IEEE TRANSACTIONS ON APPLIED SUPERCONDUCTIVITY, VOL. 15, NO. 3, SEPTEMBER 2005 3821 Thin-Film Persistent Current Switch Priti Balchandani, Rodney H. Torii, and Roger Shile Abstract—We have developed a fast, low power heat switch for switching a niobium thin film between the normal and superconducting state. The sputtered niobium film (400 nm thick, 100 m wide) has a critical current density of 5 1010 Am 2 . Switching is produced by joule heating a small section of the niobium film with a titanium thin-film resistor. With the heat switch in vacuum, the minimum heater power needed to switch to the normal state was 4.5 10 5 W. A simple three-dimensional thermal model shows that the minimum power is primarily determined by the thermal conductivity of the substrate. We have achieved response times less than 10 6 s. (a) Index Terms—Current injection, current switch, heat switch, thin film. I. INTRODUCTION A PERSISTENT current switch is used to switch a small portion of a superconducting circuit between the normal and superconducting state [see Fig. 1(a)]. One way to do this is by local heating (heat switch), driving the superconductor normal when the temperature exceeds the critical temperature. For a number of persistent current applications [1]–[3], a fast, low power heat switch is very useful. This thin-film heat switch was developed for the Satellite Test of the Equivalence Principle (STEP) experiment. STEP proposes to test the universality of free-fall by comparing the acceleration of test masses in orbit about the earth. The magnetic fields produced by persistent currents are used to constrain as well as detect test mass motion [1]. Thin-film heat switches are required to inject these persistent currents. We set design goals1 of a response time less than 10 s and a maximum power of 10 W. This performance is significantly better than previously reported by other workers [4], [5], whose switches required more than 10 W to operate, or had a response time longer than 10 s. The thin-film heat switch shown in Fig. 1(b) easily meets these requirements. A thin-film heat switch has two important advantages: 1) the heater can be put in very good thermal contact with the superconductor and 2) the heat switch can be made very small. As an added benefit, a thin-film heat switch can be easily manufactured and integrated with thin-film circuits. A thin-film resistor (heater) above the superconductor provides local heating. The heater is electrically Manuscript received September 9, 2004; revised February 14, 2005. This paper was recommended by Associate Editor M. Mueck. This work was supported by NASA under JPL Contract 959723. P. Balchandani is with the Lucas Center, Stanford University, Stanford, CA 94305-5488 USA (e-mail: pritib@stanford.edu). R. H. Torii is with STEP 250, Stanford University, Stanford, CA 94305-4085 (e-mail: torii@step.stanford.edu). R. Shile was with the Gravity Probe B Program, Stanford University, Stanford, CA 94305 USA. Digital Object Identifier 10.1109/TASC.2005.847491 1These requirements were set by the STEP experiment. (b) Fig. 1. (a) Schematic of a heat switch in a persistent current circuit. Local joule heating produced by R drives a portion H of the circuit normal\resistive allowing current I to be injected into inductive load L. A persistent current is trapped when H is allowed to return to the superconducting state. (b) 3-D solid model of the heat switch test circuit on a borosilicate substrate. isolated from the superconductor by a thin dielectric layer. Reducing the overall size of the heat switch decreases the thermal time constant of the heater and the superconductor. More importantly, for a given heater power, decreasing the heater area increases the heat flux into the substrate, enabling a faster response time. II. HEAT SWITCH FABRICATION We used sputtered niobium for the superconductor since we plan to use niobium for our persistent current circuits. In addition, we assume that the response time and power decrease as the niobium cross-section decreases in size. Thus, the test circuit was designed with a slightly larger cross-section than that required by our application. Heat switch test circuits 13 13 mm were fabricated on 76 mm diameter, 0.9 mm thick borosilicate wafers (21 test circuits per wafer).2 We chose a glass substrate for its low thermal conductivity. To begin the manufacturing process, 400 nm of niobium was dc magnetron sputtered onto the bare wafer. The niobium had of 8.9 K and a residual resistivity ratio ranging between a 2.5 and 5. Without breaking vacuum, 50 nm of gold was then sputtered onto the niobium. The gold layer makes possible a superconducting electrical connection to the niobium via a wire bond [6] and protects the niobium when etching the dielectric (the dielectric on the contact pads is etched to allow electrical contact to the niobium). Using a positive photoresist and stan2The wafers were purchased from Polishing Corporation of America, Santa Clara, CA. 1051-8223/$20.00 © 2005 IEEE 3822 IEEE TRANSACTIONS ON APPLIED SUPERCONDUCTIVITY, VOL. 15, NO. 3, SEPTEMBER 2005 dard lithographic techniques, niobium circuit elements 100 m wide and 2.1 mm long were patterned on the niobium-gold. The gold was wet etched using iodine and potassium iodide.3 The niobium was then patterned using a sulfur hexafluoride reactive ion etch. The gold is only needed on the contact pads since, as stated earlier, it is used to make electrical contact to the niobium. Thus, to prevent any thermal interference with the heater, the gold was removed from the niobium circuit. The room temperature resistance of the niobium circuit was typically 14.3 (resistivity of 2.7 10 m). A dielectric layer comprising 100 nm of silicon nitride was then deposited by plasma enhanced chemical vapor deposition (PECVD) from silane and ammonia at 280 C.4 This layer prevents the niobium from electrically shorting the heater. Thin-film heaters with a fast response time at low temperature have been in use for some time [7]. Based on this previous work, we first tried copper alloy heaters.5 A copper alloy film 100 nm thick was evaporated onto the silicon nitride. After a few thermal cycles, the copper alloy heaters failed (open circuit) apparently due to poor adhesion. The failure always occurred near a step, i.e., a transition from nitride\wafer to nitride\niobium\wafer. Rather than try to improve the film coverage over the step, we changed heater material. Titanium is an obvious choice since it has better adhesion properties and greater mechanical strength (it is also nonmagnetic which is important for our application). A titanium film 100 nm thick was evaporated onto the wafer. This thickness was a compromise between 1) a thicker film with good step coverage and a reduction in the relative contribution of the titanium oxide and 2) a thinner film with lower thermal conductance, lower heat capacity, and higher electrical resistance. A larger heater resistance makes the heater easier to operate because it requires smaller currents. Electrical contact to the titanium was made easier by depositing a gold layer 50 nm thick on the titanium before breaking vacuum. Using a positive photoresist and standard lithographic techniques, heater elements were patterned on the titanium-gold. We chose a very simple geometry for the heater, a single strip of titanium (Fig. 1). Heaters of varying widths (4 to 20 m) and lengths (113 to 170 m) were patterned on the same wafer. The titanium was wet etched using a solution (by volume) of one part hydrofluoric acid (48% assay), 20 parts nitric acid (71% assay), 54 parts glacial acetic acid, and 25 parts lactic acid (85% assay). This recipe wet etched the titanium with little undercut. The final step in the manufacture of the heater was to remove the gold from the active area of the heater to prevent shunting of the heater current. The resistivity of the tim at room temperature, reducing to tanium was 3.1 10 2.4 10 m at 4.2 K. The wafer was diced so that each of the 21 heat switch test circuits was on a separate borosilicate substrate (0.9 mm thick, 13 13 mm ). Prior to connecting electrical leads to the test circuit, the silicon nitride needed to be removed from the niobium contact pads. This was done by first protecting the niobium circuit element and heater with photoresist and then etching the sil3We used a gold etch from Transene Co., Danvers, MA. 4The PECVD process was done in a Pacific Western Systems Coyote Reactor, Mountain View, CA. 5We used manganin (84 Cu 12 Mn 4 Ni) and constantan (60 Cu 40 Ni). Fig. 2. Time response of the voltage across the niobium for a heat switch with a 4-m-wide titanium heater. The heat pulse was of amplitude 10 W and 1.6 10 s duration. The niobium bias current was 3 10 A. Thus, the effective niobium resistance is 0.13 in the normal state. 2 2 icon nitride in a reactive ion etch using a tetrafluoromethane and oxygen chemistry. Once the silicon nitride has been removed, superconducting wire bonds can be made to the niobium [6]. A number of different solders can be used for normal electrical connections.6 III. EXPERIMENTAL METHOD AND RESULTS We attached the borosilicate substrate (0.9 mm thick, 13 13 mm ) of a single test circuit to an oxygen-free high conductivity copper stage of a low-temperature probe. The probe was inserted into a dewar of liquid helium with the heat switch in vacuum. The copper stage, in good thermal contact with the liquid, cooled the heat switch to 4.2 K. A thermometer and heater on the copper stage were used to monitor and control temperature. A voltage pulse was applied to the heater using a 16 bit digital-to-analog (D/A) converter controlled by computer software. The resistance of the niobium was obtained by measuring the dc voltage across the niobium at a constant dc bias current (four wire measurement). The voltage was amplified by 10 before recording. In Fig. 2, the voltage across the niobium is shown for a bias current of A and an applied heat flux of 2.2 10 W m . 3 10 We could time measurements to 10 s by hardware triggering the analog-to-digital (A/D) converter with the rising edge of the clock used by the D/A converter. A data sample was by software triggering the D/A converter. initiated at The time resolution in Fig. 2 was limited to 20 s since the 6An indium alloy solder was used since a lead-tin alloy was more likely to cause the gold to detach from the niobium after a thermal cycle. The indium alloy (50 In 50 Sn) was purchased from the Indium Corporation of America, Utica, NY, and used with a rosin mildly activated flux. BALCHANDANI et al.: THIN-FILM PERSISTENT CURRENT SWITCH Fig. 3. Niobium resistance as a function of heater power for a heat switch with a 4-m-wide titanium heater. The resistance was calculated by dividing the voltage response by the bias current. Results are shown for several different bias currents. A/D converter was multiplexed to make two measurements: 1) the voltage across the niobium and 2) the heater voltage. For a single measurement, depending on the signal-to-noise ratio, we averaged ten to 100 samples. The data shown in Fig. 2 is the result of averaging ten samples. To compare the voltage response for different niobium bias currents, we defined the effective resistance as the voltage response divided by the applied bias current. The effective niobium resistance as a function of total power is plotted in Fig. 3 for a heat switch with a 4- m-wide titanium heater. The total power is the combined heat generated by the titanium heater and the self-heating of the niobium due to the bias current. The niobium self-heating was found by multiplying the bias current and the voltage response. For the largest bias current (10 A), the niobium self-heating contributes 10–20% of the heater power. As shown in Fig. 3, the resistance stays approximately the same for different bias currents at a particular power. Thus, the bias current has a small effect on the resistance of the niobium when self-heating is taken into account. This implies that the effective niobium resistance is more likely due to a local increase in temperature rather than a propagating normal state. Thus, thermal runaway is not a problem for bias currents up to 10 A and a total power of 2 10 W. The minimum power needed to switch to the normal state for a heat switch with a 4- m-wide titanium heater is 4.5 10 W. This is equal to a heat flux of 10 W-m . The highest rate at which we could acquire data by computer was 100 kHz. To study faster response times, we used a 250 MHz pulse generator in conjunction with a 100 MHz analog oscilloscope. We define the heating response time to be the amount of time the niobium voltage signal lags the heater signal, i.e., the time for the substrate to go from 4.2 K to the niobium critical temperature . This does not include the time (settling time) necessary to reach a steady-state equilibrium temper- 3823 Fig. 4. Heat switch response time as a function of heater power for a heat switch with a 4-m-wide titanium heater. The response time is shown for heating () and cooling ( ). The dotted line is the calculated response time as described in the text. . The heating settling time as measured by the niobium ature voltage signal is very difficult to model since we have no reliable way to relate the effective resistance to the temperature. to and The cooling response time can be measured from not to 4.2 K, since we lose the voltage signal at . The cooling response is not as useful as the heating response because of the . As shown in Fig. 4, the response time for uncertainty in heating (cooling) decreased (increased) with increasing heater power. IV. THERMAL MODEL We developed a thermal model based on the fact that the niobium becomes resistive when the temperature of the niobium rises above its critical temperature. The minimum power to switch to the normal state is that needed to raise the niobium film to its critical temperature. The temperature of the niobium near the heater is always higher than at the surface of the substrate. Thus, the temperature at the surface of the substrate determines when the niobium switches to the normal state. The critical switching power will be given by the minimum power needed to cause the surface temperature of the borosilicate substrate to rise above the critical temperature of niobium. We make three additional assumptions to simplify the calculation. The first is that the substrate is a semi-infinite slab. This is valid as long as the total heat applied to the substrate is small (so that the temperature rise of the substrate is negligible). Given that the time to reach steady-state equilibrium is relatively short, this requirement can be easily achieved by experiment (since the total heat applied is proportional to the pulse width). The second assumption is that the thermal conductivity and the heat capacity of the borosilicate substrate is temperature independent over the range of interest. The thermal conductivity k of borosilicate glass is 0.10 and 0.13 W m K at 4 and 10 K, respectively [8]. We use W m 3824 IEEE TRANSACTIONS ON APPLIED SUPERCONDUCTIVITY, VOL. 15, NO. 3, SEPTEMBER 2005 Fig. 5. Calculated temperature contour T (x; y ) near the surface of the borosilicate substrate for a heat switch with a 4-m-wide 113-m-long titanium heater. The maximum temperature is 9.5 K. Each contour line corresponds to a change in temperature of 0.1 K. K in all our calculations. This assumption does not hold so well for the heat capacity, which varies as the cube of the temperature. The large uncertainty introduced by the heat capacity applies only to the time-dependent solution and does not affect the steady-state solution. The third assumption is that the heat conducted parallel to the surface of the borosilicate substrate by the heater, silicon nitride, or niobium is negligible. This approximation is valid for the heater and silicon nitride considering the thickness and thermal conductivity of these materials. We have estimated that as much as 10 W may be conducted by the niobium circuit.7 The three-dimensional (3-D) thermal boundary value problem is to find the temperature distribution near the surface of a semi-infinite slab for which the surface boundary condition is defined by a heat flux (derivative of the temperature). If the plane defines the surface of the semi-infinite slab, , the then for a heater of size boundary condition at is otherwise (1) is where is the thermal conductivity of the slab, is a uniform heat flux. is equal to the temperature, and heater power divided by the area of the heater. The initial condition for the slab is that at , the slab is at a uniform temperature . The heat equation is [9] (2) where is density and is specific heat at constant pressure. For constant , the above equation becomes 7The thermal conductivity of the niobium thin film was estimated using the measured resistivity and the Wiedemann–Franz relation. This was a worst case estimate to check the maximum effect of the niobium. (3) BALCHANDANI et al.: THIN-FILM PERSISTENT CURRENT SWITCH 3825 where is the thermal diffusivity. The partial derivative in time can be converted to a total derivative in time by a Fourier transform. The resulting ordinary differential equation can be solved for the above boundary conditions by separation of variables. The temperature is8 (4) For steady-state conditions, i.e., , this reduces to (5) According to (5), for a fixed heat switch geometry, the power depends linearly on , i.e., an increase in requires an equivalent increase in . This means that for a silicon substrate, with to W m K between 4 and 10 K [10], the heat switch would operate at a power which is approximately 1000 to 6000 times larger than borosilicate glass. It has been estimated [4] that for a sputtered 400-nm-thick niobium thin film on silicon, the heat switch requires a power of 0.3 W. This being linearly dependent on . Thus qualitatively supports borosilicate substrate, with its low value of , is an appropriate choice for our low power heat switch. For a heat switch with a m and heater 4 m wide and 113 m long, we set m. The result for is drawn as a contour plot in Fig. 5. The highest contour (white) corresponds to a temperature of 9.5 K. The temperature changes by 0.1 K from line to line. This plot shows that the heater produces a very local rise in temperature. This means that for a good heat switch design, the heater should extend somewhat beyond the width of the superconductor. for tiIn Fig. 6, we plot the critical temperature profile tanium heaters of three different sizes. The critical temperature profile was calculated by setting the heat flux for each heater size equal to the measured critical power for that heater. The following heaters were used: 4 m wide, 113 m long; 10 m wide, 120 m long; and 10 m wide, 170 m long. The measured critical power for each heater geometry was: 10 W; 10 W; and 8.2 10 W. This corresponds to a heat flux of 10 5 10 , and 4.8 10 W m , is along the 100 m width of the niobium respectively. circuit, parallel to the length of the heater. As described earlier, we made test circuits with different heaters on the same wafer. Thus, the thermal conductivity and the critical properties of the 8The same solution was found by the Green’s function method. The following Green’s function G was derived: G= 1 (t 0 t )) (4 e 2 e + e : Fig. 6. Calculated temperature profile T (x) at critical power near the surface of the borosilicate substrate. The temperature is along the 100 m width of the niobium circuit. Results are shown for the following three titanium heaters: ( ) 4 m wide, 113 m long; ( ) 10 m wide, 120 m long; and ( ) 10 m wide, 170 m long. } niobium circuits are probably very similar for each heater test circuit. Therefore, the temperature profiles at the critical power (i.e., near the critical temperature) for the test circuits should be of similar magnitude. It can be seen that the absolute magnitude is close to the critical temperature expected for nioof bium even though the heat conducted by the niobium has been ignored. The effect of the niobium is to decrease and broaden . We estimate a decrease in of 20–30%. The time-dependent solution for temperature is (6) where Erf is the error function and is the thermal diffusivity. As stated earlier, the strong temperature dependence of the heat capacity, leading to a temperature-dependent , has been ignored. The specific heat of borosilicate glass is 0.17 kg at 4 and 10 K [11]. We set and 4.20 J K m s in all our calculations. The time-dependent temperature is shown in Fig. 7 for the following three titanium heaters: 4 m wide, 113 m long; 10 m wide, 120 m 10 m wide, 170 m long. The temperature is callong; and culated near the origin and at critical power. As shown in Fig. 7, the temperature rises faster as the heater becomes smaller. We defined the heating response time as the time for the substrate to go from 4.2 K to the niobium critical temperature . To see how response time varies with heater power, we have calculated the temperature rise for the 4- m-wide heater at the following heater powers: 1) 5.5 10 W, dotted line; 2) 6.1 10 W, dashed line; and 3) 8.8 10 W, dot-dash line. As shown in 3826 IEEE TRANSACTIONS ON APPLIED SUPERCONDUCTIVITY, VOL. 15, NO. 3, SEPTEMBER 2005 ondary is connected to the input coil of a dc SQUID. This will allow us to study the effect of the thin-film heat switch on the flux trapped in . ACKNOWLEDGMENT The authors would like to thank D. Gill and T. Carver for depositing the metal films. REFERENCES Fig. 7. Calculated temperature as a function of time for the following three titanium heaters: () 4 m wide, 113 m long; ( ) 10 m wide, 120 m long; and ( ) 10 m wide, 170 m long. The temperature is calculated near the origin and at critical power. The temperature is also calculated for the W, dotted 4-m-wide heater at the following heater powers: 1) 5.5 10 line; 2) 6.1 10 W, dashed line; and 3) 8.8 10 W, dot-dash line. } 2 2 2 Fig. 7, for a given critical temperature , an increase in shifts the temperature upwards, producing a shorter response time. The response time at minimum or critical power tends to infinity and is poorly defined. This rapid increase in the response time is a useful indicator when measuring the critical power. The calculated response time, as shown in Fig. 4 as a dotted line, is in qualitative agreement with experiment [ heating response time]. The cooling response time was not calculated since the initial temperature is poorly defined (by theory and experiment). The increase in cooling response time with increasing power is due to the increase in the initial temperature. V. CONCLUSION We have shown that a thin-film heat switch can easily achieve a level of performance that would be very challenging for a wire heat switch [5]. The heat switch using the smallest heater (4 m wide, 113 m long) performed the best, with a response time of 6 10 s at a power of 5 10 W. The response time can be reduced to 10 s with a 50% increase in heater power. Devices with similar response times but higher input power have been fabricated [12]. In general, it is always of some benefit to decrease the power consumption of a device while making it faster. For a device that operates in a low-temperature environment, the benefit is obvious since the total allowed dissipated power can be small [1]. The second practical consequence has to do with time. A shorter response time means that the switch can be activated more often in the allotted time. This means that a greater number of adjustments to a persistent current can be made, and there may be additional benefits to the specific application by having a short switching time. We plan to add our thin-film heat switch to circuits similar to the circuit shown in Fig. 1(a). In one circuit, is the primary of a thin-film flux transformer. The sec- [1] P. W. Worden, Jr. and C. W. F. Everitt, Experimental Gravitation, B. Bertotti, Ed. New York: Academic, 1974. [2] J. T. Anderson, B. Cabrera, M. A. Taber, S. B. Felch, and J. Tate, “Lowfrequency noise reduction in SQUID measurements using a laser-driven superconducting switch. Part A: Direct input circuit switching,” Rev. Sci. Instrum., vol. 60, p. 202, 1989. [3] H. J. Paik, “Superconducting tunable diaphragm transducer for sensitive acceleration measurements,” J. Appl. Phys., vol. 47, p. 1168, 1976. [4] T. R. Stevenson, “First coincidence experiments between cryogenic resonant-mass gravitational wave detectors. II Development of a thin-film superconducting transducer for a gravitational wave antenna,” Ph.D. dissertation, Stanford Univ., Stanford, CA. [5] M. Urata, T. Yazawa, H. Maeda, T. Tomisaki, S. Kabashima, K. Sasaki, and T. Kumano, “Stabilization of Cu–Ni based persistent current switch,” IEEE Trans. Appl. Supercond., vol. 3, no. 1, pp. 586–589, Mar. 1993. [6] T. Ogashiwa, A. Inoue, and T. Masumoto, “Superconductive wire bonding with high reliability at 4.2 K,” Jpn. J. Appl. Phys., vol. 28, p. 2479, 1989. [7] R. J. von Gutfeld and A. Nethercot Jr., “Heat pulses in quartz and sapphire at low temperatures,” Phys. Rev. Lett., vol. 12, p. 641, 1966. [8] R. C. Zeller and R. O. Pohl, “Thermal conductivity and specific heat of noncrystalline solids,” Phys. Rev. B, vol. 4, p. 2029, 1971. [9] H. S. Carslaw, Conduction of Heat in Solids. Oxford, U.K.: Clarendon, 1986. [10] G. K. White and S. B. Woods, “Thermal conductivity of germanium and silicon at low temperatures,” Phys. Rev., vol. 103, p. 569, 1956. [11] G. K. White, Experimental Techniques in Low-Temperature Physics. Oxford, U.K.: Clarendon, 1987. [12] A. C. Leuthold, R. T. Wakai, G. K. G. Hohemwarter, and J. E. Nordman, “Characterization of a simple thin-film superconducting switch,” IEEE Trans. Appl. Supercond., vol. 4, no. 3, pp. 181–183, Sep. 1994. Priti Balchandani was born in Bhopal, India, in September, 1979. She received the bachelor of applied science degree in computer engineering from the University of Waterloo, Waterloo, ON, Canada, and the master’s of science degree in electrical engineering from Stanford University, Stanford, CA, where she is currently pursuing the Ph.D. degree in electrical engineering. Her employment experience includes internships with Cisco Systems, San Jose, CA; Fujitsu Network Communications, Richardson, TX; Canadian Space Agency, Ottawa, ON; and Nortel Networks, Brampton, ON. She is currently with the Gravity Probe-B and STEP research groups at Stanford University, Palo Alto, CA. Ms. Balchandani received the National Science and Engineering Research Council Award to conduct research with the Canadian Space Agency on spacequalification of semiconductor devices. Rodney H. Torii, photograph and biography not available at the time of publication. Roger Shile, photograph and biography not available at the time of publication.