BAKE in SRF: Thermal Treatments for Niobium Cavities

• BAKE is a thermal treatment process for niobium SRF cavities that modifies the near-surface region to suppress high field Q-slope and optimize Q0 performance.
• Different bake variants, including 120°C mild, two-step low-temperature, and mid-temperature furnace bakes, target specific improvements such as hydride suppression and high Q0.
• Post-bake treatments and testing protocols are critical for managing trade-offs like contamination, quench field, and parasitic mode suppression in SRF applications.

In superconducting radio-frequency practice, BAKE commonly denotes a post-processing heat treatment applied to niobium cavities to alter the RF-active near-surface region and thereby change the intrinsic quality factor $Q_0$, the field dependence of the surface resistance, and the achievable accelerating gradient. In the literature considered here, BAKE spans the standard $120^\circ\mathrm{C}$ mild bake, a modified $75^\circ\mathrm{C}+120^\circ\mathrm{C}$ low-temperature sequence, and medium-temperature furnace baking in the $250\text{–}400^\circ\mathrm{C}$ or $300\text{–}350^\circ\mathrm{C}$ regime; these treatments are discussed mainly for 1.3 GHz TESLA-shape cavities, but also for 650 MHz single-cell cavities and 80.5 MHz quarter-wave resonators (Trenikhina et al., 2015, Grassellino et al., 2018, He et al., 2020, Chouhan et al., 13 Feb 2026, Brown et al., 28 Feb 2026).

1. Principal bake variants

The SRF literature represented here distinguishes three principal bake classes. The first is the conventional low-temperature or “mild bake” at $120^\circ\mathrm{C}$, often for 48 h. The second is a modified low-temperature sequence in which a short dwell near $75^\circ\mathrm{C}$ precedes the conventional $120^\circ\mathrm{C}$ step. The third is the medium-temperature (mid-T) furnace bake, discussed at IHEP, FNAL, and KEK in the $250\text{–}400^\circ\mathrm{C}$ regime and implemented in several papers as a $300^\circ\mathrm{C}$ or $120^\circ\mathrm{C}$0, 3 h vacuum-furnace treatment (He et al., 2020, Grassellino et al., 2018, Chouhan et al., 13 Feb 2026).

Bake variant	Reported regime	Reported effect
Mild bake	$120^\circ\mathrm{C}$1, 48 h	Suppresses HFQS; decreases nanohydride formation
Two-step low-temperature bake	$120^\circ\mathrm{C}$2 for 4 h, then $120^\circ\mathrm{C}$3 for 48 h	Raises quench field to $120^\circ\mathrm{C}$4 in TESLA-shape cavities
Mid-T furnace bake	$120^\circ\mathrm{C}$5 generally; $120^\circ\mathrm{C}$6 or $120^\circ\mathrm{C}$7 for 3 h in specific studies	Raises $120^\circ\mathrm{C}$8; often produces anti-$120^\circ\mathrm{C}$9-slope; may reduce quench field unless post-bake surface treatment is used

These bake classes are not interchangeable. The $75^\circ\mathrm{C}+120^\circ\mathrm{C}$0 bake is presented primarily as a cure for the high field $75^\circ\mathrm{C}+120^\circ\mathrm{C}$1 slope in electropolished cavities. The $75^\circ\mathrm{C}+120^\circ\mathrm{C}$2 sequence is presented primarily as a high-gradient treatment for 1.3 GHz TESLA-shape cavities. Mid-T bake is presented mainly as a high-$75^\circ\mathrm{C}+120^\circ\mathrm{C}$3 treatment, with explicit motivation to reduce cryogenic load and operating cost in continuous-wave systems. A plausible implication is that “BAKE” in SRF usage is less a single recipe than a family of thermally defined near-surface modifications, differentiated by their dominant performance target.

2. Low-temperature baking and the suppression of high-field losses

The standard $75^\circ\mathrm{C}+120^\circ\mathrm{C}$4 mild bake is treated in the literature as a near-surface intervention on the RF penetration layer of niobium. In electropolished 1.3 GHz TESLA-shape cavities, thermometry during RF testing showed that an unbaked EP cavity developed strong localized heating above about $75^\circ\mathrm{C}+120^\circ\mathrm{C}$5, whereas an EP+$75^\circ\mathrm{C}+120^\circ\mathrm{C}$6, 48 h cavity did not. At $75^\circ\mathrm{C}+120^\circ\mathrm{C}$7, the unbaked cavity exhibited extended hot regions with temperature rise up to $75^\circ\mathrm{C}+120^\circ\mathrm{C}$8, while the baked cavity showed no such extended dissipation. Cryogenic TEM-based diffraction then found ordered nanoscale niobium hydrides in EP material at 94 K, much less hydride signal after $75^\circ\mathrm{C}+120^\circ\mathrm{C}$9 baking, and restoration of hydride formation after $250\text{–}400^\circ\mathrm{C}$0 degassing for 3 h followed by $250\text{–}400^\circ\mathrm{C}$1 BCP. Quantitatively, 68% of probed spots in EP material showed additional low-temperature reflections, versus 27% for EP120C; SEND at 94 K identified $250\text{–}400^\circ\mathrm{C}$2-NbH and $250\text{–}400^\circ\mathrm{C}$3-Nb$250\text{–}400^\circ\mathrm{C}$4H$250\text{–}400^\circ\mathrm{C}$5 in EP material and no extra hydride reflections in the mapped EP120C sample (Trenikhina et al., 2015).

The same paper reports that the native amorphous surface oxide remained about $250\text{–}400^\circ\mathrm{C}$6 thick before and after the $250\text{–}400^\circ\mathrm{C}$7 bake, identified as amorphous Nb$250\text{–}400^\circ\mathrm{C}$8O$250\text{–}400^\circ\mathrm{C}$9, while STEM-EELS showed a larger Nb $300\text{–}350^\circ\mathrm{C}$0 chemical shift after baking, interpreted as slight inward oxygen diffusion. No niobium pentoxide was found along grain boundaries in either EP or EP120C samples. Within that paper’s interpretation, the practical role of the mild bake is to suppress nanoscale hydride formation in the first $300\text{–}350^\circ\mathrm{C}$1, thereby removing the HFQS rather than merely changing oxide thickness.

A distinct low-temperature variant adds a short $300\text{–}350^\circ\mathrm{C}$2 step before the conventional $300\text{–}350^\circ\mathrm{C}$3 bake. In 1.3 GHz TESLA-shape cavities, electropolishing followed by 4 h at $300\text{–}350^\circ\mathrm{C}$4 and 48 h at $300\text{–}350^\circ\mathrm{C}$5 raised the continuous-wave accelerating field to $300\text{–}350^\circ\mathrm{C}$6, corresponding to about $300\text{–}350^\circ\mathrm{C}$7, with $300\text{–}350^\circ\mathrm{C}$8 maintained to quench. Two cavities reached $300\text{–}350^\circ\mathrm{C}$9; two others reached $120^\circ\mathrm{C}$0 but were limited by field emission. The paper attributes the improvement mainly to a lower temperature-dependent component of the surface resistance, while the field dependence of the residual part remained similar to standard $120^\circ\mathrm{C}$1-baked cavities (Grassellino et al., 2018).

Low-temperature BAKE is not frequency-independent in its reported effects. For 80.5 MHz FRIB quarter-wave resonators at 4.3 K, a 48 h, $120^\circ\mathrm{C}$2 bake-out reduced the surface resistance by a factor of 2 to 3, reduced the medium-field $120^\circ\mathrm{C}$3-slope by 38% on average, lowered the thermal-feedback slope parameter $120^\circ\mathrm{C}$4 from $120^\circ\mathrm{C}$5 to $120^\circ\mathrm{C}$6, and increased $120^\circ\mathrm{C}$7 at the design field by about a factor of 2. The same study interprets the improvement as a reduction in the BCS contribution rather than a reduction in the residual term (Brown et al., 28 Feb 2026).

3. Medium-temperature furnace baking as a high-$120^\circ\mathrm{C}$8 treatment

Mid-T bake is presented as a vacuum-furnace heat treatment intended to raise $120^\circ\mathrm{C}$9 at useful accelerating field. At IHEP, exploratory 1.3 GHz 1-cell studies covered $75^\circ\mathrm{C}$0, $75^\circ\mathrm{C}$1, $75^\circ\mathrm{C}$2, and $75^\circ\mathrm{C}$3, each for 3 h in a small dual-vacuum furnace. All baked 1-cell cavities showed improved $75^\circ\mathrm{C}$4, clear anti-$75^\circ\mathrm{C}$5-slope over roughly 5–18 MV/m, and maximum gradients from 25.1 to 36.9 MV/m; the average $75^\circ\mathrm{C}$6 reached $75^\circ\mathrm{C}$7 at 16 MV/m, and the highest $75^\circ\mathrm{C}$8 was $75^\circ\mathrm{C}$9 at 16 MV/m for cavity S25#. IHEP then selected $120^\circ\mathrm{C}$0, 3 h for six 1.3 GHz 9-cell cavities in a new big furnace, obtaining an average $120^\circ\mathrm{C}$1 of $120^\circ\mathrm{C}$2 at 16 MV/m, with $120^\circ\mathrm{C}$3 in the range $120^\circ\mathrm{C}$4 to $120^\circ\mathrm{C}$5 for gradients between 16 and 24 MV/m. All six exceeded the LCLS-II-HE specification $120^\circ\mathrm{C}$6 at 21 MV/m, and five of six reached the CEPC specification $120^\circ\mathrm{C}$7 at 24 MV/m (He et al., 2020).

A cryomodule-scale demonstration extended mid-T bake from single-cavity qualification to system operation. The first 1.3 GHz cryomodule containing eight 9-cell medium-temperature baked cavities achieved an average cavity $120^\circ\mathrm{C}$8 of about $120^\circ\mathrm{C}$9 at 16 MV/m and about $250\text{–}400^\circ\mathrm{C}$0 at 21 MV/m in horizontal test, operated stably above 191 MV total CW RF voltage, and sustained an average cavity CW accelerating gradient above 23 MV/m. At 133 MV total voltage, corresponding to 16 MV/m average gradient, the total 2 K heat load was $250\text{–}400^\circ\mathrm{C}$1; at 173 MV, corresponding to 21 MV/m average gradient, the total 2 K heat load was $250\text{–}400^\circ\mathrm{C}$2. The paper also states that similar average cavity $250\text{–}400^\circ\mathrm{C}$3 values were observed for slow cooldown $250\text{–}400^\circ\mathrm{C}$4 and fast cooldown $250\text{–}400^\circ\mathrm{C}$5, which it presents as evidence that the mid-T bake cavity may not require fast cool-down, while also emphasizing that the result is preliminary (Zhai et al., 2023).

Mid-T bake is also presented as a process simplification relative to nitrogen doping. A 2026 study describes mid-T baking at $250\text{–}400^\circ\mathrm{C}$6 as a Fermilab-developed surface-modification method intended to enhance $250\text{–}400^\circ\mathrm{C}$7, and states that furnace-baked cavities can be processed with one bulk EP only and no post-bake light EP in some IHEP implementations. In that processing landscape, mid-T bake is positioned as a high-$250\text{–}400^\circ\mathrm{C}$8 alternative that avoids nitrogen gas exposure and the few-micron post-doping light EP associated with standard doping flows. At the same time, the same 2026 study emphasizes that $250\text{–}400^\circ\mathrm{C}$9, 3 h furnace-baked cavities often exhibit premature quench at relatively low accelerating fields unless the top RF surface layer is chemically removed afterward (Chouhan et al., 13 Feb 2026).

4. Surface resistance, near-surface structure, and post-bake surface treatment

Across the SRF bake literature, the central performance relations are written in terms of surface resistance: $300^\circ\mathrm{C}$0 For 80.5 MHz quarter-wave resonators, the low-frequency study further writes

$300^\circ\mathrm{C}$1

and notes that, because $300^\circ\mathrm{C}$2, the relative importance of BCS and residual terms changes strongly with frequency (Brown et al., 28 Feb 2026). Mid-T work at IHEP likewise attributes anti-$300^\circ\mathrm{C}$3-slope mainly to a field-dependent decrease in $300^\circ\mathrm{C}$4 between about 5 and 20 MV/m, while $300^\circ\mathrm{C}$5 increases with gradient (He et al., 2020).

The standard $300^\circ\mathrm{C}$6 bake is interpreted mainly through hydride suppression and associated hydrogen trapping in the first $300^\circ\mathrm{C}$7. The nanostructural study explicitly favors the view that HFQS comes from lossy nanoscale niobium hydrides precipitating in the near-surface region during cooldown, and that $300^\circ\mathrm{C}$8 baking decreases hydride formation and may reduce hydride size through vacancy diffusion inward and/or slight inward oxygen diffusion. The same work also notes that baked cavities often have about $300^\circ\mathrm{C}$9 higher residual resistance, which it associates with the oxygen-enriched subsurface region (Trenikhina et al., 2015).

Medium-temperature BAKE introduces a different materials issue. A 2026 study of 1.3 GHz and 650 MHz single-cell cavities baked at $120^\circ\mathrm{C}$00 for 3 h in a vacuum furnace reports that the bake raised $120^\circ\mathrm{C}$01 but also caused premature quench, suspected to arise from surface contaminants, most likely NbC. SIMS on companion niobium samples showed substantially higher NbC signal in the as-baked surface than after approximately 120 nm electropolishing, while oxygen intensity was similar in the two samples. Guided by that result, the study applied “ultralight chemical removal” of roughly 100–300 nm to the top RF surface layer. In the 1.3 GHz cavity TE1RI010, $120^\circ\mathrm{C}$02 increased from $120^\circ\mathrm{C}$03 after mid-T bake to $120^\circ\mathrm{C}$04 at 72 mT after 108 nm EP, while the quench field recovered from 22 MV/m to 32 MV/m. In the 650 MHz cavity B9AS-AES-003, the first ultralight EP of about 147 nm raised $120^\circ\mathrm{C}$05 from $120^\circ\mathrm{C}$06 to $120^\circ\mathrm{C}$07, with little additional $120^\circ\mathrm{C}$08 gain after a second 118 nm removal. The authors interpret this as evidence that the harmful contamination is confined to the outermost RF layer and that ultralight EP can remove it while preserving the beneficial mid-T-modified state beneath (Chouhan et al., 13 Feb 2026).

A broader implication is that BAKE is not only a thermal process but also a surface-state selection problem. Low-temperature baking is reported to suppress one deleterious near-surface phase population, whereas some furnace mid-T recipes can introduce another deleterious top-layer residue. The practical distinction is therefore not simply “baked” versus “unbaked,” but which bake, at which temperature, and with what post-bake surface condition.

5. Test infrastructure, metrology, and cryomodule integration

The very success of mid-T bake in producing high-$120^\circ\mathrm{C}$09 cavities created a distinct metrology problem in continuous-wave vertical testing. During CW RF cold tests of 1.3 GHz 9-cell cavities in liquid helium at 2 K, IHEP observed spontaneous excitation of parasitic modes with resonance frequencies close to the main $120^\circ\mathrm{C}$10 mode, especially the $120^\circ\mathrm{C}$11 mode at 1297 MHz. This parasitic mode, with quality factor of order $120^\circ\mathrm{C}$12, produced erroneous cavity characterization because the power meter measured both the $120^\circ\mathrm{C}$13 mode and the $120^\circ\mathrm{C}$14 mode together; the paper states that in this situation “The calculate value $120^\circ\mathrm{C}$15 increased and the $120^\circ\mathrm{C}$16 decreased.” The mode was associated with field-emission electrons, appeared from about 15 MV/m in one discussion and about 18 MV/m in the improved-system discussion, and could receive power on the order of dozens of watts. IHEP ultimately added a direct RF feedback loop to isolate and suppress the parasitic mode, after considering and rejecting repeated HPR or changed input coupling as the preferred remedy (Mi et al., 2021).

This test-stand result is operationally important because it links bake success to infrastructure requirements. The paper is explicit that parasitic-mode excitation was encountered in “high Q cavities” and that the higher-$120^\circ\mathrm{C}$17 condition created by mid-T bake is where the issue appears. A plausible implication is that BAKE can reveal weaknesses in diagnostics that were not limiting for EP-baseline cavities.

Cryomodule integration adds another layer of BAKE-specific systems behavior. In the eight-cavity IHEP cryomodule, most cavities showed nearly no multipacting quench, but processing time was dominated by a few outliers and by ancillary-component issues such as HOM coupler feedthrough overheating and input coupler cold-window heating. The module nevertheless achieved 12-hour stable operation at 133 MV with all eight cavities at 16.0 MV/m and without any trip, while the measured radiation at 16 MV/m with all cavities powered was $120^\circ\mathrm{C}$18, well below the specification $120^\circ\mathrm{C}$19. This indicates that BAKE performance cannot be evaluated solely at the level of vertical-test $120^\circ\mathrm{C}$20; it also depends on assembly cleanliness, auxiliary thermal contacts, magnetic hygiene, and whether the cryomodule can preserve the baked cavity state during string assembly and cooldown (Zhai et al., 2023).

6. Frequency dependence, unresolved issues, and scope limitations

The bake literature does not support a single universal outcome across all SRF frequencies and geometries. The 80.5 MHz quarter-wave resonator study found no improvement in cavity performance after a 3 h bake-out at $120^\circ\mathrm{C}$21, explicitly “in contrast to observations for higher-frequency cavities.” In that system, the successful bake was the 48 h, $120^\circ\mathrm{C}$22 low-temperature bake, whereas the medium-temperature bake produced no measurable benefit in $120^\circ\mathrm{C}$23, BCS resistance, or medium-field slope (Brown et al., 28 Feb 2026). This directly limits any claim that mid-T bake is generically advantageous.

Several bake papers also leave critical recipe information unstated. The IHEP parasitic-mode study makes mid-T bake central to the emergence of the test problem and shows a strong $120^\circ\mathrm{C}$24 improvement over an EP baseline, but it does not report the precise furnace temperature, whether one or multiple temperatures were used, bake duration, pressure or vacuum level, residual gas composition beyond general cleanliness considerations, deliberate atmospheres, or post-bake handling details. It therefore documents the operational consequence of obtaining very high $120^\circ\mathrm{C}$25, not a complete process specification (Mi et al., 2021).

Even where recipes are specified, open questions remain. The $120^\circ\mathrm{C}$26 study does not map an optimum dwell-time window around $120^\circ\mathrm{C}$27, and explicitly treats its hydride-suppression interpretation as plausible rather than fully proven. The mid-T contamination study identifies NbC as the main suspect but still uses cautious language such as “possibly NbC” and “most likely NbC.” The cryomodule paper identifies preliminary evidence that mid-T baked cavities may not require fast cooldown, but also calls for further study of cooldown rate, remnant magnetic field, thermal current, and flux expulsion. The IHEP furnace-bake program reports residual resistance values $120^\circ\mathrm{C}$28, higher than the $120^\circ\mathrm{C}$29 quoted for FNAL and the $120^\circ\mathrm{C}$30 quoted for KEK, and therefore leaves room for further process optimization (Grassellino et al., 2018, Chouhan et al., 13 Feb 2026, He et al., 2020, Zhai et al., 2023).

Taken together, these results define BAKE in SRF not as a single mature recipe but as a technically differentiated class of thermal treatments whose reported benefits include suppression of HFQS, reduction of BCS-dominated losses, generation of anti-$120^\circ\mathrm{C}$31-slope, and record or near-record CW performance, while whose reported liabilities include recipe sensitivity, contamination-driven early quench, frequency dependence, and the need for upgraded test infrastructure once very high $120^\circ\mathrm{C}$32 is achieved.