3.1.1.

HgCdTe detector development plan

To make HgCdTe viable for use on Origins will require ROIC stability improvements, including major development programs that address decreasing read noise, alleviating nonlinearity effects, reducing pixel-to-pixel variations, and resetting anomalies. The remaining challenges are to: (1) increase the full well, which may require a selectable capacitor/pixel addition to the ROIC, which is fairly straight forward and (2) redesign the ROIC to decrease read noise and change pixel size while not increasing dark current or adversely affecting the image quality [modulation transfer function (MTF) or point spread function (PSF)]. The second challenge can be investigated and worked out with multiple production iterations (see task 4).

HgCdTe has benefitted from existing development work for JWST ($5.0\mu \mathrm{m}$ cutoff), NEOSM ($10\mu \mathrm{m}$ cutoff), and UR’s $15\text{-}\mu \mathrm{m}$ cutoff development aimed at supporting spectroscopic detection of biosignatures via space missions, such as Origins. (The ARIEL mission will also require similar detectors to those on NEOSM.) For the past three years, UR has been extending this technology to longer wavelength cutoffs. The results for $13\text{-}\mu \mathrm{m}$ cutoff wavelength detector arrays were very promising with dark currents, well depths, and operabilities^23–25 similar to those of the earlier $10\text{-}\mu \mathrm{m}$ cutoff wavelength detector array produced for NEOSM. For the 15- to $16\text{-}\mu \mathrm{m}$ cutoff wavelength detector arrays produced in the second half of the UR development program,²⁶ the dark currents were higher and the well depths were lower than Origins requirements. However, those requirements are set by the needed sensitivity at shorter wavelengths, i.e., it would be possible to use HgCdTe arrays to cover from 2.8 to $16\mu \mathrm{m}$ using two or three cut-off wavelength arrays, e.g., 6, 11.5, and $16\text{-}\mu \mathrm{m}$ cut-off wavelength detector arrays, and still meet Origins requirements at each of those bands.

The growth of HgCdTe detectors is often fraught with difficulties, e.g., low well depth and high dark current due to tunneling currents, softer material at longer wavelengths, which makes hybridization more difficult, cluster defects such as those from Hg precipitates, thinning of substrates, and targeting the cut-off wavelength. Although each of these issues may be understood in principle, it is difficult to fully constrain all of them in practice, which results in lower yield of detectors, which in turn, results in higher costs to produce the flight detectors for any given program. Taking what is learned during both JWST’s and NEOSM’s production runs of flight detector arrays (heritage reduces cost and risk) will aid in improving detector yield.

The next steps are to address the remaining requirements for Origins: stability, read noise, and well capacity. At low enough temperatures (35 K for $10\text{-}\mu \mathrm{m}$ cutoff and 27 K for $13\text{-}\mu \mathrm{m}$ cutoff), the dark currents are bias dependent, i.e., over the course of an integration, charge accumulation will cause the dark current to decrease. This is an inherent instability in that kind of dark current. One possibility is that dark current stability can be reached by operating at a sufficiently high temperature (28 K) so the detector is thermally dark current limited (G–R) instead of bias-dependent dark current limited (trap-assisted tunneling), while still meeting the $0.3{\mathrm{e}}^{-}/\mathrm{s}$ dark current requirement for Origins. It will need to verify whether or not a more stable but slightly higher dark current is capable of producing better overall stability as compared to the instabilities due to the bias-dependent dark current. Teledyne HxRG ROICs, generally used for low background space astronomy, are source follower per detector pixel and the lowest noise and low-power choice. Although a capacitive transimpedance amplifier capacitive transimpedance amplifier ROIC could provide constant bias, which might initially seem the most stable choice, other reasons (e.g., high power—typically Watts rather than milliwatts—and glow) make this ROIC an unsuitable choice for cryogenic temperature space missions, at least as they are currently envisioned. The high power occurs because all pixels are powered during the read of each pixel. Some detector groups have discussed turning off the power to all but the pixel being read out; however, then turning on the power to each pixel would be unstable and still generate more power than a few milliwatts. Extra closed cycle cryo-refrigerators would be required, adding cost and weight to the mission. The high current would also lead to ROIC glow, which affects the background flux levels in an unstable way. The most stable performance from arrays comes with not undergoing any changes from integration to integration (including keeping the same background power and signal position on the detector, which is exactly how exoplanet transit observations are done). First efforts will be directed toward improving the HxRG ROICs before considering a completely new cryogenic ROIC design.

Redesigning the ROIC will, by definition, lower the complete detector array’s technology readiness level (TRL) to 4. Once the design is complete and tested in a lab environment, it will be back at TRL 5. We expect that the primary need for retesting will be susceptibility to cosmic ray damage, as HgCdTe detector material is susceptible to damage from cosmic rays. The current $10\text{-}\mu \mathrm{m}$ cutoff HgCdTe arrays have been qualified to TRL 6 for the NEOSM mission after proton irradiation experiments where $<1\%$ of the pixels were degraded (increased dark current) after $1.5\times$ and $2.0\times$ lifetime total ionizing does for the L1 environment.²⁷ The HAWAII-2RG HgCdTe arrays have been qualified to $>\mathrm{TRL}6$ for JWST, OCO-2, HST, and ECHO.²⁸ The redesigned detector array/ROIC will then need to pass proton radiation testing, thermal cycling, and vibration testing to once again reach TRL 6. With sufficient funding, this process will take $\sim 4$ to 5 years.

3.1.2.

HgCdTe detector manufacturing

The primary objective is to improve the overall radiometric stability of the HgCdTe detector arrays. This requires several tasks, as detailed below.

Task 1. The most important objective is to decrease the noise, both the long-term (between exposures taken over one day) $1/f$ noise component and the short-term (single image) nonrandom components of the read noise. The $1/f$ noise component directly impacts the long-term radiometric stability since it adds uncertainty to the measurement that cannot be removed via repeated sampling to reduce the noise. Popcorn (random telegraph) noise is another form of both traditional read noise and $1/f$ noise that follows a power law distribution in both frequency of transitions and voltage of transitions. The lowest frequency popcorn noise has the largest voltage transition. There are a few methods to reduce $1/f$ noise. First, the use of buried channel metal oxide silicon field effect transistor (MOSFETs) can reduce the $1/f$ and popcorn noises. Second, switching to an all P-channel metal–oxide–semiconductor (PMOS) process [i.e., not fully complementary metal–oxide–semiconductor (CMOS)] can reduce the number of defects introduced during processing, i.e., fewer dopants and thus fewer contaminants entering the silicon. The process itself could be cleaned to the point of removing the contaminant responsible for the popcorn noise. Unless the read noise is very large, exoplanet transit observations will be photon shot noise limited due to bright host stars in the raw data.

The source-follower FET has a thermal shot-noise component due to the current that flows in the FET. That thermal noise is white across frequencies and thus may be reduced by decreasing the bandwidth of the FET. The reduction of the bandwidth is achievable via increasing the number of ROIC outputs while decreasing the rate with which each pixel is read, i.e., $20\mu \mathrm{s}/\text{pixel}$ instead of 10. This option increases the power consumption per array due to the increased number of outputs, however, that is still within our budget of 100 mW per array. This option is unlikely to be needed since the thermal shot noise will of the ROIC will be reducible with multiple sampling and will be well below the photon shot noise of the individual observations.

Both the dopant and concentration of the dopant in the silicon wafer can be adjusted to improve the cryogenic performance of the ROIC.²⁹ In addition, it is possible to improve the foundry process to make each layer more planar, i.e., place more oxide between FETs and traces such that the overlying oxide on those traces is flat and plane parallel with the starting silicon wafer. Without such a process improvement, the oxide becomes thinner at the corners of the traces (as seen in cross section) and leads to microshorts between crossing traces or the substrate.³⁰ Adjustments to the combined parameters of the insulating oxide thickness, its planarization, the dopant, and its concentration will help to reduce microshorts within the ROIC, which, in turn, will reduce currents in the substrate, thereby reducing clock feed-through, which is a nonrandom component of the read noise.

Task 2. Implement design improvements to improve the stability of clocks and biases, which, in turn, will improve the overall detector stability. This is primarily a strict analysis of the design used to reduce stray capacitances and any potential shorts or crosstalk between adjacent circuits, e.g., move shift registers away from sensitive bias lines and output amplifiers (source follower FETs), adding more metal layers for shielding, employing lower voltages for switches, and dual polarity (differential) switches. The designs will also need to minimize the resistance of long lines for voltage supplies to decrease the voltage drop that occurs from one side of the ROIC to the other. The increase in pixel size will provide the ability to reduce resistance on long supply lines (reset voltage and drain voltage) using larger width traces, while also spacing those traces better to decrease the amount of stray capacitance. The improvements that were introduced into the HAWAII-4RG ROIC design to minimize clock feed-through will also be utilized in these designs. Finally, this task period will also cover optimizing the external electronics, such as voltage supplies and preamps, whose performance will also be critical in achieving the required detector stability.

Task 3: Improve the pixel-to-pixel uniformity. An increase in pixel size will allow for better pixel-to-pixel uniformity since the FETs will be well above the design rule limits for the chosen foundry, and thus inherent foundry process variations in the FET dimensions will be a smaller fraction of the area of the FETs. It should be noted that the densified pupil design also mitigates the requirement of better pixel-to-pixel uniformity because the densified pupil spectrum always forms on the same position of the detector as long as the target object is in the field of view of the MISC-T instrument.

Task 4. Change the pixel pitch from $18\mu \mathrm{m}$ to an optimized value derived by the final optical design of the MISC-T instrument. The existing HAWAII-2RG $2\mathrm{k}\times 2\mathrm{k}\text{pixel}$ ROICs used for HgCdTe arrays have $18\text{-}\mu \mathrm{m}$ pixel pitches, which may not be optimum—for example, pitches that are approximately diffraction-sized do not produce good image quality (e.g., MTF³¹). Further, increasing the size of the pixel pitch will allow more room within each pixel to accommodate changes made to satisfy the other tasks, both above and below.

Increasing the pixel pitch is not without its disadvantages. The final ROIC, produced after Origins selects its optimal design, will be large in physical area, and therefore, require stitching of reticles during the foundry process in order to create the full array. The optics for any future telescope will be more difficult to design given that the array extends over a larger area of the focal plane, thus putting more strain on the optical designer to deliver a flat focus over that larger area, although it should be noted that slower optical systems are generally less susceptible to aberrations. Furthermore, the size of the actual detector die will cover a larger portion of the HgCdTe wafers, which results in greater risk of nonuniformity of the detector and thus impacts the yield. With a larger detector die to be hybridized to the ROIC comes a larger force needed to bond all the pixels. This will require a short development period in order to be able to reproducibly hybridize the larger HgCdTe die to the larger ROIC die.

A pixel pitch increase would normally increase the dark current, i.e., larger p–n junctions for larger pixels. However, the size of the p–n junction needs not change, hence, dark current will not change. Despite having smaller p–n junctions relative to the pixel area, a pixel pitch increase will not adversely affect the lateral charge collection by a large amount due to Teledyne’s proprietary techniques in growing the HgCdTe. The MTF will degrade slightly for shorter wavelength light ($<6\mu \mathrm{m}$) due to the larger path lengths for diffusion. However, for wavelengths $>6\mu \mathrm{m}$, the MTF will actually improve for larger pixel spacing due to the fact that the pixel size will be larger than the wavelengths of the collected photons, minimizing diffraction effects.

The interpixel capacitance (IPC) will also be affected by the change in pixel size. However, an increase in pitch will result in a decrease in IPC, which is a beneficial outcome, and an increase in the pixel pitch would allow more room to accommodate an increase in the capacitance allowing a commensurate increase in well depth as described in Task 5.

All of these advantages and disadvantages will need to be properly weighed during a trade study on the pixel size. Although the current goal is to produce $15\text{-}\mu \mathrm{m}$ pixels (as given by the optics team to the detector team), it is likely that Origins will decide to use pixels that are larger than that.

Task 5. Add (switchable if needed) internal capacitance to each unit cell (pixel) to increase the well capacity and reduce the bias dependence of DQE and dark current by effectively reducing the contribution of the detector to the total capacitance. This unfortunately will also increase the effective read noise since the read noise is actually a voltage noise despite always being quoted in electrons. The increased read noise must be traded against the gain in well capacity in order to find the optimal value for the added internal capacitors.

Task 6. Once a final ROIC design is made, it can be qualified to TRL-6. Part of that qualification process involves proton radiation testing, which can be conducted on bare ROICs.

Task 7. The final objective is to produce hybridized long wavelength infrared ($11\text{-}\mu \mathrm{m}$ cutoff) HgCdTe detector arrays utilizing the best ROIC design. These arrays can then be tested against the Origins requirements. A small number of these arrays will be subjected again to radiation testing, thermal cycling, and vibe/shake testing to complete the full array qualification to TRL-6.