Near-Infrared Integral-Field Spectrograph (NIFS):
An Instrument Proposed for Gemini
Peter J. McGregor , Peter Conroy , Gabe Bloxham , Jan van Harmelen, PASA, 16 (3), 273.

Next Section: Timeline
Title/Abstract Page: Near-Infrared Integral-Field Spectrograph (NIFS):
Previous Section: The NIRI Legacy
Contents Page: Volume 16, Number 3

NIFS Design Concept

The NIFS optical design has been introduced in §2. We now describe each section of the instrument in greater detail.

Offner Relay

The Offner relay (Offner 1975) is a simple, high performance system for reimaging a narrow field at the same scale while forming a real pupil that can be used to baffle the system. A conventional spectrograph forms a pupil on the grating. However, it is difficult to use a grating pupil as a cold stop in a near-infrared spectrograph. The Offner relay provides a convenient solution because the IFU requires only a small field.

Integral Field Unit Specification

The first near-infrared IFU spectrograph was the 3D instrument (Weitzel et al. 1996), constructed by the Max Planck Institut für Extraterrestrische Physik and used recently on the Anglo-Australian Telescope. A limitation of this instrument is that each reformatted slitlet is reimaged to only one detector pixel. A mechanism for half-stepping the grating is then required to fully sample the spectrum. NIFS will avoid this complication by mapping each slitlet to two detector pixels in the spectral direction. The

2048 x 2048 pixel detector array then accommodates 32 slitlets, each 0.1'' (2 pixels) wide by 3.2'' (64 pixels) on the sky. The field of view of the IFU is

3.2'' x 3.2''. An effective slit width of 0.1'' was chosen in order to probe to spatial scales approaching the 0.07'' diffraction-limited image core size expected from the Gemini AO system at 2.2 $\mu$ m.

Integral Field Unit Philosophy

The IFU is the most critical NIFS component. The reflective ``staircase'' IFU approach is favored over optical fibres, for example, because it is proven technology requiring the least development investment. Optical fibre solutions are complicated by the need to feed fibres with fast beams. Reflective IFUs are intricate devices, and some explanation is required to demonstrate how the NIFS IFU functions, and why a superior design can be achieved by mating it to a dedicated spectrograph, rather than retro-fitting an IFU to an existing spectrograph.

The first ``staircase'' IFU was used in the 3D instrument (Fig. 5). This uses a 16 element reflective image slicer in which each of the flat mirrors is tilted at the appropriate angle to form the virtual slit. A 16 element segmented mirror then feeds light from each virtual slitlet into the spectrograph. The requirements that the virtual slit be flat and that the virtual pupil matches the telescope exit pupil required that the flat segments of the segmented mirror be positioned along a hyperbola. This simple design works well for small IFUs and small detectors. However, the long virtual slit possible with a

$2048\times2048$ detector array means that the distance from the slicer to the extremity of the segmented mirror becomes large and the f/16 beams from the Gemini telescope diverge to a size larger than each mirror segment. Specifically for NIFS, over the required distance of 32 mm the f/16 beams would diverge to a width of 2 mm. The 2 mm long slitlets would then fill 4 mm at the segmented mirror, overfilling their respective segmented mirror elements by a factor of two.

**Figure 5:** Schematic of the IFU approach used in 3D.
$\begin{figure} \centering\leavevmode \epsfysize=100mm \epsfbox{nifs_3d.ps}\end{figure}$

This problem has been addressed by the Astronomical Instrumentation Group at the Durham University (Content 1997; Fig. 6). They propose using reimaging optics to form separate pupils on each element of the segmented mirror. This means that in NIFS, for example, each 2 mm long slitlet could form an image only 2 mm in diameter on the segmented mirror with minimal overlap between segments. The Durham design then uses spherical elements in a pupil mirror array to form a real image of the f/16 slit, and a further array of field mirrors at the real slit image to relocate the system pupil on the grating.

**Figure 6:** Schematic of the IFU approach proposed by Content (1997).
$\begin{figure} \centering\leavevmode \epsfysize=100mm \epsfbox{nifs_content.ps}\end{figure}$

Implementing this design in GNIRS requires large off-axis angles between the optical components, which introduce significant aberrations. By designing the IFU as an integral part of the spectrograph, we can achieve a simpler optical design with better performance (Fig. 1).

NIFS IFU Implementation

Detail of the NIFS IFU is shown in Figure 7. The 11 mm x 11 mm imager slicer consists of 32 flat mirrors each 0.33 mm thick. The displacement angles on the image slicer elements are small so that defocus of the f/90 image plane is not significant relative to the diffraction-limited image size (Fig. 8). The image slicer will be manufactured by polishing individual glass plates to the required angles, and gluing or clamping the stack of 32 mirrors in a metal mount at the required angles (Fig. 9). The 0.33 mm thickness of each slicer element is sufficient to hold an accurate optical polish. The assembled mirror unit will then be gold coated.

**Figure 7:** Ray trace of NIFS IFU. The f/90 image slicer is at left. The f/90 pupil is imaged onto the pupil mirror array which reimages the reformatted focal plane at f/16 onto the field mirror array. The field mirror array then feeds telecentric beams into the spectrograph. The field lens and collimator fold mirror of the spectrograph are shown at right.
$\begin{figure} \centering\leavevmode \epsfxsize=\textwidth \epsfbox[20 170 560 305]{nifs_ifu.ps}\end{figure}$

**Figure 8:** Diffraction profiles at J1 (1.05 $\mu$ m), J2 (1.25 $\mu$ m), H (1.65 $\mu$ m), and K (2.20 $\mu$ m) for an object centered in a 0.1'' wide slitlet. The slitlet width is indicated by dashed lines.
$\begin{figure} \centering\leavevmode \epsfxsize=\textwidth \epsfbox[75 170 580 400]{nifs_airy.ps}\end{figure}$

**Figure 9:** Image slicer required for the NIFS IFU. The slicer segments have widths of 330 $\mu$ m forming a
10.9 x 10.9 mm surface.
$\begin{figure} \centering\leavevmode \epsfysize=80mm \epsfbox{nifs_slicer.ps}\end{figure}$

The distance between the image slicer and the pupil mirrors is 120 mm so that the 1.3 mm diameter f/90 pupils are contained within the 1.6 mm width of each pupil mirror. The extreme f/90 pupil images are not significantly out of focus so there is no overlap with adjacent pupil images (Fig. 10). The pupil mirrors have a focal length of 18.3 mm so they reimage each slitlet with a focal ratio of f/16 at a distance of 21.6 mm. A flat row of spherical field mirrors is located at this focus. These mirrors have a focal length of 21.6 mm to feed telecentric beams into the spectrograph. The vertex of each sphere is at the same lateral displacement as the matching pupil mirror, so the field mirrors must be decentered by increasing amounts.

**Figure 10:** Illumination pattern on the pupil mirrors of the NIFS IFU. All pupil images have a diameter of 1.3 mm, less than the mirror spacing of 1.6 mm.
$\begin{figure} \centering\leavevmode \epsfxsize=\textwidth \epsfbox[20 175 560 285]{nifs_pupil.ps}\end{figure}$

The pupil mirror and field mirror arrays will either be manufactured as micro-mirror arrays, or assembled from individual polished glass mirrors and glued or clamped in place. Micro-mirror arrays will be easier to align, but we are concerned that scattered light from the machined optical surfaces will be problematic, especially for the rejection of terrestrial OH emission. Individual glass mirrors can be polished with greater accuracy as a larger unit and sliced to the required dimensions (Figs. 11 & 12), but these are expected to be more difficult to align. A prototype polished glass mirror IFU assembly is shown in Fig. 13.

**Figure 11:** The array of centered pupil mirrors may be constructed by slicing larger glass mirrors.
$\begin{figure} \centering\leavevmode \epsfysize=100mm \epsfbox{nifs_pupil_array.ps}\end{figure}$

**Figure 12:** The array of decentered field mirrors may be constructed by slicing decentered sections from larger glass mirrors.
$\begin{figure} \centering\leavevmode \epsfysize=100mm \epsfbox{nifs_field_array.ps}\end{figure}$

**Figure 13:** Prototype polished glass mirror IFU assembly showing image slicer at left and pupil and field mirror arrays at right.
$\begin{figure} \centering\leavevmode \epsfxsize=\textwidth \epsfbox{nifs_ifu_assembly.ps}\end{figure}$

Diffraction

The image diffraction full width at half maximum at 2.2 $\mu$ m on Gemini is $\sim$ 0.07''. Consequently, a spectrograph with a 0.10'' slit is essentially diffraction limited in the K band. Slit diffraction causes divergence in the spectral direction of rays through the slit in excess of the geometrical focal ratio; it can be thought of as focal ratio degradation speeding the beam in the spectral direction. This spreads the beam in the IFU in the direction perpendicular to the slicing direction and can be accommodated by increasing the size of each element of the pupil mirror array in this direction. Since the NIFS spectrograph already accommodates a 64 mm long reformatted slit, the NIFS collimator optics are quite capable of accepting this diffracted light. Rays outside the geometrical focal ratio then strike the grating outside the 30 mm diameter geometrical pupil. Passing this diffracted light to the camera requires using a wider grating than would otherwise be the case.

Calculation of the diffraction angle at the slit is complicated because the image slicer is at a focus. A perfect image would have no phase coherence across the slit. In practice, there will be significant phase coherence across the slit due to a combination of the telescope diffraction and seeing effects uncorrected by the AO system. If we assume full phase coherence, diffraction at 2.2 $\mu$ m through a 330 $\mu$ m wide slit produces a first minimum at an angle of $\sim$ 0.38 $^\circ$ to the optical axis. This corresponds to a vertical displacement on the pupil mirror array of $\pm$ 0.8 mm which is to be compared to the pupil size of $\pm$ 0.67 mm. This corresponds to a motion of $\pm$ 18 mm in the spectral direction on the grating which is to be compared to the geometrical grating pupil size of $\pm$ 15 mm. The decision of how much of this diffracted light to accept will be based on design trade-offs such the ability to accommodate larger optics and their cost.

Grating Choices

The primary science drivers for NIFS require a two-pixel spectral resolving power of $\sim$ 5000 to detect black holes in galactic nuclei and study the dynamical evolution of disk galaxies at high redshift where galactic masses may be lower than for present-day galaxies, and hence the rotational velocities encountered may be lower. A two-pixel resolving power of $\sim$ 5000 is also sufficiently high to resolve much of the terrestrial OH airglow emission which plagues low resolution near-infrared spectroscopy. Greater sensitivity can be achieved by masking the OH emission-lines recorded in high resolution spectra in software. With a 2048 pixel detector, nearly complete spectra of the H, and K photometric bands can be recorded with a two-pixel spectral resolving power of $\sim$ 5340. Two 2048 pixel regions are required to record all of the J photometric band. Thus all regions of the 0.94-2.42 $\mu$ m wavelength range within standard photometric passbands can be recorded using just four fixed angle gratings.

An Ebert angle (i.e., camera-collimator angle) of 30 $^\circ$ is required by the NIFS mechanical configuration. The off-the-shelf gratings listed in Table 2 then produce the required central wavelengths in first order with grating angles of $\sim$ 20 $^\circ$ . A camera focal length of 290 mm images the full H and K photometric bands onto the detector with a two-pixel resolving power of 5340 ( $\Delta v = 56$ km s^-1). The 18 $\mu$ m detector pixels map to 0.05'' on the sky with a collimator focal length of 480 mm.

**Table 2:** NIFS Grating Parameters
Grating	$\lambda_c$	n	$\theta_{blaze}$	$\theta$	R	$\Delta v$	Range
	( $\mu$ m)	(l/mm)	(deg)	(deg)		(km s^-1)	( $\mu$ m)
J1	1.05	600	17.5	19.0	5090	59.0	0.94-1.16
J2	1.25	600	17.5	22.8	6100	49.2	1.14-1.35
H	1.65	400	13.9	20.0	5340	56.2	1.49-1.80
K	2.20	300	17.5	20.0	5340	56.2	2.00-2.42

The grating wheel will also contain a screen that blocks the spectrograph camera while dark and bias frames are exposed, and a mirror that images the ``staircase'' slit onto the detector undispersed. The latter will be used to assist in determining the correct detector focus.

Detector Choice

The primary science drivers for NIFS require maximised spatial coverage and spectral resolution, while making no specific requirement for wavelength coverage beyond 2.5 $\mu$ m. NIFS will therefore use the

2048 x 2048 18 $\mu$ m pixel HgCdTe HAWAII-2 array under development by Rockwell. These devices offer the largest available format and best noise performance of contenders in the 1-2.5 $\mu$ m range.

The Rockwell

$2048\times2048$ array will have been used with SDSU-2 controllers by the University of Hawaii before NIFS is commissioned. NIFS will also use this controller.

Optical Performance

The NIFS optical design has been progressed to a level where it gives ``two-pixel'' optical performance. No specific optimization of the IFU components has yet been performed. Spot diagrams on the detector at 1.50, 1.65, and 1.80 $\mu$ m in an H band spectrum are shown in Figure 14. This figure also shows images of the reformatted slit at these wavelengths. The slit curvature seen in Figure 14 is due to the ``staircase'' slit format combined with spectral line curvature induced by the grating.

**Figure 14:** Preliminary spot diagrams at wavelengths of 1.50, 1.65, and 1.80 $\mu$ m on the NIFS detector. The enclosing boxes are 36 $\mu$ m on a side corresponding to 2 x 2 detector pixels. The optical design is yet to be fully optimized. The frame at top right shows the locations on the f/90 image slicer mirror of each field position shown. The frame at top left shows images of the reformatted slit on the detector at the three wavelengths.
$\begin{figure} \centering\leavevmode \epsfxsize=\textwidth \epsfbox[115 20 465 430]{nifs_spots.ps}\end{figure}$

Efficiency

The NIFS system efficiency has been estimated assuming telescope mirror reflectivities of 95%, gold mirror reflectivities internal to the dewar of 98%, and single layer MgF₂ anti-reflection coatings on all lenses. A grating efficiency of 50% has been adopted, and the detector quantum efficiency is assumed to be 70% over the full 0.9-2.5 $\mu$ m wavelength range. With these assumptions, the NIFS system efficiency including the Gemini telescope optics but excluding the transmission of the AO system is $\sim$ 0.16 e^-/photon over this wavelength range. The detailed efficiency as a function of wavelength is shown in Figure 15 with and without anti-reflection coatings on the lenses.

**Figure 15:** Total NIFS plus Gemini efficiency as a function of wavelength with single layer MgF₂ anti-reflection coatings on all lenses ( *solid line*) and without anti-reflection coating ( *dashed line*).
$\begin{figure} \centering\leavevmode \epsfysize=90mm \epsfbox[20 150 530 510]{nifs_transmission.ps}\end{figure}$

Sensitivity

In calculating the NIFS sensitivity, we adopt a read noise of 9 e^-, a dark current of 0.015 e^- s^-1 pix^-1, and a maximum frame integration time of 1 hr.

Measurements of stellar velocity dispersion in galactic nuclei are limited by the sensitivity for detecting continuum sources. As noted in §3.1, the central surface brightness of typical black hole candidate galaxies in the K band is $\sim$ 13.7 mag arcsec^-2. This will produce $\sim$ 0.12 e^- s^-1 pixel^-1. The background photo-current at 2.3 $\mu$ m is predicted to be $\sim$ 0.1 e^- s^-1 pixel^-1 due predominantly to thermal emission from the telescope. In a 1 hr integration, the signal-to-noise ratio (SNR) achieved per spectral and spatial resolution element is $\sim$ 30. This is degraded to $\sim 20$ after sky subtraction. This SNR is more than adequate for stellar velocity dispersion measurements since the intrinsic width of the CO (2-0) bandhead is $\sim$ 500 km s^-1. A similar time will be needed for off-source sky measurement, making the total integration time per object $\sim$ 2 hr.

Atmospheric OH line emission is the dominant noise source in the J and H bands. Consequently, limiting observations will be restricted to regions between these strong OH lines. The dominant emission mechanisms in regions between strong OH lines are not well determined. Non-thermal sky continuum emission or a multitude of weak OH emission lines may define the limiting sky brightness in the J and H bands. However, it is likely that the limiting detected background will be set by OH line emission scattered within the dewar. If only 10% of the total OH emission entering the dewar is scattered across the detector, the background photo-current produced will be comparable to the expected detector dark current of $\sim$ 0.015 e/s/pixel. The strongest OH emission is in the H band. A model NIFS H band spectrum showing the expected OH emission-line photo-currents is presented in Figure 16. These are still sufficiently low that individual integrations will be limited by uncertain factors such as the cosmic ray event rate and on-chip amplifier glow rather than detector saturation.

**Figure 16:** Model H band background spectrum. The spectrum is dominated by terrestrial OH emission. The continuum between strong OH lines is uncertain, but will probably be dominated by scattered light.
$\begin{figure} \centering\leavevmode \epsfysize=90mm \epsfbox[20 150 530 510]{nifs_h_background.ps}\end{figure}$

For detecting H $\alpha$ at high redshift, we consider the limiting detectable emission-line flux at wavelengths avoiding the strong OH emission lines. The 1 $\sigma$ noise/pixel in a 1 hr integration is $\sim$ 13.8 e^-/pixel. A 5 $\sigma$ detection of a 50 km s^-1 wide spectral line requires the detection of $\sim$ 138 e^-, which will be achieved with a line flux of

$\sim 7.4 \times 10^{-26}$ W cm^-2 at 1.31 $\mu$ m, corresponding to the wavelength of H $\alpha$ at z = 1. The H $\alpha$ surface brightness through a 1 pixel long segment of the 2 pixel wide slit is then

$\sim 1.5 \times 10^{-23}$ W cm^-2 arcsec^-2. Sky measurements will be obtained from the object exposures for sufficiently small objects (typical half-light radius is $\sim$ 0.77''). The effect of sky subtraction on SNR can be offset by averaging two pixels along the slit. In 1 hr of on-source integration on a z = 1 galaxy with the average H $\alpha$ surface brightness of

2.5 x 10^-23 W cm^-2 arcsec^-2 predicted in §3.2, we would achieve a 7.5 $\sigma$ detection of H $\alpha$ per

0.1'' x 0.1'' spatial element. The integrated spectrum formed by averaging over the $\sim$ 186 spatial elements within a typical half-light radius would have a SNR on H $\alpha$ of $\sim$ 102 if the line was only 50 km s^-1 wide, and a SNR of $\sim$ 60 in each peak of a horned profile if approximately one-third of the H $\alpha$ emission was in the peak.