betaFeedback

WorksPhysics and Astronomy

Electrothermal behavior of superconducting nanowires buried in a commercial CMOS process

Abstract

Superconducting films native to a commercial CMOS process enable co-integration of superconducting devices (high-kinetic-inductance elements and nanowire detectors) with cryogenic CMOS on one monolithic die. Such films are buried in the front-end-of-line, beneath the back-end-of-line metal stack, an environment that might quench a self-heated hot spot. Current-voltage characteristics are reported for titanium nitride (TiN) nanowires in the polycrystalline-silicon resistor layer of a 22 nm FD-SOI process, and measured retrapping currents are compared with thermal models for longitudinal conduction to the contacts and interfacial phonon cooling into the stack. The wires are strongly hysteretic, with switching-to-retrapping current ratios I_sw/I_r of 7.4-9.8, sustaining a normal hot spot. For two wires of length ell = 50 micrometers and widths W = 0.50 and 1.00 micrometers, over bath temperatures T_b = 0.15-1.19 K, I_r follows the interfacial form (T_hs^n - T_b^n)^(1/2) with n = 4 (T_hs, the effective hot-spot temperature), and is strongly inconsistent with the (T_hs - T_b)^(1/2) constant-kappa longitudinal-conduction limit; free-exponent fits give n = approximately 4.8-5.0. An effective cooling coefficient Sigma_K = 7.9(3) W/(m^2 K^4) describes the data: 31(21) times below the acoustic- (diffuse-) mismatch value for a single clean interface, and approximately 12 times below the lowest previously reported comparable nanowire coefficient. The model-based film thermal conductivity is kappa = approximately 20 mW/(m K). Across lengths ell = 1-50 micrometers, I_r matches neither ell^(-1) of pure longitudinal conduction nor ell^0 of pure interfacial cooling; a two-channel model reproduces I_r(ell) with crossover length ell* = approximately 3.5 micrometers predicted, to an O(1) prefactor, from Sigma_K and kappa.

Commercial complementary metal oxide semiconductor (CMOS) processes routinely use transition-metal nitrides and silicides as diffusion barriers, liner metals and passive components.[1, 2, 3] Several become superconducting at cryogenic temperatures[4, 5, 6, 7]—a property incidental to processes designed for room-temperature operation, in which the films are chosen for work function, adhesion and sheet resistance. Recently Swift et al. used one such film as a circuit element.[8] They exploited the high kinetic inductance (≈ 1 nH per square) of the titanium nitride (TiN) sub-film inside a polycrystalline-silicon resistor (poly-resistor) of a GlobalFoundries 22 nm fully depleted silicon-on-insulator (FD-SOI) process. The resulting “superinductor” is four orders of magnitude more compact in area than the spiral inductors conventionally available on such a process. It is used to read out a silicon metal–oxide–semiconductor (MOS) quantum dot in an integrated radio-frequency single-electron transistor, establishing the film as a usable circuit material for superconductor-enhanced CMOS circuits.

The significance extends beyond passive inductors. The disordered superconductors of exactly this class (e.g., TiN, NbN, NbTiN, WSi) are the workhorse materials of superconducting nanowire single-photon detectors[9, 10, 11] and of nanowire logic and memory.[12, 13, 14] They are chosen for the same properties that make them good superconductors: high normal-state resistivity, short coherence length and low electron diffusivity. A CMOS process that natively contains such a film is therefore a candidate platform for monolithically integrating nanowire detectors and superconducting logic and memory circuits with their cryogenic CMOS readout electronics.

The environment of the film matters for that prospect. The front end of line (FEOL) contains the transistor-level structures at the wafer surface—active silicon, isolation, gate stack and silicide—while the back end of line (BEOL) is the multilevel metal-and-dielectric interconnect built on top of it. The TiN layer studied here is a few-nanometer-thick film within the body of a poly-resistor.[15] It lies on shallow-trench-isolation oxide and is capped by doped polysilicon, silicided at the contact ends. Above it the pre-metal dielectric separates the FEOL from the first metal level of the BEOL stack.[8] The TiN layer is buried in the FEOL beneath the BEOL stack, encapsulated in amorphous dielectric with oxide below and metal interconnects and density fill above (Figure 1).

Cross-section along the wire, not to scale

Figure 1. Cross-section along the wire, not to scale. The TiN film lies on the shallow-trench-isolation (STI) oxide, is capped by doped polysilicon with silicide over the two contact ends, and is buried under the pre-metal dielectric (PMD), the BEOL metal levels (M1, M2), and inter-metal dielectric (IMD); isolated M1/M2 blocks are floating density fill, and dots mark where the stack continues. A hot spot at ThsT_{\mathrm{hs}} loses heat along the film to the contacts (set by κ\kappa) and vertically into the surrounding layers (set by ΣK\Sigma_K). ℓ\ell is the length between the silicide pads; WW runs into the page.

The buried environment makes the electrothermal behavior of the TiN film uncertain. In superconducting nanowire detectors, photon absorption can nucleate a transient resistive region whose subsequent evolution (e.g., collapse, growth, or latching) is governed by electrothermal feedback,[11, 16, 17] related feedback is also exploited in nanowire logic and memory.[12, 13, 14] The retrapping current IrI_r—required to sustain an existing normal (resistive) domain—lies below the switching current IswI_{\mathrm{sw}} at which a cold superconducting wire becomes resistive. Whereas IswI_{\mathrm{sw}} is governed by superconducting-instability mechanisms such as depairing, phase slips, or vortex entry, IrI_r is determined by heat removal. Whether a film buried close to the BEOL metal stack can sustain a hot spot is therefore not obvious from its geometry alone and requires experimental characterization together with electrothermal modeling.

Heat can leave the hot spot by two routes, which are considered separately. A retrapping current IrcondI_r^{\mathrm{cond}} is defined for longitudinal conduction along the film to the contacts, while a distinct retrapping current IrintI_r^{\mathrm{int}} is defined for interfacial cooling from the film into the surrounding stack. Unadorned IrI_r denotes the measured retrapping current. Because the resistive hot spot is not isothermal, its interior can be warmer than TcT_c even though the normal–superconducting boundaries occur near TcT_c. Accordingly, ThsT_{\mathrm{hs}} is treated as an effective hot-spot temperature and fitted to the measured Ir(Tb)I_r(T_b),[18] rather than being fixed at TcT_c.

First, consider longitudinal conduction.[19] In the symmetric one-dimensional limit, a normal domain of length ℓhs=xℓ\ell_{\mathrm{hs}}=x\ell, occupying a fraction xx of the wire, has resistance xRnxR_n, while the remaining superconducting material forms two arms of length (1−x)ℓ/2(1-x)\ell/2 that conduct to contacts held at the bath temperature TbT_b. Each arm conducts 2κA(Ths−Tb)/[(1−x)ℓ]2\kappa A(T_{\mathrm{hs}}-T_b)/[(1-x)\ell] away from the hot spot, so the two together remove 4κA(Ths−Tb)/[(1−x)ℓ]4\kappa A(T_{\mathrm{hs}}-T_b)/[(1-x)\ell]; equating this to the Joule power I2xRnI^2xR_n gives x(1−x)=4κA(Ths−Tb)/(ℓ2Rn)x(1-x)=4\kappa A(T_{\mathrm{hs}}-T_b)/(\ell^2R_n). Since x(1−x)x(1-x) is maximized at x=1/2x=1/2, where it equals 1/41/4, the smallest current that can sustain a stationary hot spot is

Ircond=4κA(Ths−Tb)ℓRn=4Wℓκt(Ths−Tb)R□.(1)I_r^{\mathrm{cond}}=4\sqrt{\frac{\kappa A(T_{\mathrm{hs}}-T_b)}{\ell R_n}}=\frac{4W}{\ell}\sqrt{\frac{\kappa t(T_{\mathrm{hs}}-T_b)}{R_{\square}}}. \tag*{(1)}

where A=WtA=Wt is the wire cross-sectional area, Rn=R□ℓ/WR_n=R_{\square}\ell/W is the normal-state resistance of the full wire, R□R_{\square} is the sheet resistance, and κ\kappa is an effective along-film thermal conductivity over the temperature range of the superconducting arms (Figure 1). If κ\kappa varies with temperature, the factor κ(Ths−Tb)\kappa(T_{\mathrm{hs}}-T_b) in Eq. (1) is replaced by ∫TbThsκ(T) dT\int_{T_b}^{T_{\mathrm{hs}}}\kappa(T)\,\mathrm{d}T.[19] Equation (1) therefore represents the constant- or temperature-averaged-κ\kappa limit of longitudinal cooling, rather than a universal temperature dependence for heat conduction along the wire.

Second, vertical phonon-mediated cooling removes heat from the film into the surrounding stack,[18, 20] with an effective areal flux ΣK(Tn−Tbn)\Sigma_K(T^n-T_b^n) and n=4n=4. For a normal segment of length ℓhs\ell_{\mathrm{hs}}, both the Joule power I2Rnℓhs/WI^2R_n\ell_{\mathrm{hs}}/W and the vertical cooling power ΣKWℓhs(Ths4−Tb4)\Sigma_KW\ell_{\mathrm{hs}}(T_{\mathrm{hs}}^4-T_b^4) scale with ℓhs\ell_{\mathrm{hs}}; the segment length therefore cancels, giving:

Irint=WΣK(Ths4−Tb4)R□.(2)I_r^{\mathrm{int}}=W\sqrt{\frac{\Sigma_K(T_{\mathrm{hs}}^4-T_b^4)}{R_{\square}}}. \tag*{(2)}

Heat taking this route leaves the film through two parallel vertical paths: downward through the amorphous dielectric toward the silicon handle, and upward through the doped polysilicon cap. The downward path is expected to dominate, as the upward one meets the poor sub-kelvin conductivity of the polysilicon and pre-metal dielectric and ends largely in floating metal fill rather than a dedicated heat sink. As used here, ΣK\Sigma_K is therefore an effective stack-level coefficient per unit TiN area: it folds several series bottlenecks—multiple interfaces, transport through amorphous dielectric, and the absence of a dedicated upper heat sink—into a single number, and is not the microscopic thermal-boundary conductance of any individual interface.

The TiN films are found in poly-resistor structures in a GlobalFoundries 22 nm FD-SOI process.[8] The wires studied here have Tc=1.2 KT_c = 1.2\ \mathrm{K} and an assumed film thickness t=3 nmt = 3\ \mathrm{nm}.[15, 21] Four-point II–VV measurements were performed on two devices of width W=0.5W = 0.5 and 1.0 μm1.0\ \mu\mathrm{m}, both of length ℓ=50 μm\ell= 50\ \mu\mathrm{m}, on one die, at bath temperatures ranging from 0.150.15 to 1.19 K1.19\ \mathrm{K}. The measured normal-state sheet resistance is R□=492 Ω/□R_{\square} = 492\ \Omega/\square.

IswI_{sw} and IrI_r are defined as the largest positive currents within the zero-voltage band on the up-sweep and down-sweep branches respectively; the band is set to 2% of the full-scale voltage. IrI_r is reproducible sweep to sweep. IswI_{sw} is not: it varies because some individual sweeps switch prematurely at lower currents. The IswI_{sw} values quoted below are therefore the maxima measured over the low-temperature plateau, Tb≤0.6 KT_b \le0.6\ \mathrm{K}.

Figure 2 shows that the wires enter a self-sustaining resistive state, hysteretic between the switching and retrapping currents. The narrow device gives Isw=1.065 μAI_{sw} = 1.065\ \mu\mathrm{A} and Ir=0.109 μAI_r = 0.109\ \mu\mathrm{A}, a ratio of 9.8. The wide device gives Isw=1.615 μAI_{sw} = 1.615\ \mu\mathrm{A} and Ir=0.219 μAI_r = 0.219\ \mu\mathrm{A}, a ratio of 7.4.

Figure 2. Hysteretic current–voltage characteristics of TiN nanowires, ℓ=50 μm\ell= 50\ \mu\mathrm{m}, at Tb=0.15 KT_b = 0.15\ \mathrm{K}. (a) W=0.5 μmW = 0.5\ \mu\mathrm{m}: the up-sweep switches at Isw=1.055 μAI_{sw} = 1.055\ \mu\mathrm{A} and the down-sweep retraps at Ir=0.109 μAI_r = 0.109\ \mu\mathrm{A}. (b) W=1.0 μmW = 1.0\ \mu\mathrm{m}: Isw=1.590 μAI_{sw} = 1.590\ \mu\mathrm{A}, Ir=0.219 μAI_r = 0.219\ \mu\mathrm{A}. Vertical dotted lines mark the extracted currents. Each IswI_{sw} above is that of the single trace shown; because individual sweeps scatter downward through premature switching, the text instead quotes the maximum over the Tb≤0.6 KT_b \le0.6\ \mathrm{K} plateau. Thresholds are taken on the positive branch; IrI_r agrees between polarities to within 4%.

Averaged over the measured bath-temperature range, Tb=0.15 KT_b = 0.15\ \mathrm{K}–1.19 K1.19\ \mathrm{K}, the measured ratio Ir(1.0 μm)/Ir(0.5 μm)I_r(1.0\ \mu\mathrm{m})/I_r(0.5\ \mu\mathrm{m}) is 2.036±0.0342.036 \pm0.034, consistent with the linear scaling in WW shared by IrcondI_r^{\mathrm{cond}} and IrintI_r^{\mathrm{int}}. The temperature dependence, by contrast, differs between the two models: both give Ir∝(Thsn−Tbn)1/2I_r \propto(T_{hs}^{n} - T_b^{n})^{1/2}, with n=1n = 1 for the constant-κ\kappa longitudinal-conduction limit and n=4n = 4 for the conventional interfacial phonon-radiation model. Fitting Ir=C(Thsn−Tbn)1/2I_r = C(T_{hs}^{n} - T_b^{n})^{1/2} with CC and ThsT_{hs} both free parameters decisively rejects the n=1n = 1 constant-κ\kappa longitudinal-conduction limit, while n=4n = 4 (interfacial phonon radiation) follows the measurements over the full bath-temperature range, Tb=0.15 KT_b = 0.15\ \mathrm{K}–1.19 K1.19\ \mathrm{K} (Figure 3). Allowing the exponent itself to float returns n=4.81(24)n = 4.81(24) and 4.96(17)4.96(17) for the two devices. An explicit n=5n = 5 fit follows the temperature dependence better than n=4n = 4 (RMS residual 0.9–1.5% versus 1.4–1.8%). Nevertheless, the two parameterizations give nearly equal cooling fluxes that differ by less than 3% at Tb=0.15 KT_b = 0.15\ \mathrm{K}, indicating that the inferred magnitude of heat removal is insensitive to this choice. The n=4n = 4 form is retained as the conventional phonon-radiation parameterization for direct comparison with Ref. 18 and the acoustic- and diffuse-mismatch benchmarks. Because the TiN film is ultrathin, reduced-dimensional phonon modes could in principle modify the temperature dependence of the conventional three-dimensional phonon-radiation model. The fitted exponent n≈4.8n \approx4.8–5.05.0 shows no evidence for a reduced-exponent regime over the measured temperature range, although the exponent alone does not uniquely identify the microscopic thermal bottleneck.

Figure 3. IrI_r normalized to its 0.15 K value against TbT_b, for both ℓ=50 μm\ell= 50\ \mu\mathrm{m} devices, with Ir=C(Thsn−Tbn)1/2I_r = C\left(T_{hs}^{n} - T_b^{n}\right)^{1/2} fitted in the prefactor CC and the hot-spot temperature ThsT_{hs} and normalized at the same point, for n=1n = 1 (constant-κ\kappa longitudinal-conduction limit), 2, and 4 (conventional interfacial phonon-radiation model). Among the fixed exponents shown, n=4n = 4 best follows the data over the full range.

Setting IrintI_r^{\mathrm{int}} equal to the measured IrI_r and rearranging Eq. (2) gives

ΣK=Ir2R□W2(Ths4−Tb4).\Sigma_K = \frac{I_r^2 R_{\square}}{W^2\left(T_{hs}^4 - T_b^4\right)}.

Here, ThsT_{hs} is fitted independently for the two devices in the n=4n = 4 model, yielding Ths=1.304 KT_{hs} = 1.304\ \mathrm{K} and 1.314 K1.314\ \mathrm{K} for W=0.50 μmW = 0.50\ \mu\mathrm{m} and 1.00 μm1.00\ \mu\mathrm{m}, respectively. Their mean, Ths=1.309 KT_{hs} = 1.309\ \mathrm{K}, is then held fixed when ΣK\Sigma_K is evaluated point-by-point at each measured TbT_b for both devices. The 62 points below TcT_c yield 7.9(3) W/(m2K4)7.9(3)\ \mathrm{W}/(\mathrm{m}^2\mathrm{K}^4), with no resolved systematic dependence on temperature or width (Figure 4). The parenthetical uncertainty is the sample standard deviation of the point-by-point values, while the corresponding standard error is approximately 0.04 W/(m2K4)0.04\ \mathrm{W}/(\mathrm{m}^2\mathrm{K}^4).

Figure 4. Effective cooling (Kapitza) coefficient ΣK\Sigma_K extracted point by point from Eq. (2) for both ℓ=50 μm\ell=50\,\mu\mathrm{m} devices. The 62 extractions below TcT_c give one temperature-independent value, 7.9(3) W/(m2 K4)7.9(3)\,\mathrm{W}/(\mathrm{m}^2\,\mathrm{K}^4) (solid line), 31(21)×\times below the tabulated acoustic- (diffuse-) mismatch value for a single clean TiN/SiO2_2 interface [18].

Equivalently, comparison of the fitted form

Ir=C(Ths4−Tb4)1/2,I_r=C\left(T_{hs}^{4}-T_b^{4}\right)^{1/2},

with Eq. (2) gives

C=WΣKR□,ΣK=C2R□W2.C=W\sqrt{\frac{\Sigma_K}{R_{\square}}},\qquad\Sigma_K=\frac{C^2R_{\square}}{W^2}.

The fitted prefactors independently give ΣK=7.90 W/(m2 K4)\Sigma_K=7.90\,\mathrm{W}/(\mathrm{m}^2\,\mathrm{K}^4) and 7.88 W/(m2 K4)7.88\,\mathrm{W}/(\mathrm{m}^2\,\mathrm{K}^4) for the two devices, consistent with the point-by-point mean. Thus, the point-by-point extraction also provides a consistency check that the effective cooling coefficient is independent of TbT_b and WW.

For TiN on SiO2_2, Ref. 18 tabulates ideal thermal-boundary coefficients of 243 W/(m2 K4)243\,\mathrm{W}/(\mathrm{m}^2\,\mathrm{K}^4) for the acoustic-mismatch model[22] and 164 W/(m2 K4)164\,\mathrm{W}/(\mathrm{m}^2\,\mathrm{K}^4) for the diffuse-mismatch model[23]. These single-interface values exceed the effective cooling coefficient extracted for the buried CMOS stack by factors of 31(21)×\times for the acoustic- (diffuse-) mismatch model. A fairer clean-limit benchmark for a buried film is the dominant downward stack of Figure 1: combining the thermal resistances of the layers and interfaces in series gives ideal effective coefficients of 62.8 W/(m2 K4)62.8\,\mathrm{W}/(\mathrm{m}^2\,\mathrm{K}^4) (acoustic-mismatch) and 51.4 W/(m2 K4)51.4\,\mathrm{W}/(\mathrm{m}^2\,\mathrm{K}^4) (diffuse-mismatch), corresponding to factors of 8.0(6.5)×\times above the measured ΣK\Sigma_K for the acoustic- (diffuse-) mismatch model. Dane et al. fitted ΣK\Sigma_K for 15 single-layer NbN nanowires on six substrates and obtained values consistent with the acoustic-mismatch prediction [18]. The buried CMOS TiN film measured here lies a factor of 12 below the lowest effective cooling coefficient among those 15 directly comparable wires.

What has been established so far applies to long wires. For ℓ=50 μm\ell=50\,\mu\mathrm{m}, IrI_r is set by vertical phonon-mediated cooling into the surrounding stack (Figure 4). Conduction to the contacts scales with the cross-section W×tW\times t while vertical cooling scales with the footprint W×ℓW\times\ell, so their relative importance scales primarily with t/ℓt/\ell, making wire length the experimental control parameter for the crossover. The data that follow instead vary ℓ\ell at fixed TbT_b, spanning ℓ=1\ell=1–50 μm50\,\mu\mathrm{m} at W=0.50W=0.50 and 1.00 μm1.00\,\mu\mathrm{m}, measured at Tb=0.15 KT_b=0.15\,\mathrm{K}[24].

Where the crossover falls is set by the film thermal conductivity. It is estimated using the expressions of Ref. 18, combining a Bardeen–Rickayzen–Tewordt electronic contribution[25] with the normal-state Wiedemann–Franz conductivity and a Casimir-limit phonon contribution. Given a measured sheet resistance R□R_{\square}, the inferred normal-state resistivity is ρn=R□t\rho_n=R_{\square}t, which is in turn used for the Wiedemann–Franz electronic thermal conductivity κn=L0T/ρn=L0T/R□t\kappa_n=L_0T/\rho_n=L_0T/R_{\square}t. For t=3 nmt=3\,\mathrm{nm}, the resulting model-based estimate is κ(Tc)=19.9 mW/(m K)\kappa(T_c)=19.9\,\mathrm{mW}/(\mathrm{m\,K}), of the order measured in other superconducting thin films at sub-Kelvin temperatures[26].

An independent retrapping-current analysis by Swift, performed assuming t∼5 nmt\sim5\,\mathrm{nm}, yielded κ≈9 mW/(m K)\kappa\approx9\,\mathrm{mW}/(\mathrm{m\,K})[24]. Rescaling that estimate to t=3 nmt=3\,\mathrm{nm} while keeping the measured R□R_{\square} gives κ≈15 mW/(m K)\kappa\approx15\,\mathrm{mW}/(\mathrm{m\,K}), of the same order as the present model-based value. Since the electronic contribution dominates, the product κt\kappa t entering Eqs. (3) and (5) is nearly independent of the assumed film thickness.

Treating longitudinal conduction and vertical cooling together requires the introduction of the linearized thermal healing length η\eta[27, 28], which is the distance heat spreads along the film before it escapes vertically into the surrounding stack:

η=κtnΣKThsn−1=0.92 μm,(3)\eta=\sqrt{\frac{\kappa t}{n\Sigma_K T_{hs}^{n-1}}}=0.92\,\mu\mathrm{m}, \tag*{(3)}

with the measured ΣK\Sigma_K, exponent nn, and computed κ\kappa. A hot spot much longer than η\eta loses its heat vertically into the surrounding stack and never reaches the contacts. One shorter than a few η\eta loses an appreciable fraction longitudinally, out through the contacts at either end, and needs more current to sustain itself. Ir(ℓ)I_r(\ell) therefore increases at shorter lengths and plateaus at longer ones. Combining the two limiting loss terms into a single compact interpolation gives the two-channel form IrtotI_r^{\mathrm{tot}}:

Irtot(ℓ)=WΣK(Ths4−Tb4)R□1+(ℓ∗ℓ)2,(4)I_r^{\mathrm{tot}}(\ell)=W\sqrt{\frac{\Sigma_K\left(T_{hs}^{4}-T_b^{4}\right)}{R_{\square}}}\sqrt{1+\left(\frac{\ell^*}{\ell}\right)^2}, \tag*{(4)}
ℓ∗=2κt(Ths−Tb)ΣK(Ths4−Tb4).(5)\ell^*=2\sqrt{\frac{\kappa t\left(T_{hs}-T_b\right)}{\Sigma_K\left(T_{hs}^{4}-T_b^{4}\right)}}. \tag*{(5)}

with ℓ∗=3.8 η=3.5 μm\ell^*=3.8\,\eta=3.5\,\mu\mathrm{m} estimated from the ΣK\Sigma_K and κ\kappa fixed above. Equation (4) is phenomenological: it lumps the hot spot into a single temperature ThsT_{hs} rather than solving for the profile along the wire, and the prefactor in Eq. (5) is a heuristic O(1)O(1) constant carried over from the single-channel hot-spot profile, so ℓ∗\ell^{*} is an order-of-magnitude prediction. For ℓ≫ℓ∗\ell\gg\ell^{*} it returns the length-independent IrintI_{\mathrm{r}}^{\mathrm{int}} of Eq. (2). For ℓ≪ℓ∗\ell\ll\ell^{*} the contacts dominate and it recovers the ℓ−1\ell^{-1} scaling of Eq. (1), with an amplitude that depends on the hot-spot profile assumed.

Equation (4) is applied to these structures in Figure 5. The data fall steeply over the first decade of length, roll over at a few micrometres, and are flat by 30 μm30\,\mu\mathrm{m}; the two widths run parallel, separated by their width ratio. Equation (4) is fitted to each width with two free parameters, the plateau current I∞I_{\infty} and the crossover length ℓ∗\ell^{*}. The fits return I∞=0.105I_{\infty}=0.105 and 0.211 μA0.211\,\mu\mathrm{A} and ℓ∗=4.6\ell^{*}=4.6 and 4.4 μm4.4\,\mu\mathrm{m}. The plateau currents agree with the I∞=0.108I_{\infty}=0.108 and 0.217 μA0.217\,\mu\mathrm{A} obtained from the 50 μm50\,\mu\mathrm{m} measurements, and the fitted ℓ∗\ell^{*} agrees with the predicted ℓ∗≈3.5 μm\ell^{*}\approx3.5\,\mu\mathrm{m} to within the O(1)O(1) prefactor of Eq. (5). The unusually long crossover follows directly from the weak vertical cooling: at fixed κt\kappa t, ThsT_{\mathrm{hs}}, and TbT_{\mathrm{b}}, Eq. (5) gives ℓ∗∝ΣK−1/2\ell^{*}\propto\Sigma_{\mathrm{K}}^{-1/2}. Three cooling coefficients therefore set three crossover lengths. The measured ΣK=7.9 W/(m2 K4)\Sigma_{\mathrm{K}}=7.9\,\mathrm{W}/(\mathrm{m}^{2}\,\mathrm{K}^{4}) gives ℓ∗=3.5 μm\ell^{*}=3.5\,\mu\mathrm{m}. The ideal buried stack, 62.8(51.4) W/(m2 K4)62.8(51.4)\,\mathrm{W}/(\mathrm{m}^{2}\,\mathrm{K}^{4}), gives 1.2(1.4) μm1.2(1.4)\,\mu\mathrm{m}. A single clean interface, 243(164) W/(m2 K4)243(164)\,\mathrm{W}/(\mathrm{m}^{2}\,\mathrm{K}^{4}), gives 0.6(0.8) μm0.6(0.8)\,\mu\mathrm{m}, quoted throughout for the acoustic- (diffuse-) mismatch model. This benchmark is derived here from Eq. (5), not a crossover length measured by Ref. 18. An otherwise similar film with conventional strong vertical thermal coupling would therefore reach its Ir(ℓ)I_{\mathrm{r}}(\ell) rollover at a wire length about six times shorter: stronger vertical cooling removes heat within a shorter healing length η\eta, which shortens ℓ∗\ell^{*} and moves the knee in Ir(ℓ)I_{\mathrm{r}}(\ell) to shorter wires. The 2-D substrate model gives a characteristic lateral spreading length of approximately 234 μm234\,\mu\mathrm{m}, well above the 11 to 50 μm50\,\mu\mathrm{m} wire-length range studied here. The substrate therefore acts as an approximately uniform thermal bath over the measured devices, and the few-micrometer crossover is instead set by the competition between longitudinal conduction along the TiN film and vertical cooling into the surrounding stack.

Figure 5. Length dependence of IrI_{\mathrm{r}} across the TiN test structures of Ref. 24, measured at Tb=0.15 KT_{\mathrm{b}}=0.15\,\mathrm{K}. Markers are measured data. Solid curves are Eq. (4) with amplitude and ℓ∗\ell^{*} fitted per width; dotted lines are the corresponding ℓ≫ℓ∗\ell\gg\ell^{*} interfacial floors.

The extracted ΣK\Sigma_{\mathrm{K}} depends on the modeling conventions used to reduce the data, chiefly the hot-spot-temperature assignment. The extraction relation constrains only the product ΣK(Ths4−Tb4)\Sigma_{\mathrm{K}}(T_{\mathrm{hs}}^{4}-T_{\mathrm{b}}^{4}), so the inferred coefficient scales inversely with the assumed thermal driving term. The headline value ΣK=7.9(3) W/(m2 K4)\Sigma_{\mathrm{K}}=7.9(3)\,\mathrm{W}/(\mathrm{m}^{2}\,\mathrm{K}^{4}) follows the fitted-ThsT_{\mathrm{hs}} convention of Ref. 18, for which the n=4n=4 fit of Figure 3 gives Ths≈1.31 KT_{\mathrm{hs}}\approx1.31\,\mathrm{K}, slightly above Tc=1.2 KT_{\mathrm{c}}=1.2\,\mathrm{K}. Fixing Ths=TcT_{\mathrm{hs}}=T_{\mathrm{c}} and refitting the two devices instead gives ΣK≈12 W/(m2 K4)\Sigma_{\mathrm{K}}\approx12\,\mathrm{W}/(\mathrm{m}^{2}\,\mathrm{K}^{4}).

Wire width constitutes a separate systematic uncertainty because ΣK∝W−2\Sigma_{\mathrm{K}}\propto W^{-2}; for example, a 10%10\% reduction in the assumed width would increase the inferred coefficient by approximately 23%23\%. The reported extraction nevertheless uses the drawn device widths throughout and does not assume a narrower effective current-carrying channel. Thus, the same retrapping data yield ΣK=7.9\Sigma_{\mathrm{K}}=7.9 or approximately 12 W/(m2 K4)12\,\mathrm{W}/(\mathrm{m}^{2}\,\mathrm{K}^{4}), depending on the hot-spot-temperature convention, reflecting modeling uncertainty rather than distinct physical cooling coefficients. The latter value remains approximately 20(14)×20(14)\times below the acoustic- (diffuse-) mismatch value for a single clean TiN/SiO2_{2} interface, compared with 31(21)×31(21)\times for the headline fitted-ThsT_{\mathrm{hs}} value. The inferred coefficients also remain below the range reported for NbN by Ref. 18. Electron-phonon nonequilibrium was also assessed using the two-temperature model of Ref. 18, in which the electron-phonon heat flux is tΣe−ph(Te5−Tph5)t\Sigma_{\mathrm{e-ph}}(T_{\mathrm{e}}^{5}-T_{\mathrm{ph}}^{5}), with Σe−ph\Sigma_{\mathrm{e-ph}} parameterized following Refs. 18 and 29. Using an order-of-magnitude TiN coupling estimate together with the measured ΣK\Sigma_{\mathrm{K}}, t=3 nmt=3\,\mathrm{nm}, and the fitted Ths≈1.309 KT_{\mathrm{hs}}\approx1.309\,\mathrm{K} gives Te−Tph≈31 mKT_{\mathrm{e}}-T_{\mathrm{ph}}\approx31\,\mathrm{mK} and a differential electron-phonon conductance approximately ten times the film-to-bath conductance. These numerical values are estimates for the present film. Published TiN electron-phonon relaxation times vary substantially,[30] so this calculation is used only as an order-of-magnitude consistency check.

In summary, TiN nanowires buried in the FEOL beneath the BEOL stack of a commercial CMOS process sustain self-heated hot spots with hysteresis ratios of 7.47.4–9.89.8. Transient hot-spot formation underpins nanowire detection; controlled hysteresis, nanowire logic and memory. Retrapping in the long wires is governed by vertical phonon-mediated cooling into the surrounding stack rather than by longitudinal conduction to the contacts. A single effective cooling coefficient, ΣK=7.9(3) W/(m2 K4)\Sigma_{\mathrm{K}}=7.9(3)\,\mathrm{W}/(\mathrm{m}^{2}\,\mathrm{K}^{4}), describes the data under the conventional n=4n=4 parameterization. Over a fiftyfold range of length, IrI_{\mathrm{r}} follows neither the ℓ−1\ell^{-1} longitudinalconduction limit nor the ℓ0\ell^{0} vertical-cooling limit, but crosses between them at a few micrometres (fitted 4.4 to 4.6 µm), near the two-channel crossover ℓ∗≈3.5\ell^{\ast} \approx3.5 µm predicted from the measured ΣK\Sigma_{K} and the computed κ\kappa. The unusually small effective ΣK\Sigma_{K} therefore explains both the low long-wire retrapping current and the extended length scale over which contact cooling remains important.

Acknowledgments

This work was supported by the U.S. Department of Energy, Office of Science, Advanced Scientific Computing Research (ASCR), High Energy Physics (HEP) and Basic Energy Sciences (BES), under Contract No. DE-AC02-06CH11357 at Argonne. Work performed at the Advanced Photon Source, a U.S. Department of Energy Office of Science User Facilities (SUF). Work at University of Chicago is supported by the Divisions of Chemistry (CHE) and Materials Research (DMR), National Science Foundation, under grant numbers NSF/CHE-1834750 and NSF/CHE-2335833. M.F.G.Z. acknowledges a UKRI Future Leaders Fellowship [MR/V023284/1]. A.G.-S. acknowledges an Industrial Fellowship from the Royal Commission for the Exhibition of 1851. This work was primarily supported by the CMOS+X project funded by DOE’s Genesis Mission.

Author Declarations

Conflict of Interest

The authors have no conflicts to disclose.

Author Contributions

Mohamed Gharib: Methodology (equal); Software (equal); Formal analysis (lead); Writing – original draft (equal); Writing – review & editing (equal); Visualization (lead). Leonid Popryho: Software (equal); Validation (lead); Writing – review & editing (equal). Salma Abdelzaher: Writing – review & editing (equal). Tejas Guruswamy: Methodology (equal); Writing – review & editing (equal); Supervision (equal); Funding acquisition (equal). Tomas Polakovic: Methodology (equal); Writing – review & editing (equal); Funding acquisition (equal). Umeshkumar Patel: Methodology (equal); Writing – review & editing (equal); Funding acquisition (equal). Orlando Quaranta: Methodology (equal); Writing – review & editing (equal); Funding acquisition (equal). Thomas Cecil: Methodology (equal); Writing – review & editing (equal); Funding acquisition (equal). Clarence Chang: Methodology (equal); Writing – review & editing (equal); Funding acquisition (equal). Yu-Sheng Chen: Investigation (equal); Writing – review & editing (equal). Thomas H. Swift: Formal analysis (lead); Investigation (equal); Data curation (lead); Writing – review & editing (equal). Grayson M. Noah: Investigation (equal); Writing – review & editing (equal). Alberto Gomez-Saiz: Investigation (equal); Writing – review & editing (equal); Visualization (equal). John J. L. Morton: Investigation (equal); Writing – review & editing (equal); Visualization (equal). M. Fernando Gonzalez-Zalba: Investigation (equal); Writing – review & editing (equal); Visualization (equal). Antonino Miceli: Conceptualization (equal); Writing – original draft (equal); Writing – review & editing (equal); Supervision (equal); Project administration (equal); Funding acquisition (equal). Inna Partin-Vaisband: Conceptualization (equal); Writing – review & editing (equal); Supervision (equal); Project administration (equal); Funding acquisition (equal).

Data Availability

The data that support the findings of this study are available from the corresponding authors upon reasonable request.

References

  1. [1]M. Wittmer, “TiN and TaN as diffusion barriers in metallizations to silicon semiconductor devices,” Applied Physics Letters 36, 456–458 (1980).DOI
  2. [2]C. Lavoie, F. M. d’Heurle, C. Detavernier, and C. Cabral, Jr., “Towards implementation of a nickel silicide process for CMOS technologies,” Microelectronic Engineering 70, 144–157 (2003).DOI
  3. [3]T. Guruswamy, L. Gades, A. Miceli, U. Patel, and O. Quaranta, “Beamline spectroscopy of integrated circuits with hard X-ray transition edge sensors at the Advanced Photon Source,” IEEE Transactions on Applied Superconductivity 31, 1–5 (2021).DOI
  4. [4]M. R. Vissers, J. Gao, D. S. Wisbey, D. A. Hite, C. C. Tsuei, A. D. Corcoles, M. Steffen, and D. P. Pappas, “Low loss superconducting tantalum nitride coplanar waveguide resonators,” Applied Physics Letters 97, 232509 (2010).DOI
  5. [5]S.-P. Chiu, C. C. Tsuei, S.-S. Yeh, F.-C. Zhang, S. Kirchner, and J.-J. Lin, “Observation of triplet superconductivity in CoSi₂/TiSi₂ heterostructures,” Sci. Adv. 7, eabg6569 (2021).DOI
  6. [6]N. R. Groll, J. A. Klug, C. Cao, S. Altin, H. Claus, N. G. Becker, J. F. Zadzinski, M. J. Pellin, and T. Proslier, “Tunneling spectroscopy of superconducting MoN and NbTiN grown by atomic layer deposition,” Applied Physics Letters 104, 092602 (2014).DOI
  7. [7]S.-P. Chiu, C.-J. Wang, Y.-C. Lin, S.-T. Tu, S. K. Sahu, R.-T. Wang, C.-Y. Wu, S.-S. Yeh, S. Kirchner, and J.-J. Lin, “Electric and superconducting properties of CoSi₂ films on silicon — an unconventional superconductor with technological potential,” Chinese Journal of Physics 90, 348–363 (2024).DOI
  8. [8]T. H. Swift, F. Olivieri, G. Azipurua-Iraola, J. Kirkman, G. M. Noah, M. de Kruijf, F.-E. von Horstig, A. Gomez-Saiz, J. J. L. Morton, and M. F. Gonzalez-Zalba, “A superconductor in a deep sub-micron integrated circuit,” arXiv:2507.13202 (2025), 10.48550/arXiv.2507.13202.arxiv.org/abs/2507.13202
  9. [9]G. N. Gol’tsman, O. Okunev, G. Chulkova, A. Lipatov, A. Semenov, K. Smirnov, B. Voronov, A. Dzardanov, C. Williams, and R. Sobolewski, “Picosecond superconducting single-photon optical detector,” Applied Physics Letters 79, 705–707 (2001).DOI
  10. [10]A. E. Lita, A. J. Miller, and S. W. Nam, “Counting near-infrared single-photons with 95% efficiency,” Opt. Express 16, 3032–3040 (2008).pubmed.ncbi.nlm.nih.gov/18542389
  11. [11]C. M. Natarajan, M. G. Tanner, and R. H. Hadfield, “Superconducting nanowire single-photon detectors: physics and applications,” Supercond. Sci. Technol. 25, 063001 (2012).DOI
  12. [12]O. Quaranta, S. Marchetti, N. Martucciello, S. Pagano, M. Ejrnaes, R. Cristiano, and C. Nappi, “Superconductive three-terminal amplifier/discriminator,” IEEE Transactions on Applied Superconductivity 19, 367–370 (2009).DOI
  13. [13]A. N. McCaughan and K. K. Berggren, “A superconducting nanowire three-terminal electrothermal device,” Nano Letters 14, 5748–5753 (2014).
  14. [14]R. Baghdadi, J. P. Allmaras, B. A. Butters, A. E. Dane, S. Iqbal, A. N. McCaughan, E. A. Toomey, Q.-Y. Zhao, A. G. Kozorezov, and K. K. Berggren, “Multilayered heater nanocryotron: A superconducting-nanowire-based thermal switch,” Phys. Rev. Appl. 14, 054011 (2020).DOI
  15. [15]D. James, “High-k/metal gates in the 2010s,” in 25th Annual SEMI Advanced Semiconductor Manufacturing Conference (ASMC 2014) (2014) pp. 431–438.DOI
  16. [16]A. J. Kerman, J. K. W. Yang, R. J. Molnar, E. A. Dauler, and K. K. Berggren, “Electrothermal feedback in superconducting nanowire single-photon detectors,” Phys. Rev. B 79, 100509 (2009).DOI
  17. [17]A. J. Annunziata, O. Quaranta, D. F. Santavicca, A. Casaburi, L. Frunzio, M. Ejrnaes, M. J. Rooks, R. Cristiano, S. Pagano, A. Frydman, and D. E. Prober, “Reset dynamics and latching in niobium superconducting nanowire single-photon detectors,” J. Appl. Phys. 108, 084507 (2010).DOI
  18. [18]A. Dane, J. Allmaras, D. Zhu, M. Qnen, M. Colangelo, R. Baghdadi, N. Baškent, and K. K. Berggren, “Self-heating hotspots in superconducting nanowires cooled by phonon black-body radiation,” Nat. Commun. 13, 5429 (2022).DOI
  19. [19]M. Tinkham, J. U. Free, C. N. Lau, and N. Marković, “Hysteretic I–V curves of superconducting nanowires,” Phys. Rev. B 68, 134515 (2003).DOI
  20. [20]W. J. Skocpol, M. R. Beasley, and M. Tinkham, “Self-heating hotspots in superconducting thin-film microbridges,” J. Appl. Phys. 45, 4054–4066 (1974).DOI
  21. [21]G. Aizpurua-Iraola, T. H. Swift, F.-E. von Horstig, D. Prete, J. Kirkman, G. M. Noah, F. Olivieri, A. Gomez-Saiz, and M. F. Gonzalez-Zalba, “A superconducting phase transition single-electron transistor,” (2026), arXiv:2608.27045 [quant-ph].arxiv.org/abs/2608.27045
  22. [22]S. B. Kaplan, “Acoustic matching of superconducting films to substrates,” Journal of Low Temperature Physics 37, 343–365 (1979).DOI
  23. [23]E. T. Swartz and R. O. Pohl, “Thermal boundary resistance,” Rev. Mod. Phys. 61, 605–668 (1989).
  24. [24]T. H. Swift, Characterisation of superconductors and quantum dots in 22nm-node CMOS devices, Ph.D. thesis, University College London (2025).
  25. [25]J. Bardeen, G. Rickayzen, and L. Tewordt, “Theory of the thermal conductivity of superconductors,” Phys. Rev. 113, 982–994 (1959).DOI
  26. [26]A. V. Feshchenko, O.-P. Saira, J. T. Peltonen, and J. P. Pekola, “Thermal conductance of Nb thin films at sub-kelvin temperatures,” Sci. Rep. 7, 41728 (2017).DOI
  27. [27]A. J. Kerman, E. A. Dauler, J. K. W. Yang, K. M. Rosfjord, V. Anant, K. K. Berggren, G. N. Gol’tsman, and B. M. Voronov, “Constraint-limited detection efficiency of superconducting nanowire single-photon detectors,” Applied Physics Letters 90, 101110 (2007).DOI
  28. [28]D. Y. Vodolazov, “Single-photon detection by a dirty current-carrying superconducting strip based on the kinetic-equation approach,” Phys. Rev. Appl. 7, 034014 (2017).
  29. [30]A. Kardakova, M. Finkel, D. Morozov, V. Kovalyuk, P. An, C. Dunscombe, M. Tarkhov, P. Mauskopf, T. M. Klapwijk, and G. Goltsman, “The electron-phonon relaxation time in thin superconducting titanium nitride films,” Applied Physics Letters 103, 252602 (2013).DOI

Paper details

vlnc:W_u4s69aw4
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents