Design and optimization of high temperature optocouplers as galvanic isolation

The commercial InGaN-based (blue and green) and AlGaInP-based (red) multiple quantum well (MQW) lighting emitting diodes (LEDs) were studied in a wide range of temperatures up to 800 K for their light emission and detection (i.e., LEDs operated under reverse bias as photodiodes (PDs)) characteristics. The results indicate the feasibility of integrating a pair of selected LEDs to fabricate high temperature (HT) optocouplers, which can be utilized as galvanic isolation to replace the bulky isolation transforms in the high-density power modules. A detailed study on LEDs and PDs were performed. The external quantum efficiency (EQE) of the LED and PDs were calculated. Higher relative external quantum efficiency (EQE) and lower efficiency droops with temperatures are obtained from the blue and green LEDs for display compared with the blue one for lighting and red LED for display. The blue for lighting and red for display devices show superior responsivity, specific detectivity (D*), and EQE compared with blue and green for display when operated as PDs. The results suggest that red LED devices for display can be used to optimize HT optocouplers due to the highest wavelength overlapping compared with others.


Results
LEDs characterizations. Figure 1a shows a typical temperature-dependent EL spectra of the BD LED in temperature ranging from 77 to 800 K with an injection current of 0.3 mA to avoid the self-heating of the LEDs with a turn-on voltage of under 2.0 V 22,23 . The tail of the EL spectra on the high-energy region (i.e., lower wavelength) became broader as the temperature was increased due to the increasing number of the electronhole pairs in the conduction and valance bands at elevated temperatures. Although the EL spectra decrease significantly at high temperatures, a clear EL signal is observed up to 800 K. Other LEDs (BL, GD, and RD) show similar behaviours in terms of dropping in the EL intensity and boarding the EL spectra as the temperature is increased. It is observed the peak wavelength is temperature-dependent as well, and it is varied depending on the device type, as shown in Fig. 1b. The temperature-dependence peak wavelength of LEDs influences the paring of LEDs and PDs. The InGaN-based LEDs show less temperature-dependent peak wavelength than the AlGaInPbased LEDs. Samples BL, BD, and GD and exhibit blueshifts at low temperatures. A redshift is observed above 300 K. On the other hand, only a redshift is observed in sample RD. As the temperature is increased from 77 to 800 K, the peak wavelength changes by 19,15,13, and 85 nm for samples BL, BD, GD, and RD, respectively. Figure 2a shows temperature-dependent EQE for sample BD. The EQE is normalized by the peak EQE at 77 K. The EQE is increased by increasing the current density, and it is decreased by increasing the temperature. A similar trend is observed for the other LED devices. The relative EQE of sample B is reduced by 80 % by increasing the temperature from 77 to 800 K.
The current-dependent efficiency droop, the so-called "J-droop, " and the temperature-dependent efficiency droop, often referred to as the "T-droop. " Both the J-droop and T-droop are essential factors determining the performance of LEDs, so there is a need to investigate and compare the InGaN-based blue for light and display and green for display LEDs and the AlGaInP based red LED. The J-droop and the T-droop can be calculated as follow 24 : The InGaN based BL, BD, and GD LEDs and the AlGaInP based RD LED are compared. Figure 2b shows the efficiency droop in the LEDs, calculated from the EQE 24 . The J-droops of the BL, BD, GD, and RD are 0.03, PDs characterization. The dark I-V characteristics of BL, BD, GD, and RD are studies, and they show similar behaviour with increasing the temperature. The typical temperature dependence dark I-V characteristics of BL as an inset in Fig. 3a. The leakage currents originate from surface recombination and bulk leakage. Leakage mechanisms consist of thermally excited carriers, generation-recombination in the depletion region, and tunnelling. Dark current generally increases with both applied bias voltage and with temperature. Moreover, the dark current flows through the same path as the photocurrent, so it will also have a shot noise term correlated with it. Since photodetectors contain parallel resistances, which are junction leakage resistances, there will be thermal noise present in the dissipative resistances of the photodetectors. As the temperature increases, the thermally excited carriers increase, leading to the rise of thermal noise. At high temperatures, the dark currents become a significant factor in evaluating the performance of devices, so it is necessary to understand the behaviour of the dark currents as temperature increases.  www.nature.com/scientificreports/ Figure 3a shows the temperature dependence leakage current density of the samples BL, BD, GD, and at − 9 V bias voltages (except for the RD, which is at − 7 V bias). BL shows the lowest leakage current density, while RD demonstrates the highest leakage current density as the temperature increases from 77 to 800 K. The leakage current is a function of both temperature and applied bias, implying a combination of tunnelling and thermally generated currents. The leakage current density of devices is increased by around four orders of magnitude by increasing the temperature from 77 to 800 K. It is also observed that a rapid increase (two orders of magnitude) in the leakage current density of the InGaN-based samples when the temperature is increased from 100 to 200 K. This rapid increase in the leakage current is due to the thermal ionization of carriers from deep traps and trap assisted tunnelling process.
The temperature-dependent thermal noises of the devices are plotted in Fig. 3b. GD exhibits the lowest thermal noise, while BL shows the highest thermal noise. Both Samples BD and GD reveal an increase in thermal noise by three orders of magnitude, while BL and RD show a rise in the thermal noise as the temperature increases from 77 to 800 K. Figure 4a the temperature-dependent responsivity of all samples is plotted at 77 K (lowest temperature), 300 K (room temperature), 500 K (targeted operating temperature for the optocoupler), 800 K (highest temperature). Higher spectral response is noted from BL and RD. As the temperature increases, a redshift in the spectral response is more pronounced in BL and RD. The room temperature (RT) peak responsivity is 405, 418, 443, and 620 nm for BL, BD, GD, and RD, respectively. The peak position for sample BL changes from 383 to 428 nm by increasing the temperature from 77 to 800 K. The responsivity (@ 405) increases from 0.029 to 0.097 A W −1 by elevating the temperature from 77 to 600 K and decreases to 0.056 A W −1 at 800 K. The peak location for sample BD shifts from 402 to 418 nm by increasing the temperature 77-400 K and then moves to 394 nm at 600 K. The maximum responsivity from this device is 0.037 A W −1 at RT. The peak responsivity of sample GD changes from 430 to 450 nm by raising the temperature from 77 to 700 K and then shifts to 443 nm at 800 K. The peak responsivity (@ 443 nm) increases from 0.016 to 0.026 A W −1 as the temperature is elevated from 77 to 400 K and then decreases to 0.017 A W −1 by raising the temperature to 800 K. The peak position of RD moves from 598 to 667 nm when the temperature increases from 77 to 700 K and then changes to 642 nm at 800 K. The peak responsivity (@ 620 nm) increases from 0.003 to 0.235 A W −1 by raising the temperature from 77 to 400 K and droops to 0.189 A W −1 at 800 K.
The insets of Fig. 4a show the temperature-dependent peak responsivity at a different bias (0, 2, − 4, and − 6 V) for BL and RD. The peak responsivity of BL, BD, and GD show a slight increase with bias conditions, while RD indicates almost insensitivity to the bias conditions. The EQEs for all samples are shown in dotted lines in Fig. 5a. The highest and lowest EQE are observed from RD and GD, respectively. Figure 4b shows the figure of merit, D*, of all LEDs, plotted at 77 K, 300 K, 500 K, 800 K using the 1 Hz equivalent noise bandwidth at the zero-bias condition. Overall, BD and GD show a lower signal to noise ratio compared with samples BL and RD. Furthermore, the peak D* of samples BD and GD are very close across all temperatures. The peak D* for BL decreases from 1.37 × 10 10 to 3.94 × 10 9 cm Hz 1/2 W −1 by elevating the temperature from 77 to 800 K. However, RD shows an increase of the peak D* from 2.04 × 10 10 to 5.21 × 10 10 cm Hz 1/2 W −1 by raising the temperature from 77 to 400 K. Then, increasing the temperature above 400 K leads to a decrease in peak D*. The peak D* is 7.66 × 10 9 cm Hz 1/2 W −1 at 800 K for RD. The insets of Fig. 4b show the D* of samples BL and RD with different biases. Both samples show high peak D* at zero bias conditions. Moreover, the same  www.nature.com/scientificreports/ behaviour is observed for all LEDs. Table 1 lists the maximum values of responsivity and detectivity for various LEDs at 300, 500, and 800 K. The normalized emissions and spectral responses of the LEDs at 77, 300, 500, and 700 K are plotted together in Fig. 5 to determine the best match of the LEDs and PDs for the optocouplers. It can be observed that the spectral response of the LED shifts towards shorter wavelengths (higher energy photons) as compared with its emission spectra as the LED cannot detect photons of lower energy than its bandgap. Table 2 shows the peak emission and spectral response wavelengths of the LEDs at 77, 300, 500, and 77 K. Overall, the perfect overlap is by employing samples RD-RD (LED-PD).
The second option is by integrating samples BL-GD. Moreover, there is a possibility for coupling is by using BD-GD. Fig. 5a, b show the potential for coupling at 77 and 300 K when GD-RD are integrated. Fig. 5c shows the best candidates for the LEDs and PDs to fabricate HT optocouplers (i.e., operating at 500 K). Typically, the selection is based on the following parameters: (1) the wavelength; (2) maximize the signal to noise ratio (D*); (3) the forward current of the LEDs low as possible; (4) long-term reliability of the LEDs (degradation of LEDs due to operating time is acceptable).

Discussions
Based on the results from the bandwidth and responsivity of LEDs, it is observed that not all LEDs can serve as PDs. Commonly, the most extended emitted wavelength corresponds to approximately the highest responsivity of the LEDs. This behaviour agrees well with our results with one exception, which is BL. A previous study on AlGaN MQW PDs shows that the responsivity of PDs increases by increasing the number of wells and thickness of wells along with decreasing the width of the barriers 25 . First, as the number of wells increases, the effective absorption and efficiency increase, consequently increasing the responsivity. Second, when the thickness of wells increases, the transition energy between the electron and hole decreases resulting in increasing the responsivity. Third, the number of electron-hole pairs that contribute to photocurrent increases by reducing the thickness of the barriers. As the width of the barriers reduces, tunnelling through the barriers increases, leading to increase responsivity. The structures of the LEDs are shown in Fig. 6a. The main difference between samples BL, BD, Table 2. Shows the peak emissions and spectral response at 77, 300, 500, and 800 K for the devices.

Peak emission wavelength (nm)
Peak spectral response wavelength (nm) 77 K 300 K 500 K 700 K 77 K 300 K 500 K 700 K   BL  446  448  457  464  402  405  418  420   BD  470  469  474  480  402  418  418  420   GD  530  529  539  544  420  437  443  450   RD  606  630  658  684  605  620  www.nature.com/scientificreports/ and GD is the pre-quantum layer. The pre-quantum well layer adds more thickness to the active region. Sample A and C have the same number of wells, but they are different in the widths of the well and barriers as well as the pre-quantum well layer. The total thicknesses of the barriers are 44 and 30 nm for samples BL and GD, so the difference is 14 nm. On the other hand, the overall widths of the wells are 167 and 140 nm for samples BL and GD, so the difference is 27 nm. Therefore, it is expected that the thickness wells have more impact than the thickness of barriers. As a result, sample BL higher responsivity than sample GD due to the higher thickness in their wells. Now, the differences between samples BD and GD are QWs number and width of wells. Samples BD and GD have 9 and 10 QWs, respectively. Also, the thickness well is 12.5 and 14 nm for samples BD and GD, respectively. Consequently, sample GD has higher responsivity than sample BD. In short, BL has the highest responsivity among the InGaN/GaN MQW LEDs. The D* depends on the responsivity and the area of the device. Since BL and RD have the highest responsivity and largest devices, they have higher D* than BD and GD. Furthermore, samples BD and GD show similar D* as they have the same device size.
Although the general responsivity of PDs increases as the reverse voltage increase, some PDs show insensitivity to the reverse voltage depending on the diode type. On the other hand, D* decreases for all PDs by applying the reverse bias. A high detectivity at zero biased condition suggests a high rate of change of noise than that of responsivity when the applied bias increases. This behaviour indicates that the detectivity of the MQW structure is limited to the bias-induced internal noise.

Conclusion
The work demonstrates the possibility of fabricating HT optocouplers with targeting operating temperature of 500 K. Four LEDs are studied based on their emissions and spectral responses. BL, BD, GD, and RD are used in this article. The LEDs show EL spectra up to 800 K with relatively high EQE. BD and GD show higher EQE and lower efficiency droop with temperatures as compared with BL and RD. Also, this work proves that LEDs can function as PDs. In contrast with the LEDs, BL and RD LEDs perform superior PDs than others. The measured responsivity varies from 0.031 to 0.25 A W −1 at 500 K, depending on the type of LED. The figure of merit, D*, changes from 1.87 × 10 9 to 4.07 × 10 10 cm Hz 1/2 W −1 at 500 K depending on the device. The EQE in temperature ranges from 77 to 800 K changes from 5 to 47 %, depending on the PD type. Finally, the wavelength (matching) overlap between the LEDs and PDs for different temperatures is demonstrated. First, the first possible couplings (LED-PD) are (RD-RD) and (GD-RD) in temperature ranging 77-300 K. Above 300 K, the potential couplings are (RD-RD), (BL-GD), (BD-GD). Therefore, (RD-RD) confirms that it can be employed by crossing all temperatures due to the high overlap between its emission and spectral response. In short, the present results indicate that LEDs can function as LEDs and PDs and use them to fabricate HT optocouplers, which can be substituted with the bulky transforms in the high-density power modules.

Methods
The InGaN-based and AlGaInP-based multiple quantum wells (MQWs) LED structures are shown in Fig. 6a, b, respectively. Figure 6a shows the InGaN/GaN MQW LED structures for BL, BD, and GD LEDs. The room temperature peak wavelengths for BL, BD, and GD samples are 454, 472, and 523 nm, respectively. Figure 6b illustrates the Al 0.05 Ga 0.45 In 0.5 P/Al 0.4 Ga 0.1 In 0.5 P MQW LED structure for red for display (RD) with a room temperature peak wavelength of 630 nm. The detailed device structure has been discussed earlier [20][21][22][23]26 . The chip sizes are 800 × 1345 µm 2 for BL LED, 191 × 270 µm 2 for BD and GD LEDs, and 300 × 300 µm 2 for RD LED. Figure 7a-d show LED images for blue for lighting and display, green for display, and red for display, respectively. These LEDs were supplied by HC SemiTek. The optical and electrical studies were conducted with a varied range of temperatures 77-800 K using a cryostat. The detail of the LEDs and photodiodes measurements and studied are explained previously [20][21][22][23]26 .