China Space Station Telescope (CSST)

• CSST is a 2 m-class off-axis telescope providing deep, wide-field imaging and slitless spectroscopy from the near-ultraviolet to near-infrared.
• Its integrated instrument suite—including SC, MCI, IFS, CPI-C, and TS—supports diverse applications from weak-lensing cosmology to exoplanet imaging.
• The mission employs multi-tiered surveys with wide, deep, and ultra-deep fields, enabling precision cosmology, transient detection, and stellar population studies.

The China Space Station Telescope (CSST), also known as Xuntian, is a next-generation Stage-IV sky survey telescope designed to deliver deep, wide-field, high-resolution imaging and slitless spectroscopy from the near-ultraviolet to the near-infrared. It is a 2 m-class, off-axis three-mirror-anastigmat facility that co-orbits and is dockable with China’s Space Station at approximately 400 km altitude, and it is equipped with five scientific instruments: the Multi-band Imaging and Slitless Spectroscopy Survey Camera (SC), the Multi-Channel Imager (MCI), the Integral Field Spectrograph (IFS), the Cool Planet Imaging Coronagraph (CPI-C), and the THz Spectrometer (TS). Across a ten-year mission, the observatory is intended to combine simultaneous multi-band photometric imaging and slitless spectroscopic surveys over 17 500 deg² with deep and ultra-deep programs, supporting applications that range from weak-lensing cosmology and large-scale structure to Galactic archaeology, strong lensing, transient discovery, stellar populations, and exoplanet imaging (Collaboration et al., 7 Jul 2025).

1. Mission architecture and instrument suite

CSST is built around a 2.0 m unobstructed aperture and a focal length of 28 m. Its optical train is an off-axis three-mirror anastigmat, and the primary imaging region meets $R_{\rm EE80}<0.15''$, with a Survey Camera plate scale of $0.074''/{\rm pixel}$ and a total detector count of 2.6 Gpix. The total field of view is reported as 1.72 deg², while the primary imaging zone is approximately $1.1^\circ\times1.0^\circ$; the additional area accounts for wavefront-sensing, FGS, and IR sub-fields (Collaboration et al., 7 Jul 2025). Other preparatory descriptions emphasize the SC survey footprint as a $\sim 1.1\,{\rm deg}^2$ focal plane for the main survey camera, which is the quantity typically used in survey forecasts and simulations (Miao et al., 2022).

Instrument	Principal stated parameters	Stated roles
SC	NUV, u, g, r, i, z, y; GU, GV, GI; $R\gtrsim200$; FoV $\approx1.1\,{\rm deg}^2$	Wide-field imaging and slitless spectroscopy
MCI	Three channels; $7.5'\times7.5'$ per channel; 30 filters; pixel scale $0.05''$	Photo-$z$ calibration, deep imaging, stellar-population studies
IFS	FoV $\ge 6''\times6''$; $0.074''/{\rm pixel}$0; 0.35–1.0 $0.074''/{\rm pixel}$1m	Spatially resolved spectroscopy
CPI-C	Contrast $0.074''/{\rm pixel}$2 in 600–900 nm; IWA $0.074''/{\rm pixel}$3 at 633 nm	Direct imaging of mature giant exoplanets
TS	0.41–0.51 THz; bandwidth $0.074''/{\rm pixel}$4 GHz; resolution $0.074''/{\rm pixel}$5 kHz	THz spectroscopy

The SC is the central survey instrument. It combines seven-band imaging over 255–1000 nm with slitless spectroscopy in three grating channels, GU, GV, and GI, over the same broad spectral interval (Collaboration et al., 7 Jul 2025). The MCI is a simultaneous three-channel imager covering 255–430 nm, 430–700 nm, and 700–1000 nm; in one description it carries 30 filters in total, including wide, medium, and narrow bands, and provides HST-like angular sampling over a substantially larger field (2206.12068). The IFS extends the observatory into resolved spectroscopy at $0.074''/{\rm pixel}$6, the CPI-C provides high-contrast visible and near-IR coronagraphy for reflected-light exoplanet studies, and the TS targets the 0.41–0.51 THz regime (Collaboration et al., 7 Jul 2025).

This instrumental composition is unusual among survey telescopes because it couples a wide-field cosmology platform to dedicated subsystems for deep photometric calibration, resolved spectroscopy, coronagraphy, and THz line work. A plausible implication is that the mission architecture was designed to internalize calibration and follow-up capabilities that are often distributed across multiple facilities.

2. Survey design, depth, and observing modes

The main SC survey is organized into a wide-field program of 17 500 deg², a deep-field program of 400 deg², and an ultra-deep field of 9 deg² in the first two years. In the wide-field mode, each field is exposed $0.074''/{\rm pixel}$7 s per visit, with imaging depths reported as $0.074''/{\rm pixel}$8 mag (point source, $0.074''/{\rm pixel}$9), $1.1^\circ\times1.0^\circ$0 mag, and spectroscopic depths of GU = 22 mag and GV/GI = 23 mag per resolution element. The deep field uses $1.1^\circ\times1.0^\circ$1 s exposures and reaches approximately 1 mag deeper; the ultra-deep field uses $1.1^\circ\times1.0^\circ$2 s and is expected to reach $1.1^\circ\times1.0^\circ$3 mag while providing an approximately 12-day cadence for Type Ia supernova light curves (Collaboration et al., 7 Jul 2025).

A complementary survey description for supernova forecasting gives a more explicit cadence model for the photometric survey. In that formulation, the wide survey covers $1.1^\circ\times1.0^\circ$4 at medium-to-high Galactic latitude with 150 s single exposures, two visits in $1.1^\circ\times1.0^\circ$5 and four visits in NUV and $1.1^\circ\times1.0^\circ$6, for 18 total photometric exposures spread over 10 yr; the deep survey covers $1.1^\circ\times1.0^\circ$7 with 250 s visits and 72 total exposures per field over the same period (Liu et al., 2024). Preparatory simulations for the main survey similarly adopt a seven-year survey time over a ten-year mission, a wide survey of 17,500 deg², and deep fields of $1.1^\circ\times1.0^\circ$8 deg² (Wei et al., 10 Nov 2025).

The photometric depth and sample density assumed in cosmological forecasts are correspondingly aggressive. One forecast adopts wide-survey $1.1^\circ\times1.0^\circ$9 point-source depths of 25.4 AB in NUV and $\sim 1.1\,{\rm deg}^2$0, 26.3 in $\sim 1.1\,{\rm deg}^2$1, 26.0 in $\sim 1.1\,{\rm deg}^2$2, 25.9 in $\sim 1.1\,{\rm deg}^2$3, 25.2 in $\sim 1.1\,{\rm deg}^2$4, and 24.4 in $\sim 1.1\,{\rm deg}^2$5, together with an effective photometric number density of $\sim 1.1\,{\rm deg}^2$6 and a spectroscopic sample of emission-line galaxies extending over $\sim 1.1\,{\rm deg}^2$7 (Miao et al., 2022). In another overview, the wide survey is explicitly linked to a photometric galaxy density $\sim 1.1\,{\rm deg}^2$8 and a spectroscopic density of $\sim 1.1\,{\rm deg}^2$9–$R\gtrsim200$0, while the deep-field survey is described as using exposures increased by a factor of $R\gtrsim200$1 per band (Gong et al., 25 Jan 2025).

These various descriptions are consistent in the broad survey logic even when they emphasize different operational details. The wide survey supplies statistical power, the deep program extends the depth and transient cadence, and the ultra-deep or extreme-deep tiers provide calibration and legacy fields. This tiered strategy suggests that CSST is intended to operate simultaneously as a precision cosmology mission and as a general observatory with stable, reusable survey products.

3. Slitless spectroscopy, radial velocities, and calibration

The SC slitless spectroscopic channel disperses light through three dichroic grisms rather than a traditional slit, producing low-resolution spectra across GU, GV, and GI. Several papers give band limits that differ slightly in the quoted edges, but all agree on broad coverage from approximately 255 nm to 1 $R\gtrsim200$2m with resolving power $R\gtrsim200$3 (Sun et al., 2020). In one detailed simulation framework, the slitless mode uses 24 distinct gratings covering five diffraction orders and adopts the trace and dispersion model

$R\gtrsim200$4

with coefficients expanded as polynomials in CCD position; the typical fitting error is $R\gtrsim200$5 pixel in working orders (Wei et al., 10 Nov 2025).

The stellar radial-velocity performance of this low-resolution system has been studied directly using simulated CSST spectra at $R\gtrsim200$6 derived from the Next Generation Spectral Library. Using 1,000 noise realizations per star at SNR = 100, the single-exposure uncertainty was found to scale roughly as $R\gtrsim200$7. When the full 255–1000 nm spectrum is used, the reported uncertainty is $R\gtrsim200$8–$R\gtrsim200$9 for A, F, G, K, and M stars, and 4–15 km s$\approx1.1\,{\rm deg}^2$0 for hotter OB stars; the study summarizes the expected precision as about $\approx1.1\,{\rm deg}^2$1 for AFGKM stars and about $\approx1.1\,{\rm deg}^2$2 for OB stars at SNR = 100 (Sun et al., 2020). The same work reports that velocity uncertainties depend strongly on effective temperature, weakly on metallicity for FGK stars, and hardly on surface gravity. At equal SNR, GU performs best for RV work, followed by GV and then GI, because the blue region contains a richer ensemble of Balmer lines, metal features, the Balmer jump, and, in cool stars, strong molecular bands (Sun et al., 2020).

Wavelength calibration is a central complication because slitless spectroscopy precludes calibration lamps. A star-based refinement method has therefore been proposed in which each slitless spectrum is divided into narrow segments, local RV residuals are measured against template spectra, and the resulting wavelength offsets are fitted with a low-order correction,

$\approx1.1\,{\rm deg}^2$3

and

$\approx1.1\,{\rm deg}^2$4

with the coefficient fields parameterized as 2D polynomials in focal-plane position. Using only a few hundred RV standard stars, the method is reported to achieve a wavelength calibration precision of a few $\approx1.1\,{\rm deg}^2$5 for the GU band, and about 10 to 20 $\approx1.1\,{\rm deg}^2$6 for the GV and GI bands (Yuan et al., 2020).

A common misconception is to treat “best spectroscopic band” as a single category. The CSST literature does not support that simplification. For stellar RV precision at fixed SNR, GU is the most informative band (Sun et al., 2020). By contrast, end-to-end slitless-image simulations find that GI exhibits the highest detection efficiency and the highest S/N at fixed source brightness, with first-order detection densities of 4.32, 1.99, and 0.82 arcmin$\approx1.1\,{\rm deg}^2$7 for GI, GV, and GU, respectively, in a 150 s exposure, and that nearly all first-order spectra are accompanied by corresponding zeroth-order images (Zhang et al., 10 Nov 2025). These statements concern different performance metrics and are therefore compatible rather than contradictory.

For extragalactic work, the CSST Emulator for Slitless Spectroscopy (CESS) provides a fast 1D-spectrum emulator that incorporates instrumental throughput, morphology-driven self-broadening, and overlap contamination. Applied to $\approx1.1\,{\rm deg}^2$8 model galaxies with $\approx1.1\,{\rm deg}^2$9 and $7.5'\times7.5'$0, it predicts that the CSST slitless spectroscopic survey can provide secure redshifts for about one-quarter of the sample galaxies, with a total secure-$7.5'\times7.5'$1 fraction of $7.5'\times7.5'$2 under the paper’s definition (Wen et al., 2024). That forecast is explicitly conditioned on low-resolution slitless data, morphology-dependent broadening, and contamination modeling; it is not a general statement about all CSST spectroscopic use cases.

4. Photometric system, photo-$7.5'\times7.5'$3 calibration, and precision cosmology

CSST’s imaging program is centered on a seven-band broad-band system extending from the near-UV to the near-IR. One mission-level summary gives the broad-band ranges as NUV (252–321 nm), $7.5'\times7.5'$4 (321–401 nm), $7.5'\times7.5'$5 (401–547 nm), $7.5'\times7.5'$6 (547–692 nm), $7.5'\times7.5'$7 (692–842 nm), $7.5'\times7.5'$8 (842–1080 nm), and $7.5'\times7.5'$9 (927–1100 nm), with PSF EE80 $0.05''$0 in NUV–$0.05''$1 and rising to $0.05''$2 in $0.05''$3 (Gong et al., 25 Jan 2025). The MCI provides a complementary three-channel imager with medium-band filters for calibration-intensive applications, especially photometric redshifts (Cao et al., 2021).

A dedicated study of MCI photo-$0.05''$4 calibration used mock data based on the COSMOS08/COSMOS15 catalogs, 31 galaxy SED templates, instrumental and astrophysical noise, and a modified LePhare fitting procedure that includes upper limits for low-SNR bands. It reported that the current nine-medium-band MCI filter design can achieve photo-$0.05''$5 accuracy $0.05''$6 and outlier fraction $0.05''$7, while combining SC and MCI yields $0.05''$8 and $0.05''$9 (Cao et al., 2021). The same work states that these values meet or exceed the baseline requirement $z$0 and outlier rate $z$1 for weak-lensing cosmology (Cao et al., 2021).

The cosmological forecasting literature treats CSST as a multi-probe facility combining cosmic shear, galaxy–galaxy lensing, photometric galaxy clustering, spectroscopic galaxy clustering, and cluster number counts. One forecast formulates the standard tomographic power spectra $z$2, $z$3, and $z$4, a redshift-space galaxy power spectrum including RSD and Alcock–Paczynski effects, and a Fisher matrix over $z$5 plus nuisance parameters for intrinsic alignment, shear calibration, photo-$z$6 bias and scatter, galaxy bias, and instrumental noise floors (Miao et al., 2022). In that analysis, the joint combination of 3$z$72pt photometry, spectroscopic galaxy clustering, and cluster counts yields 1$z$8 uncertainties $z$9, $\ge 6''\times6''$0, $\ge 6''\times6''$1, and $\ge 6''\times6''$2, corresponding to approximately 1%, 0.2%, 4.5%, and 17% precision (Miao et al., 2022).

A broader CSST cosmology review extends the probe set to BAO, Type Ia supernovae, and cosmic voids, and summarizes the aggregate expectation as percent-level constraints on dark energy and dark matter properties, together with precise tests of gravity (Gong et al., 25 Jan 2025). That review emphasizes complementarity: weak lensing constrains the $\ge 6''\times6''$3–$\ge 6''\times6''$4 combination, clustering constrains distance and growth, clusters trace the halo mass function, and voids add independent sensitivity to large-scale structure (Gong et al., 25 Jan 2025). This synthesis suggests that CSST’s defining cosmological feature is not a single observable but the co-registration of multiple observables within one survey system.

5. Extragalactic, transient, and stellar science programs

Strong-lensing forecasts indicate that the CSST wide-field survey will observe approximately 160,000 galaxy–galaxy strong lenses over its lifespan, with mean Einstein radius $\ge 6''\times6''$5, corresponding mean velocity dispersion $\ge 6''\times6''$6, median magnification $\ge 6''\times6''$7, and unlensed source $\ge 6''\times6''$8-band magnitudes of $\ge 6''\times6''$9 AB. The simulated lensed-image SNR range of 20–1000 implies an Einstein-radius precision scaling from about 1% at SNR $0.074''/{\rm pixel}$00 down to about 0.1% at SNR $0.074''/{\rm pixel}$01, ignoring various modeling systematics (Cao et al., 2023). The same forecast states that CSST should detect rare configurations including double-source-plane lenses and spiral-galaxy deflectors (Cao et al., 2023).

For time-domain astrophysics, the survey is not dedicated to transient work, but its area and cadence still imply a large supernova yield. One photometric simulation study forecasts that, over 10 yr, CSST is expected to observe about 5 million supernovae of various types. After quality cuts, the resulting “gold” sample comprises roughly 7,400 SNe Ia, 2,200 SNe Ibc, and 6,500 SNe II candidates, with correctly classified percentages of 91%, 63%, and 93%, respectively; the same survey configuration can trigger alerts for about 15,500 SNe Ia and 2,100 SNe II candidates at least two days before maximum light, and NUV observations are expected to catch hundreds of shock-cooling events serendipitously every year (Liu et al., 2024).

Galactic stellar science is supported both by the survey design and by extensive preparatory mock catalogs. The first comprehensive Milky Way stellar mock catalogue for the Survey Camera contains approximately 12.6 billion stars down to $0.074''/{\rm pixel}$02 mag in the AB system. It was built with TRILEGAL and includes photometry, astrometry, kinematics, and stellar parameters. On that basis, the authors conclude that crowding does not by itself preclude extension of the CSST Optical Survey into low Galactic latitudes; with PSF FWHM $0.074''/{\rm pixel}$03 and photometric threshold $0.074''/{\rm pixel}$04 mag, the $0.074''/{\rm pixel}$05-band crowding limit remains $0.074''/{\rm pixel}$06 mag over most of the sky and falls to $0.074''/{\rm pixel}$07 mag only in the densest bulge fields (Chen et al., 2023). This suggests that CSST will occupy a regime between Gaia’s all-sky astrometric census and HST’s deep but narrow imaging.

Stellar-population work in dense clusters is another area where the instrument complement matters. For the MCI, one filter-optimization study finds that CMDs made with appropriate UV filters are powerful tools to disentangle populations with different abundances of He, C, N, O, and Mg, while traditional optical CMDs are essentially blind to multiple populations in globular clusters (2206.12068). A later SCam-based simulation reaches a parallel conclusion for the main survey camera: for a globular cluster at 9.6 kpc, the $0.074''/{\rm pixel}$08 colour can separate populations by $0.074''/{\rm pixel}$09 mag for red giants and up to 0.44 mag for dwarfs, and at 20 kpc the system still retains diagnostic power for populations with $0.074''/{\rm pixel}$10 and $0.074''/{\rm pixel}$11 (Li et al., 28 May 2026). In both cases, the wide field relative to HST is central to the scientific argument.

6. Simulations, data products, and methodological limitations

Because CSST has not yet produced on-orbit survey data, a substantial fraction of the literature is devoted to high-fidelity simulations, mock catalogs, and pipeline validation. One mission-wide simulation overview describes a pixel-level framework built upon the GalSim package and the latest CSST instrumental specifications. The pipeline ingests stars from Galaxia with BT-Settl spectra, galaxies from JiuTian N-body plus semi-analytical models with STARDUSTER SEDs, and quasars from SIMQSO; it models astrometry, PSF variation, optical distortions, slitless dispersion, detector non-idealities, and multiple noise sources before outputting FITS-formatted raw files (Wei et al., 10 Nov 2025). The same work reports a Cycle-9 dataset of 2,442 exposures, 50 deg², and 73,260 raw CCD files, totaling approximately 12.5 TB, and validation metrics including photometric recovery without systematic bias to $0.074''/{\rm pixel}$12, PSF interpolation size bias $0.074''/{\rm pixel}$13, and stray-light model discrepancy $0.074''/{\rm pixel}$14 relative to HST Earthshine data (Wei et al., 10 Nov 2025).

A complementary mock-catalogue paper constructs a full-sky light cone to $0.074''/{\rm pixel}$15 for galaxies and a configurable stellar catalog based on TRILEGAL or Galaxia. Galaxy SEDs are generated with the STARDUSTER deep-learning emulator, and weak-lensing shears and convergences are assigned by full-sky multi-plane ray tracing. The resulting catalogue includes positions, redshifts, stellar masses, shapes, sizes, SEDs, lensing shears, and magnifications, and is explicitly intended to support image simulations, weak-lensing forecasts, clustering, and other cosmological analyses (Wei et al., 13 Nov 2025).

Specific subsystem studies illustrate why these simulations are necessary. The slitless-spectroscopy emulator CESS quantifies morphology-dependent self-broadening and overlap contamination, finding that Sérsic index has negligible effect on line broadening, while effective radius, axis ratio, and position angle can modify the line-spread function and thus redshift completeness (Wen et al., 2024). A morphology-calibration study based on realistic mock galaxies shows that CSST deep fields recover HST-like non-parametric indicators much more accurately than the wide survey, and it provides explicit correction functions of the form

$0.074''/{\rm pixel}$16

to align CSST measurements of $0.074''/{\rm pixel}$17, $0.074''/{\rm pixel}$18, $0.074''/{\rm pixel}$19, $0.074''/{\rm pixel}$20, $0.074''/{\rm pixel}$21, and $0.074''/{\rm pixel}$22 with HST benchmarks (Luo et al., 4 May 2025).

The published performance forecasts are not presented as unconditional guarantees. The RV study explicitly cautions that its uncertainties are optimistic upper limits because they do not yet incorporate overlapping spectra in crowded fields, imperfect background subtraction, spatially variable PSFs, CCD charge-transfer inefficiency, or reduced SNR at band edges (Sun et al., 2020). More generally, the preparatory literature treats calibration strategy, reduction pipelines, stray-light mitigation, CTI correction, brighter-fatter correction, deblending, and spectral-overlap control as mission-critical rather than peripheral tasks (Wei et al., 10 Nov 2025). The overall implication is that CSST’s scientific reach depends not only on its hardware parameters—2 m aperture, $0.074''/{\rm pixel}$23 survey camera, $0.074''/{\rm pixel}$24 slitless spectroscopy, and sub-arcsecond image quality—but also on whether these simulation-driven calibration programs successfully transfer pre-launch performance into survey operations.