4.3 Μm Quantum Cascade Detector in Pixel Configuration References and Links

We present the design simulation and characterization of a quantum cascade detector operating at 4.3 µm wavelength. Array integration and packaging processes were investigated. The device operates in the 4.3 µm CO 2 absorption region and consists of 64 pixels. The detector is designed fully compatible to standard processing and material growth methods for scalability to large pixel counts. The detector design is optimized for a high device resistance at elevated temperatures. A QCD simulation model was enhanced for resistance and responsivity optimization. The substrate illuminated pixels utilize a two dimensional Au diffraction grating to couple the light to the active region. A single pixel responsivity of 16 mA/W at room temperature with a specific detectivity D * of 5 · 10 7 cm √ Hz/W was measured. Fabrication of high-performance large-format MWIR focal plane arrays from MBE-grown HgCdTe on 4 silicon substrates," J. A 256×256 element InSb focal plane array for ground-based astronomy, " First demonstration of megapixel dual-band QWIP focal plane array, Demonstration of megapixel dual-band QWIP focal plane array, " IEEE J. High operating temperature interband cascade midwave infrared detector based on type-II InAs/GaSb strained layer superlattice, " Appl. Mid-IR focal plane array based on type-II InAs/GaSb strain layer superlattice detector with nBn design, " Appl. Performance improvement of longwave infrared photodetector based on type-II InAs/GaSb superlattices using unipolar current blocking layers, " Appl. Phys. High operating temperature midwave infrared photodiodes and focal plane arrays based on type-II InAs/GaSb superlattices, " Appl. Photovoltaic quantum well infrared photodetectors: The four-zone scheme, " Appl. (4 µm) quantum cascade detector based on strain compensated InGaAs / InAlAs, " Appl. InP-based quantum cascade detectors in the mid-infrared, " Appl. " 19 µm quantum cascade infrared photodetectors, " Appl. High performance, room temperature, broadband II-VI quantum cascade detector, " Appl. Midinfrared quantum cascade detector with a spectrally broad response, " Appl. " CO2 isotope sensor using a broadband infrared source, a spectrally narrow 4.4 µm quantum cascade detector, and a Fourier spectrometer, " Appl. Diagonal-transition quantum cascade detector, " Appl. Mid-infrared quantum cascade detectors for applications in spectroscopy an pyrometry, " Appl. Quantum cascade detector utilizing the diagonal-transition scheme for high quality cavities, " Opt. Vsp—a quantum-electronic simulation framework, " J. " High performance bi-functional quantum cascade laser and detector, " Appl.


Introduction
Mid-infrared detector arrays have been investigated intensively in the past years.Their applicability in a large variety of fields including thermal imaging, remote detection, astronomy and military countermeasure systems promoted the research on so called focal plane arrays (FPA).Microbolometer, HgCdTe [1,2], InSb [3] and quantum well infrared photodetector (QWIP) FPAs have already proven their performance and are well established technologies.Despite of their performance, HgCdTe detectors are highly sensitive to their material composition and material distribution.Moreover, batch production relies on the availability of growth and bandgap compatible substrates with respect to absorption in bottom side illuminated designs.QWIP FPAs have been scaled up to provide megapixel and dualcolor imaging [4,5].Photoconductive QWIP FPAs require operation temperatures which are not accessible by thermo-electric coolers (TEC) and suffer therefore from relatively heavy space consuming casings.Type II superlattice interband detectors proposed already in 1987 [6] offer focal plane array compatibility [8], detection in the mid-infrared and long wave infrared [9,10].Operation at elevated temperatures up to 420 K was demonstrated [7].Thermal imaging up to 180 K with a specific detectivity D * of 1.05 • 10 12 cm √ Hz/W at 150 K was shown utilizing a diffusion barrier [11].Quantum cascade detectors (QCDs) offer the advantages of room temperature operation in combination with the design freedom of quantum cascade based intersubband devices and compatibility to standard InP based growth and fabrication technology.Evolved from photovoltaic QWIPs [12] they utilize a LO-phonon extraction scheme which provides more design freedom.A challenging aspect of QCD design is to maintain a high device resistance while the responsivity remains reasonably high.An advantage of QCDs over well established detector types is their designable operation wavelength, low noise photovoltaic operation mode and therefore room temperature operation capability.With a well established material system detectors for different operation wavelengths can be designed by adjusting the well and barrier dimensions, while the material properties remain the same.For mobile, battery powered applications an uncooled or thermo-electrically cooled mid-infrared imaging FPA is needed.Beside the single pixel performance, scalability to FPA dimensions of several tens of millimeters is necessary for high pixel counts.MBE and MOCVD growth on InP provides these requirements.QCDs have already proven their room temperature performance at a wide wavelength range [13][14][15] and broadband designs [16][17][18].As shown by Hofstetter et al.QCDs around 4.3 µm wavelength provide CO 2 sensing abilities [19] with high selectivity.Recently QCD performance and robustness at longer wavelengths was increased by a diagonal transition design [20].The InAs/AlAsSb material system had been introduced for short wavelength high resistance QCDs [21].Dependent on the desired application the operation wavelength and several QCD parameters have to be chosen.QCD design involves a trade-off between responsivity and device resistance which then influences all other parameters such as the number of periods N, barrier thicknesses, the absorption α and the specific detectivity D * .A description of the design parameters, their influence and optimization considerations is presented in more detail in [22][23][24].
In this paper we demonstrate a quantum cascade detector in pixel configuration processed as array [Fig.1].The detector is substrate bottom side illuminated with an Au diffraction grating processed into the top contact layer.The active region is optimized for a high device resistance.Thereby a resistance increase in the entire temperature range was achieved.Sufficient QCD resistance for operation with the custom read out integrated circuit (ROIC) can be ensured from 80 K up to 240 K.The presented device is based on the InGaAs/InAlAs material system grown lattice matched by molecular beam epitaxy (MBE) on semi-insulating InP substrate.

Heterostructure
At shorter mid-infrared wavelengths a resonant tunneling extraction scheme is more favorable than a diagonal optical transition.Due to the higher energy of the optical transition in this wavelength range back scattering to the ground level of the active region is reduced.The band structure of one period is illustrated in Fig. 2(a).The active transition takes place in well A, where A' denotes the active well of the next period.The simulated extraction efficiencies are calculated for all subsequent levels including the LO-phonon staircase (B-J).An electron in the ground level of well A can be excited to one of the three degenerated excited states in well A (delocalized between A, B and C) which have extraction efficiencies of 81.84%, 80.58% and 86.04%, respectively.These efficiencies express the probability of an electron, excited to the corresponding level, to reach and to be captured in the ground level of the next QCD period (well A').The extraction efficiency is calculated as a weighted average of these levels.The total extraction probability calculated considering the absorption strength for each level allows for extraction optimization from the active well (A/A') to the first extractor level (B/B').In order to identify bottlenecks, the extraction efficiency is calculated also for all following LO-phonon staircase levels.The transition between well J and A' is chosen to exhibit a higher energy difference to prevent thermally excited carriers from back-filling the lower staircase levels of the previous period.This is important to achieve a temperature insensitive responsivity as thermal back-filling would reduce the absorption efficiency at elevated temperatures.The device consists of 20 periods with a total active region thickness of 1.36 µm sandwiched between a 800 nm InGaAs top contact layer and a 500 nm InGaAs bottom contact layer, both Si doped to 1.5e18 cm −3 .A period of the active region consists of the layer structure 4.3/6.5/0.85/6.5/0.9/6.5/1.0/5.5/1.2/4.5/1.45/4.5/1.7/4.5/2.05/4.0/2.4/4.0/2.85/3.0nm starting with the active well A, InGaAs in bold, InAlAs normal and the 8e17 cm −3 doped well underlined.
The Vienna Schrödinger Poisson (VSP2) frameworks semi-classical Monte-Carlo transport simulator originally developed for quantum cascade laser design was extended for QCD design and simulations [20,25].The tools allow for simulation of the intersubband scattering rates for short wavelength detectors, the absorption, differential device resistance, responsivity and the specific detectivity.The model accounts for all possible scattering transitions and is based on three consecutive periods.For extraction efficiency simulation, a scattering event contributes to the current flow, if it scatters from the middle period to the right or to the left adjacent period.From the resulting scattering matrix the outbound scattering rates of the ground levels are deleted and all electrons are initially set to be in the excited state(s) of the active well (A/A').The scattering matrix is then applied n times to the initial population.After n events the electrons accumulate in the ground states which have no outbound scattering.The ratio between the amount of electrons in the ground state of A to A' then gives the total extraction efficiency.The absorption efficiency η eff is modeled for 45 • mesa devices considering the facet transmission T facet ≈ 70% and all possible optical excited intersubband transitions by the absorption coefficient α isb = i, f α i f in the active region.
where N p is the number of QCD periods and L p the length of a period.The simulated responsivity R(λ) is calculated based on the absorption efficiency.In contrast to QWIPs the capture probability p c for QCDs can be assumed to be close to unity.The peak responsivity is then given by with the total detector current I s , the total incident radiation power P s , the elementary charge q, Plancks constant h, speed of light c, absorption efficiency η abs , photodetector gain g p and the extraction efficiency p e .In-plane non-parabolicity is included in the model to correctly predict the red-shift with increasing temperatures [Fig.2(b)].The QCD differential resistance model is based on the work of Delga et al. [26].All scattering transitions between subbands are replaced by a conductance.An equivalent conductance circuit is extracted, which is solved by node analysis to obtain the device resistance.The model accounts for inter period transitions by periodic repetition of the scattering matrix.

Optical and array design
The QCD is designed as a substrate bottom side surface-normal illuminated pixel device.A QCD array was processed and flip-chip bonded to a CMOS ROIC.Since intersubband detectors are only sensitive to light with the electric field component polarized in growth direction a coupling scheme to the active region is required.Light coupling is achieved by a two dimensional 2 nd order Au grating with a grating period of p = 1.4 µm, depth d = 450 nm and duty cycle 0.4.The pixel to pixel pitch is 225 µm in both directions, with a pixel size of 109 µm × 109 µm.The pixel pitch is determined by the size of the prototype RIOC amplifier circuit for each pixel.This size can be reduced in future designs.
The optical simulations for light coupling were conducted with the active regions absorption α isb (λ) retrieved from our QCD model.Absorption was taken into account only in growth direction to account for the intersubband selection rule.For all materials the mid-infrared complex refractive indices n(λ) were taken into account.The grating depth, duty cycle and grating period were optimized for maximum coupling efficiency.The Au layer of the grating is used as the top contact of the pixels and serves as the bonding pad for Au thermo-compression flip-chip bonding.

Results
The QCD material was first characterized as a 100 µm × 100 µm mesa device.As intersubband devices are subjected to the intersubband selection rule, a 45 • facet was polished onto the sample for light coupling.The photocurrent spectra were recorded by FTIR (Fourier transform infrared) spectroscopy illuminated by a broadband globar source.In order to determine the detectors wavelength dependent responsivity R(λ) the power density spectrum and total power of the broadband globar source were recorded.With respect to the measured parameters the responsivity from Eq. ( 2) can be expressed as The photocurrent spectrum I s (λ) is obtained by the FTIR recorded photocurrent spectrum normalized by the measured total photocurrent I tot .The total incident power P i is determined by the spatial power distribution of the focused globar source and its fraction impinging on the 45 • facet.The factor M accounts for the polarization selection and the cryostat window transmission.The cryostat ZnSn window transmission was determined to be 75%.For mesa devices the polarization was accounted with a factor of 0.5 since 45 • facet devices are only sensitive to light polarized in growth direction (M = 0.75 × 0.5).Due to the 2D metal array the pixel detectors are sensitive to unpolarized light (M = 0.75).
Figure 3 shows the temperature dependent responsivity and differential resistance of a 45 • mesa device.The two pronounced valleys are due to the strong CO 2 absorption bands.In contrast to the simulation a responsivity increase between 80 K and 160 K can be observed.This may be accounted to carrier freeze-out or the increasing dominance of coherent tunneling and electron-electron scattering which are not included in the simulation model.The characteristic red-shift with increasing temperatures is due to strong in-plane non-parabolicity.Due to the optimization for higher operation temperatures, the CO 2 absorption is mostly aligned with the photocurrent peak at temperatures between 240 K and 300 K.The measured responsivity is slightly below the simulated performance illustrated in Fig. 2(b).This mismatch can be accounted to the Ti /Au metal-semiconductor contact resistances and the contact layer to first QCD period resistances, which are not included in the simulator model.From previous experiments we noticed that the design of the contact layer to the first QCD period is relevant, especially if the contact layer would be low doped [27].The pixels are processed as sparse pixels which do not cover the maximum available area.This design decision is due to the resistance limitations of the prototype read out circuit.With the pixel mesa size of 109 × 109 µm a resistance of 13.9 kΩ at 240 K operation temperature could be realized which is well within the ROIC specifications [Fig.4(b)].From the finished array single pixels were directly wire bonded.The single pixel measurements were conducted prior to flip-chip bonding.Figure 4(a) shows the single pixel responsivity of a typical processed array with a SiN x anti-reflection coating (ARC).As for the 45 • mesa devices strong CO 2 absorption is observed.The responsivity increases between 80 K and 160 K and decreases towards 18 mA/W at the design temperature.Due to the high coupling and absorption efficiency of the pixel design the responsivity decrease between liquid N 2 temperature and room temperature is less pronounced in comparison to 45 • mesa devices.The simulated peak absorption efficiency of a single pixel based on the real sample geometry is illustrated in Fig. 4(d).
Grating coupling is designed around the peak at 2350 cm −1 with the electric field component in growth direction shown in the inset of Fig. 4(d).A slightly higher doping was used in the model to match the simulations with the experiment.The optimization for increased QCD resistance results in a high detectivity even at room temperature [Fig.4(c)].
The arrays substrate side was polished and reduced in thickness to ≈ 250 µm.For the presented 8 × 8 array complete substrate removal is not necessary due to the small total size, large pixel pitch and comparable high operation temperature which results in a low ∆T ≈ 70 K.The finished 8 × 8 pixel device is packaged into a 16 pin TO-8 housing [Fig.5

Conclusion and outlook
Based on our customization of the VSP2 frameworks single particle Monte-Carlo transport simulator we have shown a high performance QCD designed at the CO 2 absorption bands around 4.3 µm.These detector specific models enabled for a resistance increase and an over all performance increase in comparison to previously shown III/V material based mesa devices at this wavelength [14,19,20,22,28].The calculation of the extraction efficiencies for every extractor level allows to identify bottlenecks and to increase the total extraction efficiency p e .Processing improvements for the grating structure could further improve pixel performance.
With the performance of our QCD design we show the potential of this detector type and envision its further improvement to be used for arrays in imaging applications.Remote sensing would benefit from spectrally narrow detection as provided by QCDs and provide additional scene information based on the design wavelength and the thereby detectable substances such as CO 2 .

Array processing
The MBE grown material is cleaned with acetone in an ultrasonic bath and isopropanol prior to all processing steps.A 420 nm SiN x hardmask is deposited for the 2 nd order diffraction grating etch process and structured by electron beam lithography.After the e-beam lithography the grating dimensions were evaluated by atomic force microscopy.The SiN x hardmask is etched with an CHF / O 2 dry etch process optimized for highly anisotropic etching.The grating is structured into the top contact layer by reactive ion etching.In the next step the pixels are structured by reactive ion etching with a SiN x nitride hardmask.The pixels are isolated from each other by a wet chemically etched 6 µm wide isolation trench which cuts through the bottom contact layer.The next step involves the deposition of an insulation layer of 350 nm SiN x and the opening of the contact windows.The Ti/Au (10 nm/440 nm) contacts are sputter deposited using a negative resist lift-off process.The relatively high Au layer thickness is a requirement for the Au thermo-compression flip-chip bonding process.After contact deposition the sample is cleaned with an ultrasonic acetone bath followed by isopropanol and cleaved to the final array size.The arrays have to be cleaved with minimal remaining edge width surrounding the array to make the ROIC's digital and analog I/O bond-pads accessible by the bond wires.The samples InP substrate bottom side is then polished by wet mechanical polishing.The 580 nm SiN x anti-reflection coating is then deposited by plasma enhanced chemical vapor deposition directly onto the polished substrate bottom side.Au thermo-compression flip-chip bonding is  done directly onto the grating surface of the pixels and the corresponding bottom contacts.Due to the relatively low mesa height no planarization step is necessary.

Fig. 1 .
Fig. 1.Scanning electron microscope image of a section of the (a) 8×8 pixel array with separate bottom contacts, (b) a detailed SEM image of a single pixel with the corresponding bottom contact (square without grating structure) surrounded by the isolation trench.The inset (c) shows the metal grating in detail.

Fig. 2 .
Fig. 2. Band diagram (a) of one QCD period with the corresponding extraction efficiencies for all levels of the extractor and LO-phonon staircase calculated for 300K.The active region consists of 20 periods.The simulated QCD responsivity (b) is based on the simulated absorption efficiency of a 45 • mesa device.

Fig. 3 .
Fig. 3. Photocurrent spectrum and temperature performance of the 4.3µm 20 period QCD design measured with a 100µm × 100µm mesa device.The illumination is through a 45 • polished facet.
(b)] with a spot welded sealed cap purged with N 2 .The assembly consists of the two stage TEC, the thermo-compression bonded array and ROIC, a Ge window soldered to the cap and a Cu cold shield [Fig.5(a)].The cold shield is thermally connected with the TEC.The Au thermo-compression bonded prototype array with the ROIC is wire bonded to the TEC element and to the pins of the TO-8 housing [Fig.5(c)].

Fig. 4 .
Fig. 4. Photocurrent spectrum and temperature performance (a) of the 4.3µm 20 period QCD design measured with a 109µm × 109µm pixel device.The illumination is through the ARC coated substrate side.The two visible valleys in the spectrum are due to the strong CO 2 absorption in the beam path.The simulated and measured differential resistance R 0 (b) show good agreement in the entire temperature range.The measured peak responsivity R Peak and the specific detectivity D * (c) were measured by a calibrated globar source.The field of view Θ for D * BLIP is π.The grating coupling simulation of the absorption efficiency (d) of a single pixel with α abs (λ) provided by the QCD simulator.The inset depicts the normalized electric field component E y due to the metal grating coupling.

Fig. 5 .
Fig. 5.The housing of the prototype consists of (a) the cap, the base with the two stage thermo-electric cooler, the bonded ROIC with the QCD array, a cold shield and the antireflection coated Ge window.The array wire bonded viewed in illumination direction is depicted in (b).The packaged array (c) is purged with N 2 .Flip-chip bonding and packaging as depicted was developed and conducted by Faunhofer IZM Berlin.