Hawaii-4 RG Detector for NIR Astronomy

• Hawaii-4 RG detectors are advanced near-infrared imaging arrays with 4096×4096 pixels, offering high quantum efficiency and low noise for rigorous scientific applications.
• They utilize a hybrid HgCdTe photodiode array bonded to a CMOS readout circuit, available in 15 µm and 10 µm pixel pitches to optimize resolution and sensitivity.
• Their state-of-the-art design, with bespoke cryogenic electronics and correction pipelines, supports high-precision exoplanet and cosmological surveys.

The Hawaii-4RG (“H4RG”) detector family comprises large-format, hybrid HgCdTe-CMOS near-infrared (NIR) imaging arrays developed for demanding astrophysical applications. With 4096 × 4096 pixels, enhanced quantum efficiency, low noise, low dark current, and resilience to environmental stresses, H4RGs underpin major missions such as MOONS and WFIRST. Key variants include the H4RG-15 (15 µm pitch) and the H4RG-10 (10 µm pitch), both configurable for 2.5 μm cut-off, with bespoke cryogenic readout electronics and advanced correction pipelines. Their architecture and performance represent the state-of-the-art in NIR scientific detectors, enabling deep multi-object spectrographs and high-precision exoplanet and cosmological surveys (Ives et al., 2020, Jr. et al., 2020, Bechter et al., 2019).

1. Detector Architecture and Key Design Features

H4RG detectors utilize a substrate-removed HgCdTe photodiode array hybridized (via indium bump-bonding) to a 64-channel (H4RG-15) or 32-channel (H4RG-10) CMOS readout integrated circuit (ROIC). The H4RG-15 has a 15 µm pixel pitch (61 mm × 61 mm die), while the H4RG-10 uses 10 µm pixels. The active material bandgap is compositionally tuned for a 2.5 µm cut-off wavelength. Each pixel incorporates a source-follower amplifier and multiplexing circuitry.

The detectors implement both unbuffered (“science”) and buffered (“low-Z”) output configurations. In unbuffered mode, 32 outputs with high impedance ($Z_\text{out}\sim100$–$200\,\mathrm{k}\Omega$) are used; in buffered mode, 64 outputs with an additional on-chip source-follower reduce $Z_\text{out}$ to $<1\,\mathrm{k}\Omega$. The buffered configuration utilizes a novel, miniaturized cryogenic differential preamplifier assembly with quad ultra-low-noise CMOS amplifiers (e.g., Texas Instruments OPA4192) and Vdda pull-up topology, optimized for <50 × 50 mm² footprint to suit the constrained optical environment of fast Schmidt camera designs (Ives et al., 2020).

2. Electro-Optical Performance Metrics

Operational performance is characterized at temperatures from 40–100 K. Key parameters are summarized below:

Metric	H4RG-15 Value	H4RG-10 Value
Pixel pitch	15 µm	10 µm
Array size	4096×4096	4096×4096
Cut-off wavelength	2.5 µm	2.5 µm
Full-well capacity	>70 ke⁻	~100 ke⁻
Read noise (CDS)	13 e⁻ rms	12 e⁻ rms
UTR-128 noise	4.5 e⁻ rms	7 e⁻ rms
Dark current (80 K)	0.009 e⁻/s/pix	<0.01 e⁻/s/pix
Quantum efficiency	~90% (J,H)	86–96% (median 91%)
Inter-pixel cap.	0.8–1.4% (rows)	1.8–2.6% (4-nn sum)
Crosstalk (buf/unbuf)	<0.01% / ~0.1%	n/a

In both devices, quantum efficiency exceeds 90% through J and H bands. For the H4RG-15, the read noise remains at 13 e⁻ rms (CDS) and drops to <3 e⁻ rms in Fowler-8 sampling. The UTR-128 slope-fitting method yields ~4.5–7 e⁻ rms total noise, with marginally higher values in longer up-the-ramp (UTR) sampling due to charge injection and glow. Dark current at 40 K is 0.004 e⁻/s/pix for good pixels; hot-pixel operability improves from 92% (90 K) to 99% (40 K). For the H4RG-10 at 100 K, median dark current is 0.002 e⁻/s/pix.

3. Noise Sources, Persistence, and Correction Strategies

Noise in H4RG detectors comprises white (Johnson and shot), pink ($1/f$), alternating column, and picture frame noise. The overall read noise per pixel is described as:

$$\sigma_\text{read}^2 = \sigma_w^2 + \sigma_\text{p,corr}^2 + \sigma_\text{p,uncorr}^2 + \sigma_\text{acn}^2 + \sigma_\text{pf}^2$$

where each term corresponds to a distinct physical noise source (Bechter et al., 2019).

Persistence, modeled as a multi-exponential trap population, is parameterized via:

$P(t) = \sum_i A_i \exp(-t/\tau_i)$

Fitting to SRH kinetics yields a trap activation energy $E_\text{trap}\approx0.13\pm0.01\,\mathrm{eV}$, with persistence peaking at 55–70 K and minimized ($<1\%$ full-well) at $T\leq50$ K or $T\geq80$ K (Ives et al., 2020). Data pipelines utilize per-pixel trap-density maps and on-the-fly subtraction to correct for persistence based on measured exposure history.

Nonlinear pixel response is corrected empirically using cubic fits ($$f_\text{corr}=f_\text{meas}+c_2f_\text{meas}^2+c_3f_\text{meas}^3$$), removing nonlinearity effects critical for precise photometry and Doppler spectroscopy (Bechter et al., 2019).

4. Readout Modes, Programming, and Operational Features

Both unbuffered and buffered readout configurations are available. In unbuffered mode, high output impedance results in capacitive crosstalk (∼0.1%) and row droop (up to 0.5% at ¾ full-well); buffered output with the custom cryogenic preamplifier reduces crosstalk below 0.01% and supports faster pixel clocks (up to 300 kHz/output) with no noise penalty (Ives et al., 2020). The “Current-Boost” enable stabilizes unbuffered outputs, and direct programming of on-chip registers (notably “Register-1”) is mandatory to achieve full well and conversion gain. Enhanced-Clocking Mode is not recommended for scientific data due to column offsets.

Management of photo-emission defects (PEDs) is achieved by row-skip and substitution; column de-select alone is counterproductive. The column-deselect and row-skipping features reduce the impact of local emission artifacts. Detailed programming and calibration, including careful use of reference pixels, are critical to maintaining optimal detector performance.

5. Charge Injection, Inter-Pixel Capacitance, and Data Calibration

Charge injection per read is minor but spatially variable: for H4RG-15, the center region yields ~0.03 e⁻/read while edges near bond pads can reach 0.65 e⁻/read. Recommendation is to keep UTR reads at ≤128 to limit glow noise.

Inter-pixel capacitance (IPC) is characterized by nearest-neighbor coupling of 0.8–1.4% (rows) and 0.5% (columns) in the H4RG-15 and 1.8–2.6% four-neighbor sum for H4RG-10 (Jr. et al., 2020). IPC induces PSF broadening and must be corrected in high-precision applications. IPC kernels are measured directly using single-pixel reset and X-ray events and show modest spatial variation and a systematic gradient towards device edges. Data pipelines apply cross-talk corrections ($\lesssim1\%$) to maintain photometric and spectroscopic fidelity.

6. Environmental Qualification, System-Level Impact, and Applications

H4RG detectors have undergone rigorous environmental qualification including: 40 thermal cycles between 295–100 K, GEVS-level sine/random vibration, acoustic testing, and proton irradiation. H4RG-10 arrays retain >99% pixel operability and exhibit minimal performance drift post-exposure: hot pixels increase by ~3.5%, dark current rises negligibly (from 0.003 to 0.012 e⁻/s), and noise increment occurs in <8% of pixels (but remains within specification) (Jr. et al., 2020). The arrays are rated at TRL-6 for mission deployment.

At the system level, the product of high QE, low dark current, and low noise yields the deep imaging (e.g., $\mu_\mathrm{AB}>26$), precise spectroscopic, and photometric capabilities required for modern NIR surveys and spectrographs (e.g., MOONS for VLT, WFIRST for cosmology and exoplanet searches) (Ives et al., 2020, Jr. et al., 2020, Bechter et al., 2019). Data calibration pipelines are equipped to handle nonlinearity, persistence, IPC, and noise characterization, critical for minimized error budgets in high-precision Doppler velocity measurements (e.g., exoplanet radial velocities down to $0.5$–$1.0\,\mathrm{m/s}$ uncertainties with sub-$1\,\mathrm{m/s}$ individual exposure precision) (Bechter et al., 2019).

7. Summary Table of H4RG Key Parameters

Characteristic	H4RG-15	H4RG-10
Pixel count	4096 × 4096	4096 × 4096
Pixel pitch	15 µm	10 µm
Cut-off wavelength	2.5 µm	2.5 µm
Detector material	HgCdTe, sub-rmvd	HgCdTe, sub-rmvd
Full-well (typical)	>70 ke⁻	~100 ke⁻
Read noise (CDS)	~13 e⁻	~12 e⁻
UTR slope fit noise	~4.5 e⁻ (UTR-128)	~7 e⁻ (55 ramp)
Dark current (good pixels)	0.009 e⁻/s/pix @ 80 K	<0.01 e⁻/s/pix @ 100 K
Quantum efficiency	∼90%	86–96% (median 91%)
Inter-pixel capacitance (IPC)	0.8–1.4% rows,	1.8–2.6% (4 neighbors)
	0.5% cols
Crosstalk (buffered/unbuf)	<0.01% / ~0.1%	n/a

These characteristics are central to the performance envelope for high-stability NIR imaging and spectroscopy applications, with system calibration and data processing methods in place to fully exploit their scientific capabilities (Ives et al., 2020, Jr. et al., 2020, Bechter et al., 2019).