Sub-50 cm/s surface recombination velocity in InGaAsP/InP ridges

Development of high speed nanoscale photonic components, in particular emitters and detectors, for integrated optical interconnects will be critical for achieving an interconnect budget of <10 fJ/bit.³ III–V semiconductors are attractive due to their high speed and bandgaps in the 1.55 and 1.3 μm telecommunication windows; in recent years, there has been interest in developing III–V photonic crystal lasers,^4,5 metal-dielectric nanocavity lasers⁶ and LEDs,⁷ nanowire lasers,⁸ antenna-LEDs,^9,10 and nanoscale detectors.¹¹ In addition to optoelectronic devices,² III-V semiconductors have been explored for nanoscale FETs due to their high mobility.^12,13

However, when scaling devices down to the nanoscale, the surface area to volume ratio increases, making the device more sensitive to surface effects. If the surface of the semiconductor is not properly terminated, the dangling bonds and impurities form an electrically active non-radiative recombination pathway, leading to lower efficiency. Surface passivation looks to mitigate these problems, and while a variety of different fabrication processes have been tested,¹⁴ they can generally be grouped into three categories: regrowth,¹⁵ field effect passivation,^13,16–20 or chemical passivation of the dangling bonds through solution treatment.^1,21 However, chemical passivation has shown stability problems without proper encapsulation.^1,22,23 One of the most promising techniques reported to date is encapsulating the active region after sulfur passivation in 50 nm PECVD SiOx, achieving a low surface recombination of 260 cm/s.²⁴ 

In this work, we report an extremely low surface recombination velocity of 45  cm/s for InGaAsP LED ridges with an increase of >$180×$ in the photoluminescence in a nanoscale device with 200 nm width using an aged ammonium sulfide solution with a well controlled atomic layer deposited oxide (<17 nm). This is currently the lowest surface recombination velocity reported for an InGaAs or InGaAsP active region.

In this report, we will discuss two samples: the first was fabricated with a four-and-a-half year old bottle of ammonium sulfide 20% in water—for brevity, we will call this $(NH4)2S#1$⁠, and the second was fabricated using a new bottle of ammonium sulfide 20% in water where we added 0.22 g elemental sulfur to 7 ml of ammonium sulfide 20% in water 20 min prior to passivation—likewise for brevity, we will refer to this process as $(NH4)2S#2$⁠. As will be shown in the report, the sample treated with the $(NH4)2S#1$ solution has the lowest surface recombination velocity. When we tried to replicate this result with a new bottle of ammonium sulfide 20% in water, the results were significantly worse. However, we were able to demonstrate comparable results with the addition of elemental sulfur [$(NH4)2S#2$ solution], which suggests that the composition of ammonium sulfide is important to achieve the low surface recombination velocity. Ammonium sulfide solutions are very complex; as they age, they have been shown to have a decreased concentration of sulifdes and increased polysulfides and hyposulfite.²⁵ The addition of elemental sulfur can increase the free sulfur as well as the degree of polymerization of an ammonium sulfide solution.²⁵ A further chemical study would be required to determine the exact composition of the ammonium sulfide solutions.

The epitaxial layers were grown using metal organic chemical vapor deposition (MOCVD) on the InP substrate. They are patterned with hydrogen silsequioxane (HSQ) electron beam lithography resist and etched into ridges with 210 nm height and 1 μm length with varying widths using an inductively coupled plasma reactive ion etcher with an Ar/H2/CH4/Cl2 chemistry. The HSQ mask is stripped with 5:1 buffered hydroflouric acid (BHF) for 2 min.

Both samples are pre-cleaned in the OPD4262 developer (dilute tetramethylammonium hydroxide) for 1 min followed by a 15 s dip in 10:1 BHF. Between each step, the sample is rinsed in DI water and dried with nitrogen. After pre-clean, the sample is then submerged for 20 min in the ammonium sulfide solution [either $(NH4)2S#1$ or $(NH4)2S#2$], then it is rinsed with isopropyl alcohol for 5–10 s and immediately loaded into an atomic layer deposition (ALD) tool.

The sample treated with $(NH4)2S#1$ was loaded into a Picosun ALD tool. The deposited stack was 30 cycles Al₂O₃ (⁠$≈$3 nm), 200 cycles TiO₂ (⁠$≈$4 nm), and 100 cycles Al₂O₃ (⁠$≈$10 nm) deposited sequentially without breaking vacuum at 250 °C. The Al₂O₃ precursors are trimethylaluminum (TMA) and DI water, and the TiO₂ precurors are titanium tetrakis isopropoxide (TTIP) heated to 80 °C and DI water.

The sample treated with $(NH4)2S#2$ was loaded into a Cambridge Fiji F200 ALD tool. The atomic layer deposition stack consisted of 30 cycles Al₂O₃ (⁠$≈$3 nm), 100 cycles TiO₂ (⁠$≈$4 nm), and 30 cycles Al₂O₃ (⁠$≈$3 nm) deposited sequentially without breaking vacuum at 200 °C. The Al₂O₃ precursors are trimethylaluminum (TMA) and DI water, and the TiO₂ precurors are tetrakis(dimethylamino)titanium (TDMAT) heated to 75 °C and DI water. Note that the two samples presented in this work were coated with two different ALD films. We performed an independent test using elemental sulfur with a six-month-old ammonium sulfide bottle [$(NH4)2S#3$] in the Picosun ALD tool with the same dielectric stack outline above (30 cycles Al₂O₃, 200 cycles TiO₂, and 100 cycles Al₂O₃). The measured surface recombination velocity was comparable to the Cambridge ALD tool.

Finally, to complete the surface passivation, both samples underwent rapid thermal annealing at 350 °C in N₂ for 5 min. A schematic of the device and a scanning electron micrograph (SEM) of the sample treated with $(NH4)2S#2$ are shown in Fig. 1.

To characterize the efficiency of surface passivation, we collected room temperature detected photoluminescence vs laser pump power (L-L) measurements and time correlated single photon counting (TCSPC) measurements of our LED ridges immediately after etching and after surface passivation. We used a 1170 nm continuous wave (CW) laser for the L-L measurements to avoid a time-dependent quantum efficiency and a femtosecond Ti:sapphire laser at 1000 nm for the TCSPC measurements. Both setups use a linear polarizer for the input to align the electric field to the length of the ridge, a 50× microscope objective to focus the laser, and a dichroic mirror paired with a 1300 nm longpass filter to remove laser light and etch stop emission.

The L-L curves from CW excitation are plotted in Fig. 2 for LED ridge widths of 40, 200, 400, and 700 nm. We saw an increase in the photoluminesence (PL) after passivation compared to the measurement immediately after etching the sample for all ridge widths.

The rate equation for CW excitation is provided below:

where G is the optical generation rate, A′ is the combination of Shockley−Reed−Hall (SRH) recombination and surface recombination, B is the radiative recombination coefficient, and C is the Auger recombination coefficient. Because we are measuring photons (⁠$∝BN2$⁠), if the carrier recombination is dominated by SRH and surface recombination (A′) we expect a slope of 2 dec/dec in the loglog L-L plot. Likewise, if the recombination is dominated by radiative recombination, we expect a slope of 1 dec/dec. As shown in Fig. 3(a) for a device with a width of 200 nm, after etching the L-L slope was 1.62 dec/dec and, after passivation, the slope decreased to 1.02 dec/dec, which indicates that for these powers, radiative recombination is dominating. Additionally, at the lowest pump power measured, we saw an increase in the PL of $180×$ for the 200 nm width. Extrapolating from these slopes we would expect the ratio to be even larger at lower pump powers.

Note that the <1.00 dec/dec slope roll-off at pump powers over $≈40μW$ in the after-passivation measurement can be explained by band-filling rather than the onset of Auger recombination. In Fig. 3(b), the portion of the spectra below the 1300 nm longpass filter is fairly small at 4 μW, but as we increased the pump power by an order of magnitude, the peak power shifted to shorter wavelengths and the portion of the spectra below 1300 nm was no longer insignificant.

To measure the time correlated single photon counting decay, we used an MPD InGaAs/InP single photon avalanche photodiode with a 4ps timing resolution. Due to the relatively long lifetimes, the device does not reach A′ dominated recombination until low carrier concentrations. This corresponds to a low photon count, making it difficult to measure a purely exponential decay without falling below the detector noise floor. So, instead of a pure exponential, we fit the temporal decay curve with both A′ and B terms: $G=A′N+BN2$⁠, referred to here as the A′B decay curve. The initial part of the A′B decay curve is dominated by radiative recombination and decays to the A′ dominated rate. This decay curve, where A′ and B both contribute, has been previously modeled,²⁶ and to confirm the quality of the fit, we introduce a model that can account for Auger recombination by directly fitting to the normalized rate equation. The derivation and details of those models can be found in the supplementary material.

An example of the decay curves for after etching compared to after surface passivation is shown in Fig. 4(a). The A′ lifetimes for the 200 nm ridge width after etch, $(NH4)2S#2$ passivation, and $(NH4)2S#1$ passivation are 0.61, 29.95, and 207.47 ns, respectively.

The connection between the surface recombination velocity (SRV) and the A′ coefficient is given as

where $τSRH,bulk$ is the bulk SRH recombination, τ_SR is the lifetime from surface recombination, v_s is the surface recombination velocity, w is the width of the ridge, and l is the length of the ridge. For nanoscale devices generally $τSRH,bulk≫τSR$⁠, so we can neglect the bulk contribution—this lets us directly convert between the measured lifetime and the surface recombination velocity, shown in Fig. 4(b). There are two regions in the graph: at narrow ridge widths, the surface recombination increases exponentially (plotted as a dashed line), then at wider ridges, the surface recombination velocity is approximately constant (plotted as a solid black line, the gray region represents ± a standard deviation).

The average fit in the constant region of the surface recombination velocity gives us $1.3×104 ±1075$ cm/s after etching (blue dots), and after passivation this decreases to 45 ± 15 cm/s for the $(NH4)2S#1$ passivation (green dots). Likewise for the $(NH4)2S#2$ passivation (orange dots), we get an average surface recombination velocity of 190 ± 42 cm/s. Our independent test treated in the $(NH4)2S#3$ solution yielded a comparable surface recombination velocity of approximately 560 cm/s.

While potential mechanisms for surface passivation were not fully explored in this paper, we believe the results were achieved by a combination of chemical passivation through sulfur-saturated ammonium sulfide and field effect passivation through a high density of negative fixed charges in the ALD trilayer—specifically in the inner Al₂O₃/TiO₂ layers.^17,20,27,28 Field effect passivation can possibly explain the deviation from the constant surface recombination velocity at narrow ridge widths (⁠$<160 nm$⁠), because narrower ridge widths lead to higher majority carrier concentration at no bias,^27,28 effectively leading to electrostatic doping. Both surface recombination and the radiative decay rate from electrostatic doping are proportional to N, so, at the narrowest ridge widths, the decay curve does not give an unambiguous fit of surface recombination. To test this idea, we measured low power L-L curves for 40, 100, and 200 nm $(NH4)2S#2$ passivated devices and converted these to η_IQE in Fig. 5. We assumed a surface recombination velocity of 190 cm/s, $B=10−10 cm3 s-1$⁠, and $Cn=Cp=8×10−29 cm6 s-1$⁠. The Auger coefficients were taken from fitting the literature data for C_n and C_p.²⁹ From the modeled η_IQE, Fig. 5(a), we see that if the surface recombination velocity was increasing without electrostatic doping (higher loss curves), we would expect an exponential decrease in the slope for lower generation rates, but for the electrostatic doping model (electrostatic curves), this would eventually become constant when the terms proportional to N dominate. We note that the experimental η_IQE, Figs. 5(b) and 5(c), more closely resembles the behavior from the electrostatic doping model rather than having a larger non-radiative $A′$ term. For comparison, we added the after-etch η_IQE, which, as expected, shows an exponential drop in efficiency for lower generation rates. Note, we did not plot the expected after etch surface η_IQE in Fig. 5(a). See the supplementary material for further modeling details.

We also tested several small process variations. Without adding elemental sulfur to $(NH4)2S$⁠, the surface recombination velocity is lower than after etching but is much higher than the $(NH4)2S#2$ process. A similar improvement was achieved by baking a $(NH4)2S$ solution on a hotplate at 30 °C for 30 min prior to soaking the sample. We have not performed a full design of experiment on the process space, so it is possible that there exists a more optimal $(NH4)2S$ passivation process [$(NH4)2S$ composition and soak time], ALD conditions (temperature, thickness, and precursor), and annealing conditions (gas, time, and temperature) that would allow for even lower surface recombination velocities.

This surface passivation is particularly promising for applications that require thin oxides like FETs or antenna-LEDs. We found that removing either the TiO₂ deposition or outer Al₂O₃ deposition during the process flow results in significantly worse surface recombination velocity. However, after N₂ annealing, the outer Al₂O₃ can be selectively etched in 30:1 buffered hydroflouric acid (BHF) with little to no degradation in the surface recombination velocity—providing a further pathway for reducing the total oxide thickness.

In summary, we found that the combination of aged ammonium sulfide and a trilayer Al₂O₃/TiO₂/Al₂O₃ combined with a rapid thermal anneal led to a record low surface recombination velocity of 45 cm/s. Importantly, because the surface is encapsulated, it shows no degradation of surface properties over several months and tests with the CW and femtosecond lasers. This surface passivation, using thin ALD oxides, can potentially enable highly efficient nanoscale devices.

See the supplementary material for field effect passivation, internal quantum efficiency measurement details, APD background subtraction, and auto-alignment.

The authors thank Professor Ali Javey for his valuable suggestions on how to replicate ammonium sulfide aging.

This work was funded by the Center for Energy Efficient Electronics Science through a grant from the National Science Foundation (NSF Award No. ECCS-0939514) and Department of Energy Office of Basic Energy Sciences under Contract No. DE-AC02-05CH11231. This material is based upon the work supported by the National Science Foundation Graduate Research Fellowship under Grant No. DGE 1106400.