Transition Edge Sensor Bolometers

• Transition Edge Sensor bolometers are superconducting detectors operating in the steep transition region to amplify small absorbed power changes.
• They employ negative electrothermal feedback through voltage biasing to enhance stability and reduce response time, crucial for diverse applications.
• Various designs integrate suspended membranes, antenna coupling, and advanced materials to optimize noise performance and scalability in large arrays.

Searching arXiv for the cited TES bolometer papers and closely related work. I don’t have a callable arXiv search tool exposed in this interface, so I’m proceeding strictly from the supplied arXiv records and details. Transition Edge Sensor bolometers are superconducting thermal detectors in which a resistor $R(T,I)$ is voltage-biased in the steep superconducting transition, thermally linked to a cold bath by a conductance $G$, and mounted on an island with heat capacity $C$. In this operating regime, small changes in absorbed power shift the TES resistance and therefore the Joule power, producing strong negative electrothermal feedback with loop gain $\mathcal{L}\equiv (P_0\alpha)/(G T_0)$ or, equivalently in closely related notation, $\mathcal{L}\approx (P_{\rm Joule}\alpha)/(G T)$ (Sonka et al., 2017, Sonka et al., 2023). TES bolometers are a standard detector technology in CMB instrumentation and also appear in sub-mm, THz, GHz, and X-ray systems, with implementations ranging from suspended membrane absorbers and antenna-coupled planar pixels to nanoscale hot-electron devices and RF-readout microbridges (Ade et al., 2015, Paolucci et al., 2020, Shitov et al., 2014).

1. Fundamental electrothermal description

The basic thermal-balance equation used across TES bolometer analyses is

$$c\,\frac{dT}{dt}=P_{\rm Joule}+P_{\rm opt}-G\,[T-T_{\rm bath}] ,$$

where $c$ is the island heat capacity, $P_{\rm Joule}=V_{\rm TES}I_{\rm TES}$ is the electrical dissipation, $P_{\rm opt}$ is the absorbed optical power, and $T_{\rm bath}$ is the bath temperature (Sonka et al., 2023). In the superconducting transition the TES has normal resistance $G$0 at $G$1 and zero resistance below $G$2, while the logarithmic temperature sensitivity $G$3 is large. Voltage bias with a nearly ideal shunt resistor stabilizes the detector because an increase in optical loading raises $G$4, lowers $G$5, and therefore reduces $G$6 (Sonka et al., 2023).

The thermal link and heat capacity set the intrinsic time constant $G$7 or $G$8, and electrothermal feedback reduces the effective response time to $G$9 in one common convention (Sonka et al., 2023). In a widely used small-signal form for CMB bolometers, the current responsivity under strong ETF approaches $C$0 (0908.1997), while a more explicit frequency-domain approximation writes

$C$1

for $C$2 (Ade et al., 2015). The TES electrical response can also be expressed through a complex impedance $C$3 or admittance $C$4, which is the basis of standard stability and parameter-extraction measurements (Sonka et al., 2017).

Noise-equivalent power is the dominant sensitivity metric. For phonon noise the frequently used form is

$C$5

with $C$6–1, while total detector noise is written as a quadrature sum of phonon, Johnson, readout, and additional terms (Sonka et al., 2023). In CMB applications photon shot noise and photon bunching are commonly included explicitly,

$C$7

and the interplay of phonon, photon, Johnson, and readout noise determines whether a given design is photon-noise-dominated on sky (Ade et al., 2015).

2. Thermal architecture, internal coupling, and stability

A central result of TES bolometer engineering is that stability is not set only by the bath conductance $C$8 and the external bias circuit. For 270 GHz CMB bolometers, increasing leg conductance to carry higher absorbed sky power caused the internal thermal link between the Ti TES and the PdAu island heat capacity to become the bottleneck (Sonka et al., 2017). Sonka et al. considered two internal thermal resistances: phonon-radiation-limited conductance in the Si$C$9N$\mathcal{L}\equiv (P_0\alpha)/(G T_0)$0 island aperture and electron-phonon coupling in the thin Ti TES. For the SiN aperture, the limiting conductance was written

$\mathcal{L}\equiv (P_0\alpha)/(G T_0)$1

with $\mathcal{L}\equiv (P_0\alpha)/(G T_0)$2 for SiN; for $\mathcal{L}\equiv (P_0\alpha)/(G T_0)$3 and $\mathcal{L}\equiv (P_0\alpha)/(G T_0)$4 this gives $\mathcal{L}\equiv (P_0\alpha)/(G T_0)$5 (Sonka et al., 2017).

The corresponding stability criterion uses $\mathcal{L}\equiv (P_0\alpha)/(G T_0)$6. Following George et al., stable ETF requires, to first order,

$\mathcal{L}\equiv (P_0\alpha)/(G T_0)$7

If $\mathcal{L}\equiv (P_0\alpha)/(G T_0)$8 falls below the loop gain achievable in the transition, the device latches or oscillates (Sonka et al., 2017). In the 270 GHz case, the baseline type A design had PdAu located $\mathcal{L}\equiv (P_0\alpha)/(G T_0)$9 from the Ti TES, so heat had to flow through the $\mathcal{L}\approx (P_{\rm Joule}\alpha)/(G T)$0 SiN island aperture; the measured $\mathcal{L}\approx (P_{\rm Joule}\alpha)/(G T)$1 led to ETF instability below $\mathcal{L}\approx (P_{\rm Joule}\alpha)/(G T)$2 of $\mathcal{L}\approx (P_{\rm Joule}\alpha)/(G T)$3 and a simultaneous biasable yield per column of $\mathcal{L}\approx (P_{\rm Joule}\alpha)/(G T)$4. In optimized type D and type F designs, PdAu overlapped the TES across oxide or both Al-TES and Ti-TES without direct metal contact, giving $\mathcal{L}\approx (P_{\rm Joule}\alpha)/(G T)$5–250, stable ETF down to $\mathcal{L}\approx (P_{\rm Joule}\alpha)/(G T)$6 of $\mathcal{L}\approx (P_{\rm Joule}\alpha)/(G T)$7, and a simultaneous biasable yield of $\mathcal{L}\approx (P_{\rm Joule}\alpha)/(G T)$8; the Ti-island time constant changed only marginally, by $\mathcal{L}\approx (P_{\rm Joule}\alpha)/(G T)$9 (Sonka et al., 2017).

A broader Al-Mn stability study formulated the problem as a two-body thermal network composed of the TES and an added heat-capacity element, the “BLING,” connected by an internal conductance $$c\,\frac{dT}{dt}=P_{\rm Joule}+P_{\rm opt}-G\,[T-T_{\rm bath}] ,$$0 (George et al., 2013). In that framework, practical design targets were $$c\,\frac{dT}{dt}=P_{\rm Joule}+P_{\rm opt}-G\,[T-T_{\rm bath}] ,$$1 and $$c\,\frac{dT}{dt}=P_{\rm Joule}+P_{\rm opt}-G\,[T-T_{\rm bath}] ,$$2, with stable devices occupying ranges such as $$c\,\frac{dT}{dt}=P_{\rm Joule}+P_{\rm opt}-G\,[T-T_{\rm bath}] ,$$3–25 ms and $$c\,\frac{dT}{dt}=P_{\rm Joule}+P_{\rm opt}-G\,[T-T_{\rm bath}] ,$$4–180 (George et al., 2013). Adding Au or PdAu heat capacity, extending BLING over Nb leads, or introducing normal-metal bars and dots on the TES were all used to trade bandwidth, transition steepness, and stability. A recurring misconception is that ETF instability is simply a readout artifact; the measurements instead show that internal thermal decoupling can dominate the onset of oscillation or latching (Sonka et al., 2017, George et al., 2013).

3. Device architectures and materials systems

Conventional TES bolometers for CMB work are typically membrane-isolated absorbers. EBEX and related designs used a gold-plated silicon-nitride spider-web absorber with an Al/Ti proximity-effect bilayer TES at the center and a Au thermal ballast or Au ring to dominate the heat capacity and stabilize the ETF bandwidth (0908.1997, MacDermid et al., 2014). In one prototype array the web diameter was 2.1 mm, the membrane was low-stress Si$$c\,\frac{dT}{dt}=P_{\rm Joule}+P_{\rm opt}-G\,[T-T_{\rm bath}] ,$$5N$$c\,\frac{dT}{dt}=P_{\rm Joule}+P_{\rm opt}-G\,[T-T_{\rm bath}] ,$$6 with 1 $$c\,\frac{dT}{dt}=P_{\rm Joule}+P_{\rm opt}-G\,[T-T_{\rm bath}] ,$$7m thickness, 6 $$c\,\frac{dT}{dt}=P_{\rm Joule}+P_{\rm opt}-G\,[T-T_{\rm bath}] ,$$8m line width, and 117 $$c\,\frac{dT}{dt}=P_{\rm Joule}+P_{\rm opt}-G\,[T-T_{\rm bath}] ,$$9m hexagonal grid spacing, and the designed average thermal conductance was $c$0 (0908.1997). EBEX later operated 504, 342, and 109 TESs at 150, 250, and 410 GHz shortly after reaching float, with measured median low-loop-gain time constants of 88, 46, and 57 ms, respectively (Collaboration et al., 2018).

A second major class is the antenna-coupled planar TES. Commercially fabricated 150 mm wafers integrated a dual-polarization, dichroic sinuous antenna, on-chip microstrip band-defining filters, and a TES thermistor on a suspended silicon-nitride membrane, with average electrical probe grade yield of $c$1 and center frequencies $c$2 and $c$3 (Suzuki et al., 2019). The BICEP2, Keck Array, and SPIDER detectors used lithographed dual-polarized slot antennas, microstrip summing trees, 3-pole bandpass filters, and a Ti TES on a released SiN island; these arrays repeatedly produced spectral bands with $c$4–$c$5 bandwidth at 95, 150, or 220 GHz, end-to-end optical efficiencies routinely $c$6 or higher, and per detector sensitivities of NET $c$7 uKrts (Ade et al., 2015). In SPT-3G, Al-Mn TES wafers replaced Ti/Au designs, giving a measured mean $c$8 mK, $c$9, thermal conductances of $P_{\rm Joule}=V_{\rm TES}I_{\rm TES}$0 pW/K at 95/150/220 GHz, and optical efficiencies of $P_{\rm Joule}=V_{\rm TES}I_{\rm TES}$1, $P_{\rm Joule}=V_{\rm TES}I_{\rm TES}$2, and $P_{\rm Joule}=V_{\rm TES}I_{\rm TES}$3 (Anderson et al., 2019).

Materials engineering has become increasingly central. Heated-sputtered hafnium TESs provide a tunable $P_{\rm Joule}=V_{\rm TES}I_{\rm TES}$4 between 140 mK and 210 mK, with $P_{\rm Joule}=V_{\rm TES}I_{\rm TES}$5, loop gain exceeding $P_{\rm Joule}=V_{\rm TES}I_{\rm TES}$6 deep in the transition, and interdigitated geometries that span normal resistances from 10 milli-Ohm to 1 Ohm for compatibility with TDM, FDM, and $P_{\rm Joule}=V_{\rm TES}I_{\rm TES}$7-mux readout (Rotermund et al., 2024). The empirical dependence

$P_{\rm Joule}=V_{\rm TES}I_{\rm TES}$8

with $P_{\rm Joule}=V_{\rm TES}I_{\rm TES}$9 in $P_{\rm opt}$0C and $P_{\rm opt}$1 in mK was associated with reduced film stress, larger crystal sizes, and a smaller relative abundance of the m-plane to c-plane $P_{\rm opt}$2 phase (Rotermund et al., 2024). By contrast, Al-Mn fabrication was emphasized for commercial throughput and for uniform multi-chroic arrays (Suzuki et al., 2019, Anderson et al., 2019).

Alternative architectures depart from the suspended-island paradigm. A nanoscale Al/Cu bilayer “nano-TES” used a $P_{\rm opt}$3 active region, thick Al Andreev mirrors, and NIS tunnel probes; for sample n-T1 at $P_{\rm opt}$4 mK, the predicted values were $P_{\rm opt}$5, $P_{\rm opt}$6s, $P_{\rm opt}$7s, and $P_{\rm opt}$8, with $P_{\rm opt}$9 GHz in calorimetric mode (Paolucci et al., 2020). High-impedance Nb$T_{\rm bath}$0Si$T_{\rm bath}$1 TES meanders eliminated delicate membrane structures by using intrinsic electron-phonon decoupling; a 204-pixel array demonstrated light absorption close to $T_{\rm bath}$2 and an NEP of about $T_{\rm bath}$3 (Nones et al., 2012). A related NbSi analysis extracted $T_{\rm bath}$4 and $T_{\rm bath}$5 for 50 nm films, and $T_{\rm bath}$6 with $T_{\rm bath}$7 for 20 nm films (Liu et al., 2014).

4. Biasing, multiplexing, and circuit nonidealities

TES bolometers are tightly coupled to their readout architecture. In frequency-domain multiplexing, each TES is placed in series with a narrow-band LC filter at

$T_{\rm bath}$8

and the summed currents are read by a SQUID amplifier (Hattori et al., 2013). APEX-SZ used seven TESs per SQUID module, with carriers at 0.3–1.0 MHz and series $T_{\rm bath}$9H spiral inductors plus tuned NP0 capacitors (Schwan et al., 2010). EBEX used digital frequency-domain multiplexing with a multiplexing factor of 16, carrier frequencies between 400 and 1200 kHz, a nulling comb $G$00 out of phase with the carriers, and autonomous remote tuning of SQUIDs and TES biases (MacDermid et al., 2014).

Scaling fMUX to larger channel counts shifts attention to parasitics. POLARBEAR-2 extended an 8-bolometer architecture to 32 bolometers per SQUID, requiring a bandwidth of 3 MHz and introducing Digital Active Nulling (DAN) to keep the net SQUID flux small at each carrier frequency (Hattori et al., 2013). Above 1 MHz, parasitic inductance and capacitor ESR degrade voltage bias, resonance $G$01, and crosstalk. The cross-talk estimate

$G$02

drove a requirement of $G$03 nH at 3 MHz, while commercial multi-layer ceramic capacitors were acceptable below 1 MHz but reached $G$04–$G$05 ESR at 3 MHz (Hattori et al., 2013).

Time-domain and microwave multiplexing place different demands on TES resistance. Time-division multiplexed readout was routine at factors of 32–128 (Suzuki et al., 2019), and in some CMB arrays an entire column of 32 TESs shared one DC bias, so a narrow stable bias region directly reduced usable yield (Sonka et al., 2017). Microwave SQUID multiplexing couples TES current to an rf-SQUID and then to a superconducting microwave resonator at 4–8 GHz, with up to $G$06 channels per coaxial line (Suzuki et al., 2019). A distinct RF-readout TES concept embedded a Nb microbridge directly into a folded quarter-wave coplanar-waveguide resonator at $G$07 GHz and used a HEMT amplifier rather than a SQUID; the measured optical sensitivity in the 600–700 GHz band was $G$08 at $G$09 K (Shitov et al., 2014).

A further nonideality is the distinction between on-island and off-island Johnson noise. In an AC-biased Thévenin-equivalent analysis, strong ETF suppresses Johnson noise generated in the bolometer resistance, but it provides no suppression of Johnson noise from external series resistance (Haan, 2 Sep 2025). In the high-loop-gain limit, the bolometer Johnson contribution vanishes while the parasitic contribution remains unsuppressed. This identifies off-island series resistance as a hard noise floor rather than a term that can be removed merely by increasing $G$10 (Haan, 2 Sep 2025).

5. Characterization, calibration, and parameter extraction

TES bolometers are characterized through I–V curves, load curves, impedance measurements, bath ramps, and controlled optical loading. For the 270 GHz instability study, the impedance measurement setup used a 3 m$G$11 shunt resistor, a small square-wave perturbation of amplitude $G$12V swept from 2.6 Hz to 1.6 kHz, and SQUID demodulation of the first-harmonic TES current. At each bias point the complex admittance $G$13 was extracted and fit with a 2-pole model to obtain $G$14, $G$15, $G$16, and $G$17; instability appeared as a collapse of phase margin and a shift of the poles toward lower frequency as bias was reduced (Sonka et al., 2017).

Large arrays complicate the separation of thermal and optical parameters. A Simons Observatory analysis introduced a joint- or “together-fits” methodology in which bath-temperature ramps and cold-load ramps are fit simultaneously with a dark-heating coefficient $G$18: $G$19 Using standard independent fits on a single bath ramp, the fitted $G$20 distribution for unmasked detectors was bimodal and differed by $G$21 mK from the masked set. With the together-fits on the standard dataset, the $G$22 difference dropped to 1.0 mK, and on the full special dataset it further improved to 0.66 mK. The same procedure eliminated the need for a separate cooldown of masked detectors, so thermal characterization of the full array could be done in a single $G$23 hr cryostat run instead of two (Sonka et al., 2023).

Daily calibration strategies can be built directly from the bias curve archive. For the CLASS Q-band array, an “I–V bin” method used daily I–V curves, binned by saturation power, to assign responsivities from repeated operating points (Appel et al., 2022). The median standard error within a bin was 0.3%, and Moon amplitudes from detector pairs sharing the same feedhorn indicated a TES calibration error of 0.5% (Appel et al., 2022). This calibration philosophy reflects a broader trend in TES arrays: rather than treating thermal characterization, optical efficiency, responsivity, and gain drift as separate tasks, current workflows increasingly solve them in a common fit or common operational archive (Sonka et al., 2023, Appel et al., 2022).

6. Scientific use, scaling, and persistent constraints

TES bolometers are central to modern CMB instrumentation because they scale to large focal planes. The Simons Observatory aims to field 70,000 TES bolometers (Sonka et al., 2023). POLARBEAR-2 planned 7,588 TES bolometers coupled to two-band polarization-sensitive antennas (Hattori et al., 2013). The Keck Array deployed 2560 detectors as 1280 matched pairs with a combined instantaneous sensitivity of $G$24 uKrts (Ade et al., 2015). SPT-3G operates 16,000 TES bolometers observing at 95, 150, and 220 GHz (Anderson et al., 2019). Commercial fabrication studies argued that a network of 3–5 production lines could deliver the $G$25 science-grade wafers required by CMB-S4 within three years (Suzuki et al., 2019). Outside broadband CMB polarimetry, APEX-SZ used 280 TES bolometers to map 48 clusters (Schwan et al., 2010), and TIME developed 1920 detectors for 183–325 GHz line-intensity mapping (Butler et al., 2 Oct 2025).

The resulting performance envelope is shaped as much by systematics as by intrinsic detector physics. Balloon-borne EBEX measurements attributed excess noise above predicted readout plus Johnson terms to in-flight EMI coupling from telemetry and attitude motors and identified unaccounted spill-over as an additional optical load (Collaboration et al., 2018). In ground-based high-frequency TES columns, variations in detector properties and optical loading within a multiplexed column restricted the common stable bias range until internal thermal coupling was improved (Sonka et al., 2017). TIME identified series resistance in Kapton flex cables as comparable to TES normal resistance, degrading electrothermal feedback until a redesigned cable reduced series resistance to $G$26 m$G$27 (Butler et al., 2 Oct 2025). These examples show that TES bolometer optimization is inseparable from cryogenic wiring, bias distribution, optical loading, and electromagnetic hygiene.

Magnetic susceptibility is another persistent constraint because TESs are intrinsically sensitive to magnetic fields. A superconducting groundplane consisting of a 65 nm sputtered Nb film on the backside of the Si$G$28N$G$29 membrane produced a measured shielding factor of $G$30–80 for external fields and reduced self-induced eddy-current losses in AC-biased detectors (Wit et al., 2022). Shielded pixels showed IV curves that overlapped up to $G$31T and maintained 5.9 keV X-ray performance with no degradation of the energy resolution or shifts of the energy scale calibration for fields of several microTesla (Wit et al., 2022). A plausible implication is that future TES bolometer systems, especially where mass, volume, or internal self-fields limit the usefulness of bulk shielding, will increasingly combine local on-chip mitigation with conventional magnetic shielding rather than relying on external shields alone.