Detector Technologies¶

ZShooter is leaning into cutting edge detector advances where detector properties enables new observing models, while keeping practical fallback paths. The detector choices are not decorative upgrades: they determine how well ZShooter can trade cadence, read noise, and spectral resolution without making the instrument itself the bottleneck.

Detector Baseline¶

The baseline detector architecture follows directly from the instrument concept: visible ZSpec channels are designed around low-noise qCCDs with spatially varying anti-reflection coatings, the infrared channels are designed around linear-mode avalanche photodiode arrays, and ZImager uses qCMOS detectors modules for fast, low-noise, simultaneous tri-band imaging. The fallback strategy is conservative: conventional CCDs and H2RG-family infrared arrays can preserve the basic instrument function, while only sacrificing some of the full cadence/noise/dark current advantages.

Detector technology summary¶
Subsystem	Baseline	Fallback / alternate	Why it matters
ZSpec optical	qCCDs with tapered ALD AR coatings	Conventional STA or Teledyne-e2v CCDs	Sub-electron effective read noise makes short sub-exposures, post-facto binning, sky-line editing, and high native resolving power scientifically useful rather than read-noise dominated.
ZSpec IR	Low-noise HgCdTe LmAPDs, one 2k × 2k-class array per YJ/H/K channel	H2RG or H4RG-family HgCdTe arrays	Much of the IR band is detector-noise limited between OH lines. LmAPD avalanche gain reduces the read-noise penalty before multiplexor/readout noise is added.
ZImager	ORCA-Quest 2-class qCMOS cameras, final model TBD	N/A, detectors in use and commercially available	Near-quantizing performance, an electronic shutter, and high frame-rate subrasters support acquisition, photometric anchoring, occultations, and compact-binary timing.

Spectral Channels¶

The post-concept-review optical prescription update moves ZSpec toward two spectrometers, each split into three channels and each formatted onto a 2k × 2k active detector area. The active detector format is deliberately technology agnostic: it can be satisfied by available detector classes and by the likely formats of the qCCDs and LmAPDs ZShooter intends to field.

Post-concept-review channel coverage¶
Spectrometer	Channel	Detector implication
Visible	Blue	Optimized for the atmospheric blue cutoff, qCCD coating and blue QE matter disproportionately.
Visible	Green	Separates the highest-throughput visible region from the blue and red channels, improving cross-disperser efficiency relative to the older two-visible-channel split.
Visible	Red	Covers the red optical arm on a 2k × 2k active format and supports red-sensitive bulk qCCD material.
Infrared	YJ	2k × 2k LmAPD target format; no detector mosaic required
Infrared	H	Improves efficiency balance and somewhat isolates high-sky-background regime design trades.
Infrared	K	Preserves K-band science while making thermal background, detector dark current, and telescope impacts tightly contained in one channel.

qCCDs & Tapered Coatings¶

A qCCD is a CCD operated with a skipper-style nondestructive readout architecture: the same pixel charge packet can be sampled repeatedly before it is finally cleared. If the samples are sufficiently independent, the effective read noise falls approximately as the square root of the number of samples. The result is not merely a lower-noise CCD; it is a detector that can approach direct charge quantization for optical photons while retaining CCD strengths such as high QE, large-area formats, linearity, and on-chip binning.

The optical spectrograph is intentionally high-resolution for a broad set of science cases, but many programs want to synthesize lower resolution after the fact. With ordinary CCD read noise, collecting a spectrum at high native resolution and then rebinning it is not equivalent to observing directly at lower resolution: the read-noise contribution has already been paid in too many pixels. qCCDs change that trade. They let ZShooter collect more spectral information at the detector and defer some of the resolution/SNR decision into the reduction, where sky lines, cosmic rays, and science-specific line windows can be handled more intelligently.

For the current ZShooter concept, the optical detector path has three separate pieces:

Differential skipper readout. Differential sampling of the signal and reference levels reduces the speed-noise penalty relative to earlier skipper implementations.
Many parallel outputs. Segmenting the serial register into many outputs keeps the readout time compatible with observational use rather than particle-physics-style integrations.
Frame-transfer operation. A frame store hides much of the readout overhead, preserving open-shutter efficiency even when the instrument is taking many short sub-exposures.

The STA5500 is an available fallback qCCD and the upcoming STA5900 qCCD is our preferred path. The STA5500 already has better performance than a conventional CCD, with a 4096 × 2048 image area (w/ matching frame store), and 1 e⁻ read noise in roughly 13 s. STA5900 is the target faster device, with 1 e⁻ read noise in roughly 2.8 s, and charge quantization in about 120 s, with both bulk red-sensitive and epitaxial blue-optimized versions.

Tapered AR Coatings¶

The qCCD coating concept further improves ZShooter’s optical grasp. In a fixed-format spectrograph, each detector location corresponds to a predictable wavelength range. A spatially varying anti-reflection coating can therefore be tuned to the local incident wavelength rather than forcing one coating stack to serve the entire channel uniformly.

The practical goal is simple: keep reflectance below the percent level across the relevant detector format. Since silicon charge collection is effectively complete when the photon is absorbed in the active volume, lowering reflectance directly improves QE. It also suppresses ghosting, fringing, and broad-band scattered-light coupling from wavelengths that do not belong at a given detector location.

Visible qCCD coating concept¶
Channel	Coating gradient	Coating bandwidth	Stray-light suppression note
ZSpec Blue	Tapered ALD stack ~3.5 nm/mm with 365 nm midpoint	full channel span is 110 nm, instantaneous coating bandwidth TODO	Early estimates suggest ~ TBD% reduction in spectrally broad scattered light
ZSpec Green	Tapered ALD stack, ~6.5 nm/mm with 500 nm midpoint	full channel span is 200 nm, instantaneous coating bandwidth TODO	Early estimates suggest ~ TBD% reduction in spectrally broad scattered light
ZSpec Red	Tapered ALD stack, ~13 nm/mm with 780 nm midpoint	full channel span is 400 nm, instantaneous coating bandwidth TODO	Red performance also depends on bulk silicon thickness, fringing, and dark current

Early explorations use a graded graded Al₂O₃ layer bounded by HfO₂ and fabricated by atomic layer deposition. The appeal of ALD is process control: thickness can be refined cycle-by-cycle, the approach is compatible with detector-sized substrates, and the coating can be tuned to the cross-dispersed format. The remaining engineering issue is calibration and repeatability whenever the material set or process changes.

qCCD Science and Operations Implications¶

Capability	Detector reason	ZShooter observing consequence
Short-exposure coadds	Read noise no longer dominates each short exposure.	Better cosmic-ray rejection in thick devices; real-time SNR assessment; less penalty for splitting exposures around changing conditions.
High native resolution with later binning	Read noise remains small even before digital binning.	Observers can preserve line information, then bin around sky features or science windows in the DRP.
Digital sky suppression	High native sampling plus low read noise makes contaminated spectral pixels less expensive.	OH-affected bins can be down-weighted or removed before rebinning to science resolution.
Self-calibration	Charge quantization exposes integer-electron peaks in image histograms.	Conversion gain, zero signal, linearity, serial CIC, and some brighter-fatter diagnostics can be inferred from science or calibration images.
AO and image-slicer compatibility	Oversampling the PSF does not automatically impose a read-noise catastrophe.	Optical architectures that collect light into more pixels become more plausible.

CCD Fallback¶

The conventional CCD fallback preserves wavelength coverage and basic spectroscopic function. It does not fully preserve the same observing model. Internal performance slides summarize the expected loss from conventional CCDs as roughly 0.2–0.4 mag. We believe this is still an acceptable fallback because the instrument remains a broad-band, high-throughput optical–IR spectrograph that remains at least as performant as existing instruments.

LmAPDs¶

ZShooter’s infrared baseline is a set of linear-mode HgCdTe avalanche photodiode arrays. In linear mode, the detector is used as an integrating array, but photoelectrons are multiplied by avalanche gain before the readout noise of the ROIC is added. This is the central difference from ordinary HxRG-style operation: the signal is amplified before the dominant read noise source enters the measurement.

Leonardo describes Saphira-family HgCdTe APDs as variable-gain detectors for 0.8–2.5 µm operation, with flexible windowing and high-speed readout. The Mike Bottom and the University of Hawaiʻi and UC Berkeley have partnered with Leonardo on a development program that has pushed the technology toward larger, lower-glow, lower-dark-current astronomy arrays. Recent work on LmAPDs reports newer devices with that employ a modified pixel structure to reduce ROIC glow, with measured glow around 0.012 e⁻ pixel⁻¹ frame⁻¹ and very low intrinsic dark current in the relevant low-background tests.

For ZShooter, the current detector target is a 2k × 2k, 15 µm-pixel LmAPD. The project assumption is that one array can serve each IR channel: YJ, H, and K, a much cleaner path than the earlier conceptual design mosaic that required three-side-buttable LmAPDs.

ZSpec IR detector assumptions¶
Parameter	LmAPD baseline assumption	HxRG fallback	Design consequence
Format	2k × 2k, 15 µm array	H2RG/H4RG-family detector options	Updated YJ/H/K split avoids the detector mosaic required by the older HK concept.
Read noise	Effective <1 e⁻ RMS in a 60 s sampling-up-the-ramp assumption	Several e⁻ for long SUTR; higher for short CDS	LmAPD gain matters most between OH lines and for synthesized lower-resolution spectra.
QE	Effective QE ~80% across Y/J/H/K after excess-noise-factor penalty	Conventional HgCdTe can have ~95% nominal QE and no avalanche excess-noise penalty	LmAPDs win when read noise dominates; HxRGs remain competitive when shot noise dominates.
Dark current	Project target <3.6 e⁻ pixel⁻¹ h⁻¹, excluding glow	Mature HxRG performance is known and around 20 e⁻ pixel⁻¹ h⁻¹	K-band and long integrations require careful dark/glow/persistence control.
Known risks	2k × 2k ROIC schedule, glow, persistence, cosmetic uniformity, pixel-to-pixel QE structure	Mature procurement and calibration path	LmAPDs are the higher-payoff, higher-maturity-risk path; HxRGs are the conservative fallback.

APD Excess Noise and Effective QE¶

Avalanche gain is not perfectly deterministic. The distribution of multiplication gains contributes an excess noise factor, which has nearly the same SNR effect as reducing QE. Project material uses an excess noise factor of about 1.2, so an APD with high physical absorption can behave like a detector with a lower effective QE. This should not be hidden: the case for LmAPDs is not that every detector property is superior to HxRGs. The case is that the read-noise reduction is large enough to dominate the SNR trade over much of ZShooter’s faint-source IR parameter space.

This is why K band deserves separate treatment. In K, telescope emissivity, sky background, thermal background, and longer integrations can move observations toward a shot-noise-limited regime, where the LmAPD advantage is reduced and the QE/excess-noise trade becomes more visible.

HxRG Fallback¶

H2RG/H4RG-family detectors afford a straightforward fallback for the infrared channels. They offer mature packaging, established calibration behavior and high QE. They also may reduce some risk associated with persistence or pixel-to-pixel avalanche variations.

The cost is read noise and dark current. Sensitivity estimates performed during the conceptual design phase quantify the sensitivity loss as roughly 0.4 mag for H2RG-style detectors, but the real operational loss is larger in the exact regimes ZShooter is trying to open: faint, short, or highly binned observations between OH lines. In those cases, sky between OH lines may be low enough that a conventional IR array may require very long integrations before shot noise overtakes read noise.

ZImager qCMOS¶

ZImager is a simultaneous tri-band science imager intended to support target verification, spectrophotometric anchoring, rapid fallback photometry when spectra are too faint, and high-cadence timing science. That places different demands on its detector than ZSpec: the priority is low read noise at high frame rate, stable photometry, compact packaging, and manageable data flow.

The current baseline is ORCA-Quest 2-class qCMOS device. Hamamatsu describes the ORCA-Quest 2 as a photon-number-resolving qCMOS camera with 0.30 e⁻ RMS read noise in ultra-quiet scan mode, a 4096 × 2304 format with 4.6 µm pixels, and improved UV QE relative to the first ORCA-Quest generation.

Use case	Detector requirement	Implication for ZImager
Deep acquisition and field verification	Low read noise and high QE in u/variable/z bands	Faint ToO fields can be verified without long acquisition overheads.
Spectrophotometric anchoring	Stable relative photometry and standard filter support	ZImager can anchor ZSpec spectra with pre/post imaging when spectroscopy is not enough.
Millisecond/subsecond timing	Fast subrasters with low read noise	Compact binaries, occultations, flickering sources, and rapid optical counterparts become real ZImager programs, not just acquisition side-products.
Data handling	High frame rates at megapixel scale	Requires early DRP/data-volume planning; high-cadence modes must be treated as a first-class data-system requirement.

Performance Implications¶

The detector choices interact with the optical design rather than sitting after it. The current performance story is therefore best written as a set of detector-dependent consequences, not as a single sensitivity number.

Choice	Primary gain	Main fallback penalty	Science cases most affected
qCCD vs conventional CCD	Sub-electron read noise, charge quantization, short-exposure coadds, high-R then bin	Roughly 0.2–0.4 mag loss plus weaker cadence/binning flexibility	Young SNe, kilonovae, binaries, primordial D/H, polluted WDs, faint nebular work
LmAPD vs HxRG	Effective <1 e⁻ read noise in IR with avalanche gain before readout noise	Roughly 0.4 mag loss and stronger read-noise penalty between OH lines	Kilonovae, GRBs, high-z SNe, TDEs, H/K diagnostics, low-resolution synthesis
AO + low-noise detectors	Smaller sky aperture and higher spectral resolution without equivalent slit-loss penalty	AO gains are reduced if detector noise dominates after sky suppression	Primordial D/H, lensed systems, faint compact sources, high-R stellar/WD work

Risk Posture¶

The optical detector risk is low because the qCCD path has a staged fallback. Earlier skipper CCDs already established the underlying charge-quantization principle; the ZShooter-relevant question is whether the differential, multi-output, frame-transfer versions arrive at the desired speed/noise point on the project schedule. If STA5900 is late or underperforms, STA5500-class devices and conventional CCDs both preserve a credible path.

The IR detector risk is more substantive. The desired LmAPD arrays are aligned with active HWO-class detector development and are exactly the kind of technology ZShooter should exploit if the schedule and procurement path close. But the 2k × 2k ground-based astronomy implementation is still maturing. Persistence, glow, pixel-to-pixel QE variation, packaging gaps, and procurement/NRE structure remain live issues. The HxRG fallback protects the instrument from a single-point detector failure, but not from loss of the most distinctive IR read-noise capability.