Volume 10 Issue 6 - September 11, 2009 PDF
High-Speed GaN-based Green Light Emitting Diodes with Partially n-doped Active Layers and Current-Confined Apertures
J.-W. Shi, Jinn-Kong Sheu*, C.-H. Chen, G.-R. Lin, W.-C. Lai

Institute of Electro-Optical Science and Engineering, College of Sciences, National Cheng Kung University

IEEE Electron Device Letters, Vol. 29, No. 2, 2008.pages:158-160

Font Enlarge
High-speed red light-emitting diodes (LEDs) often serve as key transmitter components for communication networks using plastic optical fibers (POFs). According to the media-oriented systems transport (MOST) standard they must run at a data rate of 22.5Mb/s.  However, the optical bandwidth of the red operating window (~650nm) of the current commercially used polymethylmethacrylate (PMMA) POF is narrower, and the propagation loss higher (0.125dB/m vs. 0.09dB/m) than that of the high-performance POF system. This operates under a PMMA loss minimum window, which has a wavelength of around 500nm.  The III-nitride-based green LEDs would thus be a more promising choice for such applications.  Although the material dispersion of PMMA fiber at around 520nm wavelength is higher (1.5 times higher) than at 650nm wavelength, the dispersion-free window in total dispersion spectrum, which includes waveguide dispersion and material dispersion, may be shifted to 520nm by optimizing the index profile of PMMA fiber.  In this paper we demonstrated an InXGa1-XN/GaN-based high-speed green LED with a current-confined aperture and partial n-type doping InXGa1-XN/GaN based multiple-quantum-wells (MQWs).  With this device, one can achieve an extremely high electrical-to-optical (EO) 3-dB bandwidth (~330MHz), limited only by the spontaneous lifetime of our MQWs, with a reasonable POF coupling power (~264μW) suitable for POF communications.

The LED wafers were grown by metal-organic-chemical-vapor-deposition (MOCVD) on a sapphire substrate.  The active region of the LED was a ten-period InXGa1-XN/GaN based MQW structure, which consists of a 135Å-thick GaN barrier layer and a 25Å-thick InXGa1-XN well layer in each period.  The n-type doping density was around 5×1017cm-3 in seven of the GaN barrier layers near the n-type cladding layer.
Fig.1 A conceptual cross-sectional view of the demonstrated LED. The inset shows the top-view of device after zooming in on the active current-confined aperture. For clearness, figure is not drawn according to scale.
  Conversely, the barrier layers near the p-type cladding layer were left undoped.  Although n-type doping in the InXGa1-XN/GaN based MQW layers can improve the material quality and the radiative recombination rate in MQWs, this approach may result in most of the depletion region being located in the p-type GaN layers due to that the doping density in the MQW layers is comparable to the typical ionized doping density of the p-type GaN layer.  By inserting three undoped MQW periods near the p-side of the active region, the external electric field can be concentrated more in the active layers than is the case when fully n-type doped MQWs are used in the active layers.  For high-speed applications, a current-confined aperture was realized by etching a small mesa on the topmost p-type GaN layer to restrict the current-flow path and reduce the capacitance. The use of a current-confined aperture can eliminate etching induced damage on the MQW sidewall and its influence on the internal quantum efficiency of LEDs, such as what occurs with direct etching through the active MQW layers to reduce capacitance. A conceptual cross-sectional view of demonstrated device and top-view of current-confined aperture is given in Figure 1 and its inset.  In order to study the influence of the diameter of the current-confined aperture on the speed and output power performance of the LEDs, devices with different diameters of current-confined aperture and active areas were fabricated.  The epi-layers and geometric structure of Devices A, B, C, and D were similar, with only the ratio of the diameter of the current-confined aperture to that of total circular active mesa of the bottom side being varied.  The ratios of devices A and B are 1:1 and 1:2, respectively.  For devices C and D, the ratio is the same at 1:3.  The diameter of the circular active area below the p-type GaN layer is 228μm, the same as for devices A, B, and C.  Device D has a diameter of 132μm.  As shown in Figure 1, the LED unit was integrated with two co-planar waveguide (CPW) pads, which were parallel to uniform current distribution during DC measurement.  Alternatively, for the AC measurement one of the CPW pads was left as radio-frequencies (RF) open.  Details of the fabrication processes of the demonstrated devices can be referred to our previous work.

Fig.2 The output optical power (P) vs. the bias current (I) of the devices (A, B, C, and D) measured with different diameters of current-confined apertures. The two insets show the measured I-V curves of devices A to D and electroluminescence (EL) spectra of device A measured under different bias currents.
During DC measurement the output power of the LED was collected by use of an integrating sphere or a POF (2 mm in diameter and with a 0.485 N.A), which served as an optical probe.  All four devices exhibited similar bias dependent electroluminescence (EL) spectra behavior with a similar center wavelength (~510nm).  The inset of Figure 2 shows the EL spectra of device A measured under different bias currents.  A slight blue-shift of the center wavelengths (~510nm to ~500nm) was observed when the bias current increased from 10mA to 100mA, which can be attributed to the screening of the piezoelectric field and the renormalization of quantized states in the well region under a forward bias.  Another inset of Figure 2 shows the current-voltage (I-V) curves of devices A, B, C, and D.  The turn-on voltages of these four devices were all around ~3V. The values of the differential resistances of each device are also specified in this inset.  We can clearly see, via the difference in the cross-sectional areas of the current-flowed paths, that device A has a much lower differential resistance than device D.  Figure 2 shows the optical power (P) collected by use of the integrating sphere vs. injected current (I) for the four devices. By measuring the difference in collected optical power, obtained using an integrating sphere and a POF, we can obtain the coupling efficiency of the devices: this was the same for all four devices, around 13.4%, which is close to the reported coupling efficiency (~11%) for high-speed GaN based green LEDs to POFs.  The similar coupling efficiency of all our devices can be ascribed to that their output far-field is similar and can be approximated by an ideal Lambertian far-field distribution regardless of the diameter of the light-emitting aperture. Even device C, with its small light-emitting area, had a coupled optical power (185μW under 50mA) to the POF large enough for POF communication.
Fig.3 The EO frequency responses of (a) device A and (b) device B measured under different bias currents.
Using the same measurement setup and the same bias current (~100mA) as in our previous work, we found the POF coupled optical power of the current LED (Device B) to be much higher (343μW vs. 170μW) than that of our previous fully n-type doped MQW green LED, even though it had a larger active area (~14000μm2 vs. ~10000μm2).  As can be seen in Figure 2, the P-I measurement results are superior to those found in our previous work, which implies the advantages of using partially n-doped MQW structures.  During the AC measurement we injected the RF signal into the LEDs. The modulated optical power was collected by the POF and then fed into a low noise Si based photo-receiver (New Focus: 1801) that was connected with a RF spectrum analyzer.  We only de-embedded the frequency response of the photo-receiver because the microwave loss for the K-band passive RF components used in the frequency range of interest (hundreds of MHz) for LED measurement is negligible.  Figures 3 (a) and (b) and 4 (a) and (b) show the measured electrical-to-optical (EO) frequency responses under different bias currents obtained for devices A , B, C, and D.  The maximum bias current is limited by device failure.    We can clearly see that the measured 3-dB bandwidth of each device increases significantly with a bias current density.  This phenomenon has been reported for other LEDs and can be attributed to the bi-molecular recombination probability being proportional to the injected carrier density into the active volume.  Devices C and D have a bias current of around 100mA bias and a very high current density (>2.2kA/cm2), and both could exhibit a 3-dB bandwidth as high as 330MHz.  The coupled optical power of device C under high-speed operation was around 264μW.  To the best of the authors’ knowledge, this is the highest 3-dB modulation bandwidth (330MHz) ever reported for III-nitride based LEDs.  Such a value is comparable with the highest reported 3-dB bandwidth (300MHz~500MHz) for AlGaInP based red resonant cavity LEDs (RCLEDs).
Fig.4 The EO freguency responses of (a) device C and (b) device D measured under different bias currents.
There are two major factors limiting the speed of LEDs: one is the RC time constant and the other is the spontaneous recombination time.  In order to clarify which one caused the significant bandwidth enhancement observed for devices C and D, we performed capacitance-voltage (C-V) and current-voltage (I-V) measurements to obtain the RC-limited bandwidths of the devices.  The extracted RC-limited bandwidth, which includes the 50Ωsource resistance, of devices A, C, and D were around 150MHz, 1GHz, and 3GHz, respectively.  Such a result indicates that the current-confined aperture can effectively reduce the diffusion capacitance and improve the RC-limited bandwidths.  According to the extracted RC-limited and measured 3-dB bandwidths of the devices, we can conclude that under a high bias current density, the dominant bandwidth limiting factor of device A is the RC delay time, and for devices C and D it is the spontaneous recombination time and the maximum allowable bias current density.  Based on the measured 3-dB bandwidth (330MHz), the extracted recombination time will be around 0.5ns [9], which is a reasonable number and comparable to the measurement results (~1ns) for fully n-type doped InGaN/GaN based MQWs obtained using time resolved photoluminescence technique.
< Previous
Next >
Copyright National Cheng Kung University