square (rms) voltage of 37 V and a frequency of 1 kHz. The overlap of these two curves indicates that the light emitting mechanism responsible for the EL generated by the pulsed pumping scheme is the same as that of the PL.

 

Fig. 1 Electroluminescence from pulsed excitation is similar to photoluminescence of the silicon nanocrystal devices.

Download Full Size | PDF

In Fig. 2 , we present the time-resolved voltage and EL signals. As seen in the figure, a spike in the EL from the device occurs at each voltage transition of the applied square wave. The observed behavior can be explained as follows [18,19]: The forward bias portion of the applied voltage drives the device into accumulation such that holes are injected into the nanocrystals. When the voltage switches to reverse bias, the device is driven into inversion, and electrons are injected. Hence, during the forward-to-reverse transition, electrons are injected into hole-charged nanocrystals, and the opposite occurs during a reverse-to-forward transition. Each of the transitions results in a spike in the EL signal as carriers recombine radiatively inside the nanocrystals. We found the decay time of the EL pulses to be ~50 µs. Also, the peak of the EL signal after forward-to-reverse transitions was consistently higher than the peak after reverse-to-forward transitions. Walters, et al. suggest that this occurs due to a difference in the tunneling times of electrons and holes [16].

 

Fig. 2 A time-resolved plot of the alternating driving voltage (red) and resulting electroluminescence (black) of the silicon nanocrystal LEDs.

Download Full Size | PDF

In order to verify that a bipolar effect is occurring (as opposed to injection of a single carrier type which results in luminescence), we reduced one side of the applied voltage to zero while continuing to monitor the time-resolved EL signal. Figure 3 shows that as either the positive or negative side of the applied voltage was reduced to zero, the EL signal diminished. We observed that for electric fields below 6 MV/cm (30 volts for these devices), there was no measurable EL without a bipolar voltage transition. Bipolar injection allows us to generate EL at much lower voltages, which helps reduce damage to the material from impact excitation processes and improve reliability.

 

Fig. 3 The applied voltage (above) and corresponding EL signal (below). The EL signal diminished as the negative portion of the bias was reduced. The behavior was the same when the positive portion of the bias was reduced (not shown).

Download Full Size | PDF

Based on the time-dependent behavior observed in Fig. 2, we can consider optimizing the integrated EL output by increasing the driving frequency of the device. For low frequencies, we would expect a doubling of the driving frequency (producing twice as many EL pulses per unit time) to result in a doubling of the integrated EL output. This behavior should continue until reaching an optimal frequency at which a half-period of the driving voltage becomes equivalent to the decay time of the device. Beyond this frequency we would expect the integrated EL to saturate. Based on these assumptions, we would expect the optimal driving frequency to be 10 kHz, due to the observed 50 µs decay. Figure 4 shows the integrated EL output as a function of frequency, which we measured by integrating all EL pulses during a fixed period of time while changing the frequency of the driving voltage. At low frequencies (2 – 1000 Hz), the integrated EL signal does not follow the linear behavior described previously, but instead, it follows a square root dependence on frequency. Additionally, the optimal frequency is found to be close to 50 kHz, which is faster than expected.

 

Fig. 4 Frequency dependence of pulsed EL output for silicon nanocrystal devices.

Download Full Size | PDF

In order to understand the frequency-dependent behavior observed in Fig. 4, we measured the time-resolved EL signal over one period of the supply voltage for frequencies between 50 Hz and 25 kHz. Figure 5(a) only shows the results for EL pulses corresponding to forward-to-reverse transitions, but the behavior was the same for reverse-to-forward transitions. At frequencies below 5 kHz, each EL pulse decayed exponentially from a maximum and was reduced by at least a factor of e prior to the next voltage transition. Above 5 kHz, the frequency became too fast for each pulse to decay by at least a factor of e; however, a peak still occurred followed by a partial decay of the luminescence. At 50 kHz (not shown), the limited response time of the detector resulted in a flat EL response with no peak or decay. Beyond 50 kHz, the EL response remained flat and the signal began to decrease. We fit the decay time of the EL signal at each frequency using a single exponential decay, and the results are shown in Fig. 5(b). The decay time decreased from ~70 µs at 50 Hz to less than 30 µs at 5 kHz. The decay time of the 10 and 25 kHz signals could not be extracted due to the limited response time of the detector.

 

Fig. 5 Frequency dependence of the time-resolved EL signal (a) and decay time of the EL pulse (b). The peak and decay time of the signal both decrease with frequency.

Download Full Size | PDF

Two important observations help explain the frequency-dependent behavior of the integrated EL output. First, the peak in the EL pulse decreases with frequency. This effect explains the lack of a linear increase in the integrated EL at low frequencies, because the decreasing strength of EL pulses works against the number of pulses that are integrated with increasing frequency. The decrease in the peak of the EL pulse is related to the charging time of the nanocrystals [20]. Second, the decay time of the EL signal also decreases with increasing frequency. Higher frequencies elicit a response from a smaller size distribution of nanocrystals [21], which have a faster decay time [20]. The decrease in the decay time of the EL signal allows the balance between the speed of the alternating voltage and the decay of the EL to occur at frequencies higher than expected. Thus, the frequency dependent time-resolved behavior of the EL signal verifies the integrated EL measurement.

In order to quantify the advantage of a pulsed pumping scheme over a DC pumping scheme, we measured the light output as a function of voltage for DC pumping and the light output as a function of rms voltage at 1 kHz for pulsed pumping (Fig. 6(a) ). We measured the output power using a separate optical power detector that required the measurements to be taken at 1 kHz due to speed limitations. For rms voltages below 30 V, the corresponding DC current that would flow through the devices was less than 0.15 µA/cm₂, and the corresponding DC EL output was less than 20 pW. As the voltage increased sufficiently to allow for a significant DC current, the DC driven output and the pulsed driven output began to converge. In Fig. 6(b), we present the output power enhancement in the pulsed pumping scheme as compared to the DC pumping scheme, and we note a maximum enhancement factor of 18 at 30 V. At higher voltages, this enhancement decreases due to the increasing amount of DC current flowing through the device. However, high DC driving voltages and current densities can hurt the long-term performance of Si-nc LEDs. Therefore, a pulsed pumping scheme offers an opportunity to increase the reliability of such devices by improving performance sufficiently to allow for operation at lower voltages.

 

Fig. 6 The output power as a function of voltage for a pulsed driven device and a DC driven device begin to converge at high voltages (a). In the pulsed case, the voltages correspond to the rms voltage of the square wave. Under pulsed driven excitation, the device emits more than 18 times the amount of light as under DC excitation (b).

Download Full Size | PDF

4. Conclusion

In this work, we demonstrated bipolar injection into silicon nanocrystals within a Si/SiO₂ superlattice deposited by PECVD using a pulsed pumping scheme. We explained the frequency dependence of the devices by analyzing the time-resolved EL signal. The integrated output of the devices peaked at an operating frequency of 50 kHz. We compared the output power of the pulsed driven devices to that of DC driven devices and found that the output power increased by a factor of 18 at low voltages. The presented pulsed pumping approach increases the reliability and performance of Si-nc based LEDs.