Reticulated shallow etch mesa isolation (RSEMI) for controlling surface leakage in GaSb-based infrared detectors

Transcription

1 Reticulated shallow etch mesa isolation (RSEMI) for controlling surface leakage in GaSb-based infrared detectors J. A. Nolde, 1,a) E. M. Jackson, 1 M. F. Bennett, 2 C. A. Affouda, 1 E. R. Cleveland, 1 C. L. Canedy, 1 I. Vurgaftman, 1 G. G. Jernigan, 1 J. R. Meyer, 1 E. H. Aifer 1 1 U.S. Naval Research Laboratory, 4555 Overlook Ave. SW, Washington, DC 20375, USA 2 Sotera Defense Solutions, Inc., 7230 Lee Deforest Dr. Suite 100, Columbia, MD 21046, USA Abstract: Longwave infrared detectors using p-type absorbers composed of InAs-rich type-ii superlattices (T2SLs) nearly always suffer from high surface currents due to carrier inversion on the etched sidewalls. Here we demonstrate reticulated shallow etch mesa isolation (RSEMI): a structural method of reducing surface currents in longwave single-band and midwave/longwave dual-band detectors with p-type T2SL absorbers. By introducing a lateral shoulder to increase the separation between the n + cathode and the inverted absorber surface, a substantial barrier to surface electron flow is formed. We demonstrate experimentally that the RSEMI process results in lower surface current, lower net dark current, much weaker dependence of the current on bias, and higher uniformity compared to mesas processed with a single deep etch. For the structure used, a shoulder width of 2 µm is sufficient to block surface currents. Infrared (IR) detectors incorporating p-type, type-ii superlattice (T2SL) absorbers have shown promising levels of performance because the diffusion lengths for minority carrier electrons are sufficiently long to achieve high quantum efficiency (QE) in the longwave (LW) IR. However, like bulk InAs, 1 InAs-rich T2SL absorber materials such as InAs-Ga(In)Sb and InAs- InAsSb superlattices, as well as the bulk InAsSb alloy, have high densities of surface states a) Electronic mail: jill.nolde@nrl.navy.mil Page 1 of 14

2 located on exposed mesa sidewalls. In order to maintain charge neutrality at the surface, these donor-like states induce band bending, causing the Fermi level to pin near the energy of these states, just within the conduction band. 2 For p-type material, this pinning leads to an electron inversion layer and excessive surface leakage associated with electron conduction along the sidewalls, which degrades detector performance. The problem is exacerbated in advanced FPAs, where the smaller pixels have larger surface-to-volume ratios, and are thus more readily dominated by surface current. Numerous research efforts have attempted to directly address the problem by developing methods for effectively passivating the surface states of T2SL absorber materials. Investigated approaches have included: encapsulation with polymers 3,4 or dielectrics; 5 7 epitaxial overgrowth of a larger bandgap layer; 8,9 and sulfidation. 9,10 While many of these studies have shown improvement compared to unpassivated devices, generally the observed dark currents were still much higher than theoretical limits. The use of a gate metal to modulate the trapped charge was more effective, 11 but is inconvenient for small-pixel FPAs. Although inconsistencies in the literature make it difficult to draw definitive conclusions concerning the effectiveness of the various passivation chemistries, sidewall leakage remains a serious unresolved issue for p-type III-V LWIR photodiodes. An alternative approach to suppressing surface leakage is by exploiting heterostructure design and mesa geometry, as we demonstrated previously using shallow etch mesa isolation (SEMI). 12 In this approach, individual pixels are isolated by etching just through the n-contact while the absorber remains buried. For the heterostructure with the band diagram depicted in Fig. 1(a), the pixel mesas would be isolated by etching through the n + longwave barrier (LWB) superlattice (SL) and stopping at the vertical dotted line within the low p-doped (π) LWB. This allows the longwave absorber (LWA) layer to remain buried under 200 nm of the LWB. Surface currents are strongly suppressed because the p-doped LWA is not exposed on the Page 2 of 14

3 etched surface and the LWB does not support a surface channel. The process can also be used on pbp unipolar barrier 13 or other heterojunction structures in which the metallurgical junction is located within a wide-gap layer that does not form a surface inversion layer. However, a disadvantage of the SEMI arrangement is that the unbroken narrow-gap absorber allows for crosstalk via the diffusion of minority carriers between neighboring pixels. For FPAs with a significant lateral diffusion length and small pixel pitch, crosstalk can strongly degrade the spatial resolution of the array. Furthermore, SEMI is inconvenient for use on midwave/longwave (MW/LW) dual-band (DB) structures. Here we introduce an extension of SEMI, the reticulated shallow etch mesa isolation (RSEMI) architecture 14 that fully reticulates each pixel to eliminate crosstalk and also minimizes surface currents for both single-band (SB) and dual-band detectors without the use of chemical passivation. RSEMI and SEMI have similar requirements for the detector structure, i.e., a pbp or n + -π-p - diode structure that employs wide-gap materials in all layers except the absorber. For definiteness, we will again use the n + -π-p - photodiode structure depicted in Fig. 1(a) for describing the RSEMI functionality, with the understanding that the concept applies more generally. In contrast to the conventional photodiode mesa illustrated in Fig. 1(b), RSEMI, as schematically depicted in Fig. 1(c), employs a lateral shoulder where the n + contact is recessed above the π region. This has the effect of increasing the path length for electrons flowing along the surface, on the path denoted by the arrow in Fig. 1(c), from the n + cathode to the n-type inversion layer at the absorber sidewall. Page 3 of 14

4 (a) 200 Energy (mev) (b) E f n + LWB p - LWA (InAs/GaInSb SL) InAs/AlInSb SL z ( m) (c) n + LWB p - LWB ( ) p - LWA n + LWB p - LWA p + contact p + contact FIG. 1. (a) Band diagram of the single band (SB) T2SL heterostructure diode that was studied experimentally. The horizontal dashed line is the Fermi level. The vertical dotted grey line represents a typical etch depth for the SEMI or RSEMI processes. (b) Schematic of a deep etched mesa for an SB detector. (c) Schematic of an RSEMI mesa. In (b) and (c), the shaded regions represent the surface inversion layers and the arrows indicate the paths followed by electrons flowing along the surfaces. The cross hatching under the n + region of (c) is the junction space charge region. The operating principle of RSEMI can be understood by considering 1D conduction along a straight sidewall with no shoulder. The band structure is simulated using NRL- Multibands. 15 The superlattice band parameters are calculated from 8-band k.p simulations, then Poisson s equation is solved on a 1D mesh. Full Fermi-Dirac statistics are used to compute the carrier concentrations and the resulting equilibrium band profiles. In the Page 4 of 14

5 conventional mesa structure of Fig. 1(b), the LW FPA pixel is defined by a single deep etch terminating just below the bottom of the p-type absorber. The shaded regions along the absorber sidewall indicate the electron inversion layer, while the arrow indicates the resulting pathway for surface electron flow. In order to avoid an excessive operating bias, the π layer should be approximately one depletion layer thick, or 0.3 µm for a typical p-doping level of cm 3. Figure 2 shows a 1D simulation of the band structure along the etched sidewall as a function of π layer (p = cm 3 ) thickness, assuming an effective 3D n-doping of the surface inversion layer of cm 3. While the actual electron density varies with distance from the surface, our assumption is sufficient to illustrate the concept. We also assume that the large bandgap of the π layer prevents its surface from pinning, a critical requirement that is confirmed by the experimental study discussed below. Note that when the n + inversion layer is separated from the n + cathode by several depletion widths (e.g. the blue dash-dot curve in Fig. 2, which corresponds to a π-layer thickness of 1.5 µm), the π layer fully depletes to create a significant barrier to electrons along the surface. However, this barrier becomes less effective with decreasing thickness. In particular, the π-layer thickness of about one depletion width that is required to avoid high operating bias (0.3 µm, solid black curve in the figure) induces a barrier height of < 50 mev. At the operating temperatures of interest such a barrier will do little to impede the flow of thermalized surface electrons, which is consistent with the nearly universal observation of excessive surface leakage in LW n + -π-p - structures that are deep etched with straight mesa sidewalls. Whereas the specific barrier height induced by a given π-layer width depends on the doping level, modifying the doping level only alters the length scale and not the fundamental incompatibility of the requirements for a thin (about one depletion width) π-layer to avoid high bias and a thick (several depletion widths) π-layer to minimize surface leakage. Page 5 of 14

6 Energy (mev) thickness: 0.3 m 0.7 m 1.0 m 1.5 m z ( m) FIG. 2. Band diagram along the mesa sidewall for an LW structure in which the absorber surface is pinned n type and the π layer thickness is varied. The dotted horizontal line shows the Fermi level. The RSEMI approach circumvents this limitation by creating different path lengths for electrons traversing the π layer, depending on whether they flow within the bulk or along the surface. In particular, the in-plane shoulder illustrated in Fig. 1(c) stretches the surface path, allowing full depletion of the π region between the n + cathode and absorber inversion layer and imposing a high barrier that blocks surface currents. While the 1D analysis illustrated in Fig. 2 oversimplifies the complex electric field contours of RSEMI s mesa geometry, it nonetheless provides a useful tool for estimating the shoulder width required to substantially suppress surface currents. We conclude that if the total path through the π region (vertical plus lateral) is at least a few microns long, the imposed barrier should effectively block surface leakage. Below we describe experimental demonstrations showing that RSEMI strongly suppresses surface dark currents on both SB and DB T2SL photodiode structures. Layer structures for the SB and DB diodes that were designed at NRL and grown by molecular beam Page 6 of 14

7 epitaxy at IQE, plc. are specified in Tables I(a) and I(b), respectively. The band diagram shown above in Fig. 1 is that for the LW SB diode. The midwave/longwave DB detector was designed in a back-to-back absorber configuration in which the MW SB diode was grown inverted, starting with the n + cathode and completing with its absorber, which was followed by the LW absorber. To prevent crosstalk between the absorbers, they were separated by a thin InAs/GaInSb SL that functioned as an electron barrier. Table I. Layer structures of (a) SB and (b) DB wafers used in this study. (a) (b) Layer Structure Thickness (nm) Doping (cm 3 ) Layer Structure Thickness (nm) Doping (cm 3 ) Cap InAs 10 n = Cap InAs 10 n = n+ LWB InAs / Al0.83In0.17Sb 200 n = n+ LWB InAs / Al0.83In0.17Sb 200 n = Π InAs / Al0.83In0.17Sb 300 p < π LWB InAs / Al0.83In0.17Sb 300 p < LWA InAs / Ga0.88In0.12Sb 6000 p = LWA InAs / Ga0.88In0.12Sb 1500 p = Grade InAs / Ga0.88In0.12Sb 100 e barrier InAs / Ga0.88In0.12Sb 50 p = Contact GaSb 500 p = MWA InAs / Ga0.88In0.12Sb 1500 p = π MWB InAs / Al0.88In0.12Sb 300 p < Contact InAs / Al0.88In0.12Sb 1000 n = RSEMI devices were fabricated using a citric-acid based wet-etch process, although other wet or dry etchants could also be used. The first etch forms the RSEMI shoulder and is targeted to stop 500 Å into the π layer as indicated by the dashed vertical line in Fig. 1(a). This is followed by deep etching into the backside contact layer to create the isolation mesa, with a diameter several microns wider than for the shallow etch. The deep etch fully isolates the mesas and provides electrical access to the anode. Un-annealed Ti/Au Ohmic contacts were then made to the exposed n and p-contact regions using electron beam evaporation and lift-off processing. No other passivation or encapsulation was performed. Page 7 of 14

8 Table II summarizes the cutoff wavelength (λ c, defined as the inflection point of the quantum efficiency curve when plotted versus wavenumber), absorber thickness, and median external quantum efficiency (EQE) at 80% of the cutoff wavelength for detectors fabricated from the SB and DB wafers. We found no substantial difference in the optical performance of mesas fabricated with the conventional single deep etch (labelled deep), SEMI, or RSEMI processes. In the remainder of this article, we will focus on the dark current performance and effects of RSEMI on the surface leakage. Table II. Characteristics of wafers used for RSEMI characterization. ID Cutoff wavelength, λ c (μm) Absorber thickness (μm) Median EQE at 80% λ c (%) SB DB The dark current-voltage (DIV) characteristics of up to 60 devices per test die were measured at 78 K using a Keithley 7002 switching mainframe equipped with Keithley 7158 low current switching cards. The switching matrix has a noise floor of about 1 pa, which corresponds to a current density between and A/cm 2 for the size range of the processed diodes. The operating bias (V b ) for each diode is defined as the bias required to obtain 95% of the maximum photocurrent. The current density (J b ) at V b was tabulated for diodes of each type (i.e. Deep, SEMI, RSEMI), and the median and 10 th -percentile (lowest) J b were identified. Figure 3 plots representative DIV curves at 78 K for SB detectors processed with the three different geometrical etch profiles: deep (dotted black), SEMI (dashed blue), and RSEMI (solid red). The deep and RSEMI devices resided on the same die, while the SEMI devices were processed previously from the same wafer. The vertical green line represents the Page 8 of 14

9 operating bias V b = 20 mv. While the median J b of the RSEMI devices (37 μa/cm 2 ) was only slightly lower than that of the deep and SEMI devices, the RSEMI DIV curves remained nearly flat up to 300 mv. At 200 mv, the median current density is only 28% higher than the value at V b, but is more than 300 times lower than for the strongly field-dependent deep devices. The RSEMI DIV curves are also quite uniform, with J b for 90% of the devices falling within a factor of 2 of the 10th percentile J b. Clearly, RSEMI devices show lower dark current, much less field dependence, and higher uniformity than the mesas patterned with a single deep etch. Current density(a/cm 2 ) 10 1 V b = -20mV Deep 10-5 RSEMI SEMI Voltage (V) FIG. 3. DIV at 78 K for the SB diode processed as deep (dotted black), SEMI (dashed blue), and RSEMI (solid red) mesas. Although the SEMI process cannot be used for DB structures with back-to-back absorbers since the bottom junction is not isolated between devices, RSEMI can be used to isolate the mesas electrically and optically while strongly suppressing the sidewall leakage across the upper LW junction. Figure 4(a) shows the DIV plot for the DB structure measured at 78 K, where the number of curves is reduced for clarity. The vertical blue lines denote V b for the MW ( 20 mv) and LW (+130 mv) junctions. While J b for the MW junction (negative voltages) cannot be accurately determined since at biases < 200 mv the dark currents fall below the Page 9 of 14

10 noise level of the measurement system, the results are quite similar for the two processing methods because the RSEMI shoulder only affects the surface current of the top (LW) junction. In comparison, RSEMI reduces J b by nearly a factor of five for the LW junction. Furthermore, the RSEMI current density again remains flat for several hundred millivolts, increasing by only 10% at 300 mv. (a) (b) Current density(a/cm 2 ) Deep RSEMI V b (MW) = -20mV Voltage (V) V b (LW) = 130mV Current Density ( A/cm 2 ) Deep 130mV RSEMI 130mV RSEMI 300mV P/A(cm -1 ) FIG. 4. (a) DIV at 78 K for deep (dashed black) and RSEMI (solid red) DB devices. The vertical lines show V b for the MW and LW junctions. (b) Current density at the LW V b for deep (black squares) and RSEMI mesas (red circles) and at 300 mv (RSEMI only, blue triangles and dotted line) plotted versus the perimeter to area ratio of each mesa. In order to quantify the surface leakage, Figure 4(b) plots J b versus the mesa s perimeter-to-area ratio (P/A), along with a linear least squares fit that is used to extract the surface component. The slope corresponds to the surface current, J surf, while the intercept determines the bulk current density J bulk. We can also calculate the sidewall resistance, where is Boltzmann s constant, is temperature, and is the electron charge. For the LW junction, J surf is 0.8 na/cm 2, which corresponds to a sidewall resistivity of 9 MΩ cm at V b. These values do not change significantly at higher bias, e.g., J surf increases to Page 10 of 14

11 1 na/cm 2 and 6.7 MΩ cm at 300 mv. In contrast, the surface current and resistivity for deep mesas are 85 times worse at V b. We can also estimate the fraction of the total current carried on the surface,. For diodes with a diameter of 200 µm, F surf is 86% and 7% for deep and RSEMI devices, respectively. When the RSEMI results are extrapolated down to 10 µm pixels, F surf is 44-68% accounting for the large standard error (±22%) of the small value for J surf. The results presented in Figs. 3 and 4 were obtained for mesas with an RSEMI shoulder width of 20 µm. While these structures were useful to validate the new approach, such a wide shoulder would not be practical for FPAs. Therefore, we designed a mask to pattern RSEMI mesas with four different small shoulder widths on the same DB die with nominal values of 0 (corresponding to a deep etch), 2, 5, and 10 µm, although differences in etch undercut and mask alignment caused variations of a few microns from run to run. Current density(a/cm 2 ) Deep RSEMI 2 m 10-2 RSEMI 5 m RSEMI 10 m Voltage (V) FIG. 5. DIV of the best 5 10 diodes of four different shoulder widths that are nominally 0 (deep), 2, 5, and 10 μm. The vertical line shows V b for the LW junction. Page 11 of 14

12 Figure 5 plots DIV curves for the best 5 10 diodes of each shoulder width. As before, surface leakage substantially degrades the dark current when there is no RSEMI shoulder (black lines). On the other hand, we find that all three RSEMI shoulder widths yield nearly identical curves, with the same low dark current density and surface leakage extracted from P/A data as for devices having 20-µm-wide shoulders. This implies that the minimum shoulder width required to eliminate surface leakage on these structures is less than 2 µm, which is consistent with the value µm derived from the simple 1D model discussed above. For the 5 µm RSEMI shoulder, 13 of the 15 diodes (87%) had J b within 2x of the 10 th percentile value, whereas only 50% met that specification when the nominal shoulder width was 2 µm. The less uniform results for the narrowest shoulder reflect variation of the mesa alignment and wet etch undercut across the die. A process that better controls the undercut, such as using reactive ion etching (RIE) for the second isolation etch, may significantly reduce the scatter. Using a step and repeat exposure tool with an RIE etch, 10-micron pixels with a 2 μm shoulder should be possible. We estimate the dark current for pixels with these dimensions to be less than 10 μa/cm 2. To summarize, we have demonstrated that RSEMI can dramatically reduce the dark currents of heterostructure IR detectors employing p-type InAs-rich T2SL absorbers, which are prone to strong surface leakage due to the n-type inversion layer that forms along the mesa sidewalls. RSEMI is shown to be effective in substantially reducing the surface leakage for both SB and DB diodes, and by extension, unipolar barrier device structures such as the pbp. The RSEMI process results in lower surface current, lower total dark current, much weaker dependence of the current on bias, and higher uniformity compared to deep-etched mesas. We find that for π layer doping in the low cm 3 range, a shoulder width of 2 µm is sufficient. However, at this scale the processing tolerances need to be addressed in further studies. Page 12 of 14

13 The authors are grateful for the support of our sponsors from the Office of Naval Research (ONR) and the U. S. Army Night Vision and Electronic Sensors Directorate (NVESD). We also thank IQE for growth of the wafers used in this study. 1 C. Affentauschegg and H.H. Wieder, Semicond. Sci. Technol. 16, 708 (2001). 2 W. Mönch, Semiconductor Surfaces and Interfaces, 3rd ed. (Springer-Verlag, New York, 2001). 3 E.K. Huang, D. Hoffman, B.-M. Nguyen, P.-Y. Delaunay, and M. Razeghi, Appl. Phys. Lett. 94, (2009). 4 H.S. Kim, E. Plis, N. Gautam, S. Myers, Y. Sharma, L.R. Dawson, and S. Krishna, Appl. Phys. Lett. 97, (2010). 5 S. Bogdanov, B.-M. Nguyen, A.M. Hoang, and M. Razeghi, Appl. Phys. Lett. 98, (2011). 6 R. Peng, S. Jiao, H. Li, S. Gao, Q. Yu, J. Wang, D. Wang, and L. Zhao, J. Electron. Mater. 45, 703 (2016). 7 O. Salihoglu, A. Muti, and A. Aydinli, IEEE J. Quantum Electron. 49, 661 (2013). 8 R. Rehm, M. Walther, F. Fuchs, J. Schmitz, and J. Fleissner, Appl. Phys. Lett. 86, (2005). 9 E. Plis, M.N. Kutty, S. Myers, S. Krishna, C. Chen, and J.D. Phillips, Proc. SPIE 9070, (2014). 10 N.C. Henry, A. Brown, D.B. Knorr, Jr., N. Baril, E. Nallon, J.L. Lenhart, M. Tidrow, and S. Bandara, Appl. Phys. Lett. 108, (2016). 11 G. Chen, A.M. Hoang, and M. Razeghi, Appl. Phys. Lett. 104, (2014). 12 E.H. Aifer, J.H. Warner, C.L. Canedy, I. Vurgaftman, E.M. Jackson, J.G. Tischler, J.R. Meyer, S.P. Powell, K. Olver, and W.E. Tennant, J. Electron. Mater. 39, 1070 (2010). 13 J.A. Nolde, E.M. Jackson, C.A. Affouda, R.S. Pai, S.I. Maximenko, M.K. Yakes, C.L. Canedy, I. Page 13 of 14

14 Vurgaftman, J.R. Meyer, and E.H. Aifer, in Mil. Sens. Symp. Parallel (2012). 14 E.H. Aifer, J.A. Nolde, I. Vurgaftman, E.M. Jackson, and J.R. Meyer, U.S. patent application 14/ (18 November 2015). 15 M.P. Lumb, I. Vurgaftman, C.A. Affouda, J.R. Meyer, E.H. Aifer, and R.J. Walters, Proc. SPIE 8471, 84710A (2012). Page 14 of 14

15 (a) Energy (mev) (b) E f n + LWB p - LWA (InAs/GaInSb SL) InAs/AlInSb SL p - n + LWB LWB ( ) z ( m) (c) n + LWB p - LWA p - LWA p + contact

16 Energy (mev) thickness: 0.3 m 0.7 m 1.0 m 1.5 m z ( m)

17 Current density(a/cm V 10 0 b = -20mV Deep RSEMI SEMI

18 Current density(a/cm 2 ) (a) Deep RSEMI V b (MW) = -20mV Voltage (V) V b (LW) = 130mV Current Density ( A/cm 2 ) Deep 130mV RSEMI 130mV RSEMI 300mV P/A(cm -1 )

19 Current density(a/cm Deep RSEMI 2 m RSEMI 5 m RSEMI 10 m