NIRSpec Detector Performance
The JWST NIRSpec detectors, NRS1 and NRS2, have been tested to characterize read noise, detector gain, dark current, non-linearity, saturation, quantum efficiency, persistence, and inter-pixel capacitance.
The NIRSpec detectors (sensor chip assemblies; SCAs) have been optimized to balance a variety of performance metrics such as read noise, dark current, and saturation well depth.
Table 1 presents characteristics of the two NIRSpec detectors, called NRS1 and NRS2, in the data processing and pipeline systems. The presented SCA values are relevant at the operating temperature of 42.8 K. Each NIRSpec SCA pixel corresponds to approximately 0.1" × 0.1" on the sky in imaging mode. Operable pixels are defined as having total noise less than 12 e¯ and a quantum efficiency (QE) >70% of the average QE. Non-operable pixels include hot pixels (i.e., pixels with a high dark current), dead pixels, and pixels with low QE or other defects. Each NIRSpec detector has an overall requirement of less than 6 e¯ total noise in a 1,000 s MULTIACCUM exposure. This requirement is met using the improved reference sampling and subtraction (IRS2) detector readout mode. Presented noise and dark current values were estimated using ground test data in 1,000 s MULTIACCUM "up the ramp" exposures acquired using the traditional and IRS2 readout modes.
The following information was measured during ground testing at the Detector Characterization Lab at NASA Goddard Space Flight Center, or in a flight-like configuration during instrument characterization tests in the December 2015 to February 2016 timeframe. Verification of all detector performance parameters will be carried out during on-orbit commissioning, therefore, some of these values are subject to change.
Table 1. NIRSpec NRS1 and NRS2 detector performance summaries
|Parameter||Requirement||Measured NRS1||Measured NRS2|
|Array size||two SCAs, each 2048 × 2048 pixels|
|Pixel size||Physical size: 18 × 18 µm, On the sky: 0.1" × 0.1"|
|Wavelength range||0.6–5.0µm||0.6–5.3 µm (estimated long wavelength cut-off)|
|Average quantum efficiency, 0.6–1.0 µm||>70%||79.5%||80.4%|
|Average quantum efficiency, 1.0–5.0 µm||>80%||88.0%||88.3%|
|Total noise, including electronics, using traditional readout (e¯)||<6||5.55||6.46|
|Total noise using IRS2 readout (e¯)||n/a||5.17||6.60|
|Dark current (e¯/s/pix)||<0.01||0.0092||0.0057|
|Average full well capacity (e¯)||>60,000||55,100 1||60,400 1|
|Operational temperature||42.8 K||42.8 K|
1 The detector average full well capacity (in e¯) is the saturation well depth minus the detector bias level. NRS1 does not meet the well depth requirement in full frame mode. This is a consequence of the chosen conversion gain, which is optimized for low noise. In subarray mode the conversion gain (and noise) is higher, and the well depth requirement is met.
The detector gain is the ratio of electrons to detector count (typically expressed as data numbers, or DN). This number, as a function of detector pixel, is used primarily in the data processing pipeline as part of the calculation of the shot noise, or Poisson noise, contribution to the total noise per pixel in an integration. Pixel-to-pixel gain maps for the two SCAs were constructed from data obtained during the detector characterization ground test campaign, using the classical photon transfer curve technique (Janesick, J. et al.,1987). The average gain values for both SCAs in full frame readout is near unity (Table 1). The gain map for NRS1 (Figure 1, Sirianni 2017) exhibits a very uniform distribution, with a small area of slightly lower values at the lower center caused by a known "epoxy void" effect. The map for NRS2 (Figure 1) is also uniform, though with a slight gradient at the ~5% level.
There is a different gain setting for subarray exposures which will provide an improved dynamic range for bright sources. The conversion gain is about a factor of 1.43 higher compared to full frame data for both SCAs. This was confirmed during OTIS testing.
All infrared detectors exhibit some level of dark current. By virtue of the relatively low operating temperatures, NIRSpec's detectors show extremely low dark current values. The average pixel dark current performance for both the NRS1 and NRS2 NIRSpec detectors are well within the performance requirement of <0.01 e¯/s. Figure 2 presents the dark current images in traditional readout for the NIRSpec detectors (from Birkmann 2016). A small region on detector NRS1 at the center bottom has an "epoxy void" that is apparent as a region of lower dark current (also seen in Figure 1).
All of JWST's near-infrared detectors exhibit non-linear response, as illustrated below. The thin straight line represents linear flux, the curves present the deviation from this linearity for detector operation at 3 different temperatures. Correction for this detector non-linearity performance is taken into account in the data reduction pipeline.
Charge cannot accumulate indefinitely in a single integration; there is a maximum level beyond which the signal saturates or can no longer be reliably corrected for non-linearity. Given the gain values discussed above, the saturation level for most pixels in the NIRSpec SCAs is set by the limit of the analog-to-digital converter: 216−1, or 65,535 DN. Counts at this level must be flagged and rejected by the data processing pipeline in order to derive the correct count rate (a minimum of two unsaturated groups per integration are necessary to recover the count rate). In practice, the saturation thresholds applied by the pipeline are actually below this level, and vary from pixel to pixel, in order to reject data in which the nonlinearity is too large to be corrected accurately. The average saturation level is 61,000–64,000 DN. The average detector well depth is the saturation minus the detector bias level: 55,100 and 60,400 electrons for the NRS1 and NRS2 detectors, respectively (Birkmann 2016).
Unlike with a CCD, there is no "bleeding". However, charge diffusion can impact neighboring pixels. This is also known as the "brighter-fatter" effect, see e.g. Plazas et al. 2018. Also, a highly saturated region could result in persistence that affects subsequent exposures at the same detector position. For example, in the case of solar system observations, dither patterns and target orientation should be planned such that a faint satellite does not fall onto a region that was illuminated by the planet in a previous exposure.
Detector quantum efficiency (QE)
The detector quantum efficiency (QE) is a measure of how efficient a detector is at recording incoming flux signal. NIRSpec detectors were extensively measured for QE at multiple wavelengths during detector system tests carried out in the Detector Characterization Lab (DCL) at NASA Goddard Space Flight Center. The illustration below shows relative QE for NIRSpec detector NRS2 at three operating temperatures. The gray region represents uncertainty in the measurements, and the orange line presents the performance requirements on the NIRSpec detectors. The NIRSpec NRS1 detector shows a similar QE pattern and amplitude.
JWST's near-infrared detectors can retain vestiges of earlier exposures; this effect is called persistence. The most vulnerable situation is when a very faint object is observed not long after something very bright. At the present time, the JWST scheduling system does not give an observer knowledge of what may occur in a preceding exposure, in part because that is not always predictable. Figure 5 shows how a persistence signal drops by a factor of ~200 after one hour. The initial persistence level is about 1 e¯/s/pix, and, therefore, it is anticipated that the worst effects will be seen from self-persistence within a program.
Detector inter-pixel capacitance (IPC)
Inter-pixel capacitance (IPC) is a type of capacitive coupling between neighboring pixels (see Moore et al. 2006). This phenomenon is commonly confused with charge diffusion, which involves migration of charge carriers to adjacent pixel cells which also happens at high signal levels (at or beyond full well). Instead, IPC arises from the interaction of electric fields that alter the measured voltages.
In its simplest form, this effect can be parameterized as a convolution kernel described as a 3 × 3 matrix with values totaling unity. Measurements for NIRSpec detectors show values consistent with an assumption of symmetric coupling: the kernel will have values of nearly zero in the corners, some value α in the left, right, top, and bottom positions, and the remaining 1−4α in the central pixel (α is a coupling coefficient, see Moore et al. 2006). In the NIRSpec SCAs, α ranges from 0.005 to 0.0067. In other words, approximately 2.5%–3% of the total flux measured from a given pixel gets redistributed between its four adjacent neighbors.
Uncorrected IPC can affect the spatially measured Poisson noise, which will result in overestimates of the conversion gain and reported quantum efficiencies. Figure 6 shows the IPC kernels for the NRS1 and NRS2 detectors of NIRSpec (Sirianni 2016). This correction is currently not applied in the default data processing pipeline.
NIRSpec Bias and Dark Reference Files for Build 7
NIRSpec Saturation and Non-Linearity Correction Reference Files for Build 7
Quantum efficiency overestimation and deterministic cross talk resulting from interpixel capacitance