1.1.

Sensors and Electronics

Key requirements for the LSST sensors were formulated early in the project, and a multiyear prototyping program was carried out with several suppliers. The production devices are of two types, the CCD-250 made by Teledyne-e2v (henceforth E2V) and the STA3800C, designed by Semiconductor Technology Associates, wafers fabricated at Teledyne-DALSA, and devices postprocessed, packaged, and tested at the Imaging Technology Laboratory (ITL) of the University of Arizona. Both devices share a common $4\mathrm{k}\times 4\mathrm{k}$ pixel format, $10\text{-}\mu {\mathrm{m}}^2\text{pixels}$, with 16 independently read out amplifier segments of $512\times 2\mathrm{k}\text{pixels}$. Devices are back-illuminated and fabricated on high-resistivity p-type silicon thinned to $100\mu \mathrm{m}$, which is sufficient to meet the y band throughput requirement without introducing enough charge diffusion to impact the point spread function.^8,9 Devices have four-side buttable packages and achieve $>90\%$ fill factor including nonimaging silicon area and interchip gaps. The two device types are 100% interchangeable in their mechanical and electrical interfaces to the RTM but differ in their construction (see Table 2).

Table 2

Construction differences between sensor types.

Both devices are treated with proprietary coatings on the entrance side and the substrate-facing side, leading to differences in quantum efficiency (QE) at various wavelengths. Finally, ITL device outputs are buffered by JFET source followers mounted on the flex cables that interface the sensor to the RTM electronics boards. These were found to be necessary to provide sufficiently fast video rise and fall times to meet the 2-s frame readout requirement.

To accommodate the high number of video channels in the LSST focal plane, each RTM incorporates a compact, ASIC-based control, and readout electronics system¹⁰ on three PC boards occupying the $\sim 4\text{-}\mathrm{L}$ volume in the shadow of the CCD subarray. The RTM electronics includes 144 channels of video processing (amplification, dual-slope integration filtering, 18-bit digitization, data multiplexing, and serial output link), CCD bias, timing, and control signal generation, thermal control of the CCD array, power conditioning, and monitoring and readback of several hundred temperatures, voltages, and currents. A strict power budget of $<50\mathrm{W}$ (average) is necessary to match the heat removal capacity of the cryostat refrigeration system.

1.2.

Electro-Optic Performance Requirements and Production Test Methods

The performance requirements for the RTM are summarized in Table 3.

Table 3

RTM electro-optic performance requirements and measured median performance for the 189 CCDs, 3024 video channels on 21 rafts

Since each RTM is able to function as standalone camera, the RTM electro-optic test stand has been built to simulate, as closely as possible, the conditions that will be experienced in the final LSST focal plane: early versions of the camera control¹¹ and data acquisition¹² software are used to execute exposure sequences, substantial use is made of the LSST data management software stack¹³ for image analysis, and prototype LSST power supplies are used. RTMs are housed in a vacuum cryostat with cold plates held at $-130°\mathrm{C}$ and $-60°\mathrm{C}$ to remove heat from the sensor array and electronics, respectively. (The LSST camera will hold the electronics cold plate at $-40°\mathrm{C}$, but our commercial closed-cycle cryocooler does not have sufficient cooling capacity at the higher temperature.)

The EO test methods follow conventional procedures for CCD characterization such as those described in Ref. 14. CCD clock and bias voltages and timing sequences were set to the manufacturer’s suggested values. In early raft testing, some variations around the standard settings were explored; additional work to optimize performance¹⁵ is underway but is a separate activity from the production tests reported here. Key measurements that are made are: (1) ${}^{55}\mathrm{Fe}$ exposures, providing measurements of gain (by fitting the $\mathrm{Mn}\text{-}{\mathrm{K}}_{\alpha }$ and ${\mathrm{K}}_{\beta }$ lines for reconstructed clusters), noise (from overscan pixels), and charge diffusion (from cluster sizes); (2) dark exposures of 500 s; (3) monochromatic flatfields from 350 to 1100 nm for QE; (4) monochromatic flatfield pairs at increasing exposure times for linearity, full-well capacity, and photon transfer curve; (5) superflat exposures, coadded to produce low-noise files suitable for charge transfer efficiency measurements using the extended pixel edge response method;¹⁴ and (6) 12- to 20-h runs with continuous ${}^{55}\mathrm{Fe}$ exposures, to estimate response stability. Note that according to our test specification, full well is defined as the maximum output signal level, not the signal level in which the charge transfer breaks down. Also in keeping with general practice, the “gain” of a channel is expressed in e-/ADU (actually an inverse gain). For most rafts, EO runs were carried out at two CCD temperatures ($-90°\mathrm{C}$ and $-100°\mathrm{C}$); results given in subsequent sections are for $-90°\mathrm{C}$ temperature.

The tests described here have been developed to verify, using fully automated acquisition and analysis, that the LSST science rafts satisfy the performance criteria in Table 3. Further studies of subtle characteristics (persistence, crosstalk, distortions due to static and dynamic electrostatic effects, etc.) have been carried out on single CCDs and are reported in Refs. 1617.18.–19.

2.1.

Quantum Efficiency

QE measurement uses conventional methods²⁰ and is referenced to a NIST-calibrated photodiode and corrected for flatfield irradiance falloff across the raft surface. A single number is reported for each CCD by averaging the response after correcting the individual segments’ gain and offset. Spatial nonuniformity of QE across the CCD surface is typically at the 1% level in midband, up to 3% RMS below 450 nm and above 950 nm due to window processing and fringing, respectively. Figure 2 shows the QE curves for all CCDs, separated by supplier type. Absolute QE numbers have uncertainties of 2% to 5%, whereas relative variations include instrumental drifts over the 28-month production period. In general, there is higher QE at low and high wavelengths for E2V and ITL sensors, respectively.

Fig. 2

QE versus wavelength for E2V (green) and ITL (orange) sensors (72 sensors of each type). Horizontal (magenta, green, red, yellow, brown, and gray) lines indicate the passbands and band-averaged QE requirements for (u, g, r, i, z, and y) filters.

2.2.

Noise, Full Well, Dark Current, Charge Diffusion, Charge Transfer Inefficiency

Figure 3 shows the distribution of parameters for the 189 CCDs (3024 channels) of the focal plane, separated by CCD supplier. Population statistics are summarized in Table 4. Figure 4 shows RTM-by-RTM parameter distributions.

Fig. 3

Histograms of electro-optic parameters: E2V (green) and ITL (orange). Dashed lines indicate hard (red) and soft (gray) specification limits from Table 3. Note: charge diffusion PSF measurement uses coarsely quantized bins.

Table 4

Population statistics and temperature coefficients for all LSST science rafts.

Fig. 4

RTM-by-RTM electro-optic parameters for E2V (green) and ITL (orange) rafts.

In contrast to the QE results, the electro-optic parameters show a greater difference between CCD types, with nearly nonoverlapping distributions in some cases. Furthermore, the ITL sensors show greater variability in parameters. Figure 5 shows that the read noise and full-well parameter differences are linked through the different gains of the two CCD types. Gain, read noise, and full well are all governed by the sense node capacitance,¹⁴ which apparently differs by roughly 35% between the two suppliers. The electronic gain can be controlled at the CCD level and will be adjusted during operation to ensure that the maximum CCD signal does not saturate the ADC, as this is necessary for accurate crosstalk correction.

Fig. 5

(a) Correlation between gain and (a) read noise, (b) full-well capacity.

2.3.

Correlated Noise and Effect of Reduced Readout Rate

In order to read out the CCDs in 2 s, their serial registers are clocked at a relatively rapid rate of 545 kHz, leaving little timing margin in the readout sequence to isolate CDS integration intervals from nearby clock transitions. We made two observations that suggested that clock-coupled noise was present in some devices. First, clock feedthrough was seen to be pronounced in video waveforms of the noisier devices. Second, measurements were made, which showed significant correlation between the noise waveforms of the 16 channels of individual CCDs. (See Fig. 6. Since we record the waveforms of all 144 raft video channels, we can calculate their pairwise correlation coefficients while readout out a bias image.) Comparison of the degree of intra-CCD correlation and read noise (Fig. 7) confirmed the connection.

Fig. 6

Noise correlation matrices for 20 RTMs, at 2-s frame readout time. In these plots, the correlation coefficient between all $144\times 144$ channel pairs is shown. The $16\times 16$ block diagonals show the intra-CCD coefficients; the larger $48\times 48$ block diagonals represent the correlations within a single electronics board. Values along the main diagonal = 1 (poorly resolved at this plot scale). Note different scales for E2V and ITL rafts.

Fig. 7

Read noise ($x$ axis) and intra-CCD correlation for 13 RTMs at 2- and 3-s frame readout times.

We also observed that read noise could be reduced by adding delays between the clock edges and the integration intervals while keeping the integration times constant. For RTM-10, we varied the timing to give frame readout times of 2, 3, 4, and 5 s; both the noise and the dispersion in noise decreased [Fig. 8(a)]. Thirteen rafts were then measured at both 2- and 3-s readout times. An improvement of 10% to 20% was found for the slower readout, with the noisiest channels on ITL devices benefiting the most [Fig. 8(b)].

Fig. 8

(a) RTM-10 read noise versus frame readout time. Numbers in parentheses are percent of channels with noise exceeding 9 ${\mathrm{e}}^{-}$. Error bars show the channel-to-channel spread in read noise within the raft. (b) Noise results for 13 rafts at 2- and 3-s readout time.